Table of Contents 1.0 Introduction 2.0 University of Michigan Progress Report 3.0 University of California at Davis Progress Report 4.0 University of Wisconsin Progress Report 5.0 ERIM Progress Report 6.0 George Washington University Progress Report

1.0 Introduction The Eos Simultaneity Study has been funded since June 1987 under grant NAGW-1101 from the University Applications Program at NASA Headquarters. Over the three years of this project, institutional participation in the project has included the University of Michigan, the Jet Propulsion Laboratory, George Washington University, the University of Nebraska, University of California at Davis, the University of Wisconsin, ERIM, NASA/AMES Research Center, Duke University and Aster Consulting; and subcontracts have generally been administered through the University of Michigan (with the exceptions of JPL and Duke). This executive summary is drawn from Technical Progress Reports provided by each institution in December 1989 which document accomplishments and plans for the remainder of the third year of funding and will also include results of a meeting of the investigator team held at the University of Michigan on February 14, 1990. The study has yielded a number of technical results related both to the fundamental objectives of this program and to remote sensing science in general. For example. it has been found that while the signal from Eos SAR can be expected to be diurnally variant in response to changes in leaf-angle distribution and changes in canopy and substrate moisture condition, this variation is expected to be relatively continuous in the absence of certain impulses of a meteorologic (i.e., rain) or anthropogenic (i.e., harvest or tillage) nature. Consequently, no case can be made for simultaneous observation of vegetated terrain by Eos SAR and HIRIS. The requirements for specific relative offsets in the timing of SAR and HIRIS are driven by the extent to which the data from the two instrument packages can be analyzed synergistically to provide estimates of specific biophysical attributes. This issue cannot be fully addressed at this time since robust inversion approaches for either SAR or HIRIS have not been developed and tested to date. However, great progress has been made in development and validation of direct models to predict observed radar

backscatter and optical reflectance on the basis of biophysical attributes. Much of the work in this study to date has focused upon these models. Finally, the experimental efforts of this study have significantly enhanced our capability to conduct field studies through improvement of procedures and techniques. Advances include new techniques for in situ observation of microwave dielectric properties and direct field comparisons of various methods for estimation of leaf area and the angular distributions of canopy elements. This study has provided partial support to 16 faculty and primary researchers, 16 graduate students, 16 undergraduate students, and 9 members of administrative and technical support staffs. These efforts have resulted in 18 technical presentations at workshops and symposia and 10 technical publications. The following sections present the technical progress reports for each of the participating institutions.

TECHNICAL PROGRESS REPORT on Eos Synergism Study University of Michigan M. Craig Dobson January 1990

1.0 Objectives The general objective of this study is to examine the relative timing requirements of the Eos SAR and HIRIS for observation of terrain and particularly vegetated terrain. To a large extent, this issue is determined by the time stability of the expected signal or, in other words, how quickly do backscatter (for SAR) and reflectance (for HIRIS) decorrelate in the absence of some impulse such as rain or harvest. If both signals (HIRIS and SAR) are expected to be time constant, then relative timing becomes a non-issue. On the other hand, if one or more of the signals is time variant and significant with respect to expected calibration, then what are the optimal observation times for retrieval of specific biophysical quantities and can SAR and HIRIS data be used synergistically in such retrievals. Hence, a specific objective of this study is to define the temporal variance of o~ and reflectance for natural vegetationi canopies over the periods of minutes, hours and days. A second objective is to develop retrieval algorithms for SAR and HIRIS (individually and combined in a synergistic fashion). These algorithms are to be developed via first validating direct models of reflectance and backscatter, then conducting sensitivity studies of the validated models to specific biophysical variates, and finally inverting these models, as appropriate, for use in retrieval algorithms to be tested against data. The specific objectives of the University of Michigan as part of the overall study have been to: (1) coordinate the activities of the various members of the investigator team, (2) generate experimental data sets to show the time variation of reflectance and backscatter from vegetation over the appropriate time scales using ground-based and airborne sensors, and (3) validate radar scattering models in conjunction with these data sets for use in sensitivity studies and development of retrieval algorithms applicable to Eos SAR and HIRIS.

2.0 Major Accomplishments 2.1 Specific Accomplishments (1) Under a precursor to the existent program, a truck-based experiment was conducted in summer 1986 to examine the diurnal variation in oo and reflectance from a phototropic canopy (soybeans). (2) An experiment was planned and successfully executed in summer 1987 at a Walnut orchard at the Kearney Agricultural Research Center using boom-mounted instrumentation to investigate the diurnal variation of o0 and reflectance due to moisture related properties of the canopy. (3) A radar scattering model, MIMICS, has been modified for the walnut orchard data and validated for that canopy at L-band. (4) Initial sensitivity studies have been performed at L-band for various biophysical attributes of the walnut orchard. (5) Preparations were made for an airborne experiment at the Duke University Research Forest in summer 1989. The field experiment was conducted (SAR only) and the field data has been partly processed and edited. 2.2 Detail Statement The precursor experiment in 1986 involved participants from the University of Michigan, JPL (Paris, Cimino, Cook and Smith), NASA/Ames (Vanderbilt), and George Washington University (Lang). This experiment was the first attempt to obtain intensive simultaneous field observations with both radar scatterometers and visible/infrared radiometers, and established protocols for measurement strategy. Important issues defined by this experiment were (1) the need for

adequate time calibration of the sensors to account for thermal and solar conditions, (2) appropriate sampling strategies to account for both the temporal and spatial variation in canopy attributes as regards both the sensing systems and also the "ground truth" data, and (3) procedures for data acquisition and subsequent processing. Due to the problems encountered during the experiment (both procedural and related to intermittent cloud cover), the data and results were not suitable for publication. However, it was determined that changes in leaf-angle orientation are (1) very difficult to characterize with statistical satisfaction and (2) significant determinants of radar backscatter at higher frequencies (C- and Xbands). These results of this experiment contributed to a significantly improved experiment plan for the Kearney experiment in 1987. An integrated field experiment was planned and conducted in summer 1987 at the walnut orchard and included numerous individuals from the University of Michigan, JPL, UC/Davis, NASA/Ames, George Washington University, and the University of Nebraska. The University of Michigan was primarily responsible for (1) overall experiment design, (2) acquisition of radar scatterometer data, (3) acquisition of moisture related ancillary data (i.e., soil and canopy gravimetric moisture, leaf water potential, and soil and canopy dielectric properties), and (4) editing and processing of these data. The experiment was quite successful and produced a very comprehensive data set summarized in Cimino, et al. (1988). This data was also presented at various workshops and symposia (Weber and Ustin (1988), Dobson, et al. (1988), Cimino, et al. (1988), Dobson (1988), and Ulaby, et al. (1988)). Using the field data obtained at the walnut orchard (i.e., moisture related quantities obtained by the University of Michigan and canopy architectural quantities obtained by UC/Davis) the MIMICS model was modified and used to

successfully predict the observed variation of'p (0) at L-band and also the diurnal variation of o4 over a three-day period. The variations in the dielectric properties of the woody components of the trees and the soil substrate are found to account for the observed variation in oo which ranged from 1 to 3 dB. These results have been presented at symposia McDonald, et al. (1989) and submitted for publication McDonald, et al. (1989) and Way, et al. (1990). The validated MIMICS model has also been tested for other forest stands such as the frozen and thawed winter-time conditions for white spruce, black spruce, balsam poplar and alder. In general, excellent agreement is obtained with both the canopy propagation loss and backscatter derived from SAR observations at L-, C-, and X-bands. These results have been presented at symposia Dobson, et al. (1989) and Dobson, et al. (1990) and have also been submitted for publication Dobson, et al. (1990). Importantly, it was found that MIMICS-I (which assumes a continuous and closed canopy) cannot account properly for the observed oo from sparse black spruce stands. As a consequence, MIMICS-II (for discontinuous tree canopies) has been developed and will be presented at IGARSS'90 (McDonald and Ulaby, 1990). The highly encouraging MIMICS model validation results have given credence to preliminary model sensitivity studies of L-band backscatter response to variations in the dielectric constants of trunks, branches and the soil substrate. These results have been presented at IGARSS (McDonald, et al. 1988 and McDonald, et al. 89) and submitted for publication (McDonald, et al. 1990). In an effort to extend the boom-mounted experimentation conducted at Kearney to a larger and more "natural" ecosystem, an airborne experiment was planned for loblolly pine stands at the Duke University Research Forest in summer

1989. This experiment was to involve truck-mounted scatterometry, ariborne SAR, AVIRIS and ASAS. Extensive preparatory work for the scatterometer was aborted by problems with the airborne SAR schedules and also strong theoretical indications that low altitude scatterometry (i.e., truck, platform, and helicopter mounted systems) were not suited to observing the net expected o' (for a SAR). Although SAR observations in July 1989 were aborted, an extensive set of dielectric observations were made for loblolly pines and included sampling to identify (1) depth profile of trunk dielectric, (2) within-stand variance in e*, (3) between-stand variance in E*, (4) variation in ~* of trunks as a function of vertical location, (5) diurnal variation of E*, and (6) seasonal variation of ~*. In addition, detailed observations of moisture related quantities were made in association with airborne SAR data sets obtained at Duke at the end of August and beginning of September 1989. These.data are presently being edited for use by MIMICS in prediction of the SAR response to selected test stands of variable age and density. Of note is that the diurnal variations of ~* in the trunks of loblolly pines is not as strong as that found for the walnuts; and also that there is a variation in E* with vertical height up the trunk and this variation is not monotoric which indicates that two or more competing forces are controlling water distribution in the xylem. 3.0 Planned Activities During the 1989-1990 time period the University of Michigan plans to conduct the following analyses: (1) complete editing and processing of the dielectric data obtained at Duke in 1989, (2) use the dielectric data in conjunction with stand architecture data obtained by Duke to initialize MIMICS for selected continuous stands of loblolly pine,

(3) test MIMICS-II for discontinuous stands of black spruce (from Alaska) and/or for stands of young loblolly pines (Duke) and compare to SAR and heloscat observations of o~, and (4) conduct more thorough sensitivity analyses of 00, as predicted by MIMICS to (a) leaf angle orientation, (b) leaf moisture, (c) trunk and branch dielectrics, and (d) substrate conditions including moisture, roughness and the presence of intervening layers of snow and organic litter.

4.0 Personnel Supported The following personnel have been supported by this project: Name Status Responsibility Fawwaz Ulaby PI overall direction of program Craig Dobson Co-I direction of field measurements Tom Senior Faculty advisor on certain aspects of model development Jim Weber Res. Scientist canopy water relations meas. Tom Haddock Res. Scientist scatterometter development & operation Val Liepa Faculty scatterometer calibration Kamal Sarabandi Res. Scientist Scatterometers and modeling Kyle McDonald GSRA MIMICS and field experiments Michael Whitt GSRA MIMICS and field experiments M. Ali Tassoudji GSRA scatterometer support Walid Ali Ahmad GSRA Radar calibration Roger De Roo GSRA dielectric probe calibration Ron Hatikka Sr. Technician scatterometer support Parag Mody Technician scatterometer support Jeanette Vechio Secretary secretarial support Beth Olson Secretary secretarial support Bonnie Kidd Secretary secretarial support Timarie Wilkins hourly secretarial support Martin Kuttner hourly technical lab support Mike De Liso hourly technical lab support Sebastian Lauer hourly lab and field support Ron Oliver hourly lab support Jorge Hernandez hourly lab support Alan Klingelhafer hourly lab support Roberto Mitrevski hourly lab support Andrew Isztwan hourly lab support Darelyn Crochran hourly lab support G. Eleftheriades GSRA lab support variously hourly help at Kearney Agricultural Res. Center The following students have received degrees or are expected to do so over the next year with significant support from this program. M. Ali Tassoudji - Masters June 1989 Kamal Sarabandi - PhD Sept. 1989 Kyle McDonald - PhD 1991 expected Michael Whitt - PhD 1991 expected

5.0 Publications List 1. Cimino, JoBea, et al. "Eos Synergism Study: 1987 Field Experiment Data Report", JPL Tech. Rep. March 1988. 2. Cimino, J.B., C. Bruegge, D. Diner, J. Paris, C. Dobson, D. Gates, F. Ulaby, N. Goel, E. Kasischke, D. Kimes, r. Lang, J. NOrman, and V. Vanderbilt, "Synergism Requirements and Concepts for SAR and HIRIS on Eos," International Geoscience and Remote Sensing Symposium (IGARSS'87) Digest, Ann Arbor, Michigan, Vol. 2, pp. 955-966, May 18-21, 1987. 3. Cimino, J.B., J. Paris, M.C. Dobson, D. Gates, J.A. Weber, F.T. Ulaby, S.L. Ustin, V. Vanderbilt, J. Norman, R. Lang, and E. Kasischke, "Eos Synergism Study: Overview and Objectives," International Geoscience and Remote Sensing Symposium (IGARSS'88) Edinburgh, Scotland, September 1988. 4. Cimino, J.B., J. Paris, C. Dobson, F. Ulaby, J. Waber, V. Vanderbilt, S. Ustin, J. Norman, R. Lang, E. Kasischke, and Ray Hunt, "Synergism Study: Implications of Diurnal Change in Vegetation Canopies on Eos Orbit Selection," 1989 International Geoscience and Remote Sensing Symposium (IGARSS'89), July 1014, 1989, Vancouver, B.C., Canada. 5. Cimino, J.B., J. Paris, D. Casey, F. Ahern, N. Christensen, M.C. Dobson, F.T. Ulaby, J. Weber, R. Hoffer, M. Imhoff, E. Kasischke, A. Milne, J. Richards, A. Sieber, P. Churchill, D. Simonett, C. Slaughter, L. Viereck, E. Mougin, and T. LeToan, "The Effect of Changing Environmental Conditions on Microwave Signatures of Forest Ecosystems," submitted to International Journal of Remote Sensing, special issue on Microwave Signatures of Forests Workshops. 6. Dobson, M.C., F.T. Ulaby, and J. Paris, "Radar Backscatter from Tree Canopies", Forest Signatures Workshop, JPC, Ispra, Italy, September 1988.

7. Dobson, M.C., K. McDonald, F.T. Ulaby, and J.F. Paris, "Diurnal Patterns in Multifrequency, Multipolarization Backscattering by a Walnut Orchard," International Geoscience and Remote Sensing Symposium (IGARSS'88), Edinburgh, Scotland, September 1988. 8. Dobson, M.C., "Diurnal and Seasonal Variations in the Microwave Dielectric Constant of Selected Trees," International Geoscience and Remote Sensing Symposium (IGARSS'88). Edinburgh, Scotland, September, 1988 and in preparation for submission to International Journal of Remote Sensing. 9. Dobson, M.C. and E.S. Kasischke, "Microwave Attenuation by Boreal Forest Canopies in Winter," submitted to 1989 International Geoscience and Remote Sensing Symposium (IGARSS'89), July 10-14, 1989, Vancouver, Canada. 10. Dobson, M., K. McDonald, F.T. Ulaby and J. Cimino, "Effects of Temperature on Radar Backscatter from -Boreal Forests," IGARSS'90. 11. Dobson, M.C., K. McDonald, E. Kasischke, J. Way, and F.T. Ulaby, "Effects of Temperature on Microwave Attenuation and Backscatter from Boreal Forest Stands, to be submitted to IEEE Trans. Geoscience and Remote Sensing 1990. 12. McDonald, K., M.C. Dobson, and F.T. Ulaby, "Determination of Soil Moisture Beneath a Stalk or Trunk Dominated Canopy by Radar," International Geoscience and Remote Sensing Symposium (IGARSS'88), Edinburgh, Scotland, September 1988. 13. McDonald, K.C., M.C. Dobson, and F.T. Ulaby, "Using MIMICS to Model Microwave Backscatter from Tree Canopies," 1989 International Geoscience and Remote Sensing Symposium (IGARSS'89), July 10-14, 1989, Vancouver, B.C., Canada. 14. McDonald, K. and F.T. Ulaby, "MIMICS II: Radiative Transfer Modeling of Discontinuous Tree Canopies at Microwave Frequencies," IGARSS'9().

15. McDonald, K., M.C. Dobson, and F.T. Ulaby, "Using MIMICS to Model L-Band Multi-angle and Multi-temporal Backscatter from a Walnut Orchard," submitted to IEEE Transactions on Geoscience and Remote Sensing, November 1990. 16. Sieber, A.J., J.B. Cimino, R. Brown, J. Cihlar, C. Dobson, J. Ford, D. Gates, P. Hartl, G. Hildebrandt, R. Hoffer, M. Imhoff, E. Kasischke, J.P. Malingreau, J. Megier, T. Milne, J. Paris, J. Richards, F. Rocca, B. Rock, M. Sami, D. Simonett, T. LeToan, and F.T. Ulaby, "The Global Forest Ecosystem as Viewed by ERS- 1, SIR-C and Eos," International Geoscience and Remote Sensing Symposium (IGARSS'87) Digest, Ann Arbor, Michigan, Vol. 2, pp. 967-974, May 18-21, 1987. 17. Tassoudji, A., K. Sarabandi, and F.T. Ulaby," Design Consideration and Implementation of the LCX Polarimetric Scatterometer (POLARSCAT), Radiation Laboratory Tech. Rep. 022486-T-A, EECS Dept., Univ. of Michigan, Ann Arbor, Michigan, June 1989. 18. Ustin, S.L., V.C. Vanderbilt, J. Way, F. Ulaby, C. Dobson, and D.M. Gates, "Opportunities for Eos Instrument Synergism in Monitoring of Forest Ecosystems," IGARSS'90 Geoscience and Remote Sensing Symposium, University of Maryland, College Park, Maryland, May 20-24, 1990. 19. Way, J., M. Schier, N. Christensen, C. Dobson, D. Gates, F. Ulaby, E. Kasischke, R. Lang, J. Norman, J. Paris, S.L. Ustin, V.C. Vanderbilt, and J. Weber, "Eos Synergism Study: Diurnal Change in Trees as Observed by Optical and Microwave Sensors," to be submitted to Transactions Geoscience and Remote Sensing, 1990. 20. Weber, J.A. and S.L. Ustin, "Water Relations of a Walnut Orchard: Simultaneous Measurements with Remote Sensing," IGARSS'88, Edinburgh. Scotland, September 13-16, 1988, 1988, ESA-SP284, Vol. III: 1749-1752, 1988.

EFFECTS OF TEMPERATURE ON RADAR BACKSCATFTER FROM BOREAL FORESTS M.C. Dobson, K. McDonald, F.T. Ulaby and J. Cimino Abstract: An airborne SAR campaign during March 1988 acquired P-, L-, C- and X-band data for repeated overflights of the Bonanza Creek Experimental Forest located along the Janana River near Fairbanks, Alaska. The experimental forest included many stands characteristic of the boreal forest of interior Alaska: white spruce (Picea glauca), black spruce (Pice mariana), and balsam poplar (Populus balsamifera). The airborne SAR data was complemented by extensive characterizations of forest stand parameters and microwave dielectric conditions. These scene properties are used as inputs to the Michigan Microwave Canopy Scattering (MIMICS) Model, which uses a first-order radiative transfer approach to predict radar backscatter as functions of the radar wave parameters of frequency, polarization and angle of incidence. During the airborne campaign, an early spring thaw with air temperatures of 90C was followed by more typic winter conditions with air temperatures of - 150C. SAR data were obtained for both the frozen and thawed conditions by the P-, L- and C-band polarimetric SAR operated by the Jet Propulsion Laboratory. These data sets are calibrated with respect to arrays of external calibration point targets. The image data is segmented by forest stand and the resultant average values of the radar backscattering coefficient are shown to compare favorably with the values predicted by MIMICS.

USING MIMICS TO MODEL MICROWAVE BACKSCATTER FROM TREE CANOPIES K. C. McDonald, M. C. Dobson and F. T. Ulaby The Radiation Laboratory Department of Electrical Engineering and Computer Science The University of Michigan Ann Arbor, MI 48109-2122, U.S.A. Phone: (313) 764-0500 FAX: 313/936-3492 Telex: 432-0815 UOFM-UI ABSTRACT The Michigan Microwave Canopy Scattering Model (MIMICS) is used to model microwave scatterometer data that were obtained during the August 1987 EOS Simultaneity Experiment. During this experiment, L-, C- and X-band truck-based scatterometers were used to measure radar backscatter from a walnut orchard in Fresno County, California. Multi-polarization data were recorded for orchard plots of varying irrigation levels. MIMICS, a scattering model based on radiative transfer theory, is applied to model various data sets recorded during the experiment. Groundtruth data are used as inputs to MIMICS and the resulting modeled data are compared to the measured backscatter. Data examined in this study include a series of diurnal measurments in which a single orchard plot was observed continuously over several 24 hour periods, a multi-angle data set for which this same plot was observed at varying incidence angles, and a series of data recorded on two plots of different irrigation levels. In modeling the canopy backscatter, MIIMIICS accounts for scattering contributions directly from the trees themselves, direct backscatter contributions from the underlying ground surface, and contributions resulting from interactions between the trees and the ground. The model is shown to account for variations in canopy backscatter that are driven by diurnal processes as well as by the differing irrigation levels. The distribution of branch sizes and orientations within the tree crowns is shown to be an important parameter for modeling canopy backscatter.

MICROWAVE ATTENUATION BY BOREAL FOREST CANOPIES IN WINTER M. C. Dobson The Radiation Laboratory Department of Electrical Engineering and Computer Science The University of Michigan Ann Arbor, Ml 48109-2122 Telephone: (313) 764-0500 FAX: 313/936-3492 TELEX: 432-0815 UOFM-UI E. S. Kasischke Environmental Research Institute of Michigan P.O. Box 8618 Ann Arbor, MI 48107 Telephone: (313) 994-1200 FAX: 313/994-0944 TELEX: 494-0991 ERIMARB ABSTRACT In March of 1988, a series of overflights were conducted by two airborne SARs over the Bonanza Creek Experimental Forest near Fairbanks, Alaska. The SARs included the Jet Propulsion Laboratory Airborne Imaging Polarimeter operating at P-, L-, and C-bands aboard a NASA/Ames DC-8 and the ERIM/Naval Air Development Center SAR operating at C- and X-bands aboard a NADC P-3. Although an early thaw had melted much of the water in the vegetation and produced wet surface conditions in the snow pack, the majority of the overflights occurred after an ensuing freeze. The test site consisted of numerous stands of white spruce, black spruce, balsam poplar, and alder on islands along the Tanana River. A number of these stands were instrumented with arrays of corner reflectors (both tnhedral and dihedral) and L-band active radar calibrators. The tnhedrals ranged in size from 30 cm to 120 cm and were boresighted in the expected direction of the SARs. Each stand was carefully mapped with respect to tree type, location, diameter and height. Statistics of the canopy derived from these measurements and allimetric relationships were input into the Michigan Microwave Canopy Scattering (MIMICS) Model along with pertinent dielectric information and used to estimate the two-way attenuation by the canopies. These estimates are found to compare favorably to remotely sensed estimates of the attenuation as determined by comparing the backscatter from the arrays of point targets to the backscatter from adjacent portions of each stand.

Technical Progress Report Optical Model Application in Vegetation Remote Sensing John M. Norman Department of Soil Science 1525 Observatory Drive University of Wisconsin Madison, WI 53705 December 10, 1989

1) Objectives This research activity has two objectives: 1) Comparison of predictions from the Cupid model with remotely-sensed observations from several vegetation canopies approximating full cover, and 2) inverting the Cupid model to estimate several canopy biophysical characteristics from the remote observations. A major goal of the main proposal, which is entitled "Extended Ecosystem Signatures With Application to EOS Synergism Requirements", is to conduct field experiments to measure optical, radar and thermal signatures of forest, prairie and agricultural scenes. The work being done under this subcontract uses results from these field studies to test the model, Cupid. Inputs for Cupid would be provided by those directly involved in the field experiments. Predictions of the canopy bidirectional reflectance distribution functions (BRDF) from Cupid will be compared with BRDF measurements from the field experiments. If the model performs adequately, or can be made to perform adequately in the forward direction, then we intended to invert the model to produce estimates of canopy biophysical characteristics from the measured BRDF's or directional radiances. Although the major emphasis of this research is in the optical wavelength band, we had planned to coordinate our work with Fawwaz Ulaby and Roger Lang who are developing Radar models. 2) Major Accomplishments 2.1 Specific Accomplishments: Field measurements of Walnut leaf reflectance and transmittance with the leaves'attached to the trees and comparison with laboratory measurements on excised leaves. Development of a stem water flow model for comparison with the stem dielectric constant measurements. Inversion of fish-eye photographs taken by Susan Ustin to obtain leaf area index of the Walnut stand. Modification of the Cupid model to include leaf angle distributions that are asymmetric about the azimuth. Comparison of one-dimensional Cupid model predictions of nadir reflectance with nadir measurements made by'emrn Vanderbilt. -1

2.2 Detailed Statement: Leaf reflection and transmission measurements (400nm to 1000nm wavelength band) of Walnut leaves at various locations in the canopy were made in the morning and afternoon to determine whether the laboratory measurements made on excised tissue were representative, and to determine whether the leaf properties varied during the day. The insitu field measurements and lab measurements of spectral properties were comparable in the 400 to 1000 nm wavelength band considering typical spatial variability and effects of age and dust. Therefore the lab measurements were considered reasonable in the 1000nm to 2400 nm wavelength band because the insitu measurements did not cover this wavelength band. The Spectron SE-590 failed at midday so the afternoon measurements that were taken were useless. The following table is a summary of our best estimate of the Walnut leaf spectral properties for old and young leaves in the 100% and 33% ET treatments including sun and shade leaves. TABLE 1. Walnut leaf reflectance and transmittance from 1987 measurements at the end of the field experiment. The treatment includes leaf age (current year=new, other=old), irrigation treatment (100% or 33%), and location of leaf (upper leaf in the sun=sun, lower shaded leaf=shade). Treatment MMR Chanel 1 2 3 4 5 6 7 Reflectance of top df leaf New,100%,Sun 6 12 7 43 43 35 24 Old,100%,Sun 7 11 8 51 51 41 28 33%,Sun 8 15 11 48 48 39 26 Old,100%,Shaded 7 11 9 44 44 36 24 Reflectance of bottom of leaf New,100%,Sun 9 18 12 43 43 35 23 Old,100%,Sun 9 18 13 50 50 41 28 33%,Sun 11 20 15 48 48 39 26 Old,100%,Shaded 9 16 16 45 45 36 25 Transmittance of leaf New,100%,Sun 3 13 7 47 47 38 26 Old,100%,Sun 1 3 2 33 33 26 18 33%, Sun 1 6 4 35 35 28 19 Old,100%,Shaded 1 6 3 40 40 32 22 -2

The Walnut stem dielectric measurements indicated a large change in xylem water content over a diurnal cycle of transpiration. A simple stem water conduction model was constructed to accommodate both radial and longitudinal movement of water as a function of transpiration rate. Work on this model was stopped when Ray Hunt began work on the project with JoBea Cimino because he had considerably more expertise in this area and apparently was going to do that modeling. Estimating the leaf-area-index (LAI) of the Walnut orchard is very difficult so we analyzed several fish-eye photographs that had been taken by Susan Ustin to obtain an indirect estimate of LAI. The average LAI from six fish-eye photographs was 2.0. David Goldhammer also made gap-fraction measurements at solar noon in 1987 using a large gridded canvas, and our estimate of LAI from his measurements is 2.6. These values are much lower than the LAI measurements obtained from directly sampling trees; direct sampling yielded LAI=3.7. The reason for this difference may be associated with the clumping of vegetation on branches and trees. The Cupid BRDF model was originally limited to canopies with leaf azimuth angle distributions that were symmetric about the points of the compass. We have modified Cupid to include azimuthally asymmetric leaf azimuth angle distributions. The BRDF for a canopy can be quite sensitive to whether the leaf azimuth angle distribution is symmetric. The Walnut orchard appears to be reasonably symmetric about the azimuth from canopy structure measurements so this refinement may only provide a small improvement in model predictions. The main objective of this research has been to compare predictions of canopy BRDF with measurements from the Walnut orchard experiment. Unfortunately only one nadir set of data has been made available so that is the only comparison work that we have done. The one set of nadir-view MMR measurement results that have been provided to us are for a single observation time between 1500 and 1530 Pacific Daylight Time on August 24, 1987. This data is from the 100% plot and consists of the average of 720 measurements in seven wavelength bands. These results are compared with predictions of the one-dimensional Cupid model (Norman et al. 1985) for leaf-area-indices of 2.0 and 3.7 using measured leaf spectral properties and estimated soil reflectances (Table 2). Leaf spectral properties varied with age and height and average properties were used in the model. The fraction of incident radiation above the canopy that is in direct beam is also required as an input for Cupid. Unfortunately this is very difficult to measure in each MMR wavelength band and so was not available in the data set. A detailed study at Nebraska in 1982 provided estimates of direct-beam fractions in the seven MMR wavelength bands for clear skies over a range of solar zenith angles. These values should be appropriate for clear skies at Fresno, California and are tabulated in Table 2. -3

TABLE 2. Average measured leaf reflectance and transmittance for seven MMR wavelength bands on Walnut leaves and estimated dry soil reflectances. Beam fractions of incident radiation were estimated from clear sky observations made in Nebraska in 1982 and should be appropriate for clear skies in California at similar zenith angles. MMR Wavelength Bands 1 2 3 4 5 6 7 Leaf Reflectance 6.5 11.0 7.0 45.0 42.5 34.5 22.0 Leaf Transmittance 2.5 10.0 5.0 42.0 41.5 32.5 22.5 Soil Reflectance 5.0 6.0 8.0 14.0 21.0 21.0 13.0 Beam Fraction 0.80 0.86 0.89 0.91 0.93 0.95 0.97 The results of the comparison of nadir measurements with predictions from the Cupid model are summarized in Table 3. Most of the model results fall within the range of the measured mean plus or minus one standard error. This is surprisingly close agreement between a one-dimensional model and measurements over an orchard that has distinct rows. The effect of the LAI uncertainty is minor because at an LAI of 2.0 only a small amount of soil is apparent anyway. TABLE 3. Comparison of nadir measurements of canopy bidirectional reflectance with predictions from the one-dimensional model, Cupid, in seven MMR wavelength bands on the 100% Evapotranspiration treatment in a Walnut orchard (Caldwell and Vanderbilt, 1989). MMR Wavelength Band 1 2 3 4 5 6 7 Measurements 2.8 4.5 4.0 25.8 26.1 15.0 6.6 (std. error) (0.9) (1.4) (1.3) (9.0) (8.0) (4.6) (2.1) Cupid Model LAI=2.0 2.2 3.8 2.6 23.1 23.5 16.6 8.7 LAI=3.7 2.0 3.8 2.3 27.1 25.6 16/8 8.5 -4

References: Caldwell, W. and V.C. Vanderbilt. 1989. Tree canopy radiance measurement system. Optical Engineering 28:1227-1236. Norman, J.M., J.M. Welles, and E.A. Walter-Shea. 1985. Contrasts among bidirectional reflectance of leaves, canopies and soils. IEEE Trans. Geoscience and Remote Sensing GE-23:659-667. 3) Planned Activities The primary activity for the upcoming year of research will be to compare the Cupid model predictions of BRDF with measurements from the Walnut orchard. This cannot be done until some measurement results are available. At the present time the model appears to be working adequately for nadir predictions but we have no indication about off-nadir comparisons. Originally we had planned to invert the Cupid model using measured BRDF data as another part of the research through cooperation with Dr. Naren Goel of New York. However because of problems in the contract office at NASA this work was only funded the first year. In the second year we lost the funds to do this work because of some technicalitsy in NASA. It has never been clear why we had no problem the first year and then lost the funds the second year. Therefore we are not able to do any inversion work on Cupid. The second activity is to continue to work with Susan Ustin to characterize the canopy of the Walnut orchard. We plan to modify a program that will estimate leaf area density and leaf angle distribution for a heterogeneous canopy using a three-dimensional kernal. The input for this work will be digitized fish-eye photographs from the Walnut orchard. 4) Personnel Supported Wayne Polley (Univ. of Nebraska)- Partial support from this grant to develop the stem flow model and process the leaf reflectance and transmittance data for Walnut leaves. Naren Goel (SUNY-Binghamton)- Consultant the first year to assist in the inversion of optical models such as Cupid and begin work on the joint inversion of RADAR and optical models. He participated in three meetings of the research group and demonstrated that the Cupid model is invertible. Jia-Lin Chen (Univ. of Wisconsin)- Assisted in the adaptation of the Cupid model to run on the Unix computer at the University of Wisconsin (Cupid was originally written on an IBM computer) and modified the Cupid model to accommodate azimuthally asymmetric leaf angle distributions. -5

5) Publications Martens, S.N., S.L. Ustin and J.M. Norman. 1990. Measurement of tree canopy architecture. Submitted to International J. of Remote Sensing. Ustin, S.L., S.N. Martens, J.M. Norman and D. Goldhammer. 1988. Measurement and characterization of tree canopy architecture. IGARSS. Edinburgh, Scotland. Sept. 13-16, 1988. ESA,SP-284. Vol. 3 p.1753. Abstract. Cimino, J., J.Paris, M.C.Dobson, D.Gates, J.A.Weber, F.T.Ulaby, S.L.Ustin, V. Vanderbilt, J.Norman, R.Lang and E.Kasischke.1988. Eos synergism study: Overview and objectives. Abstract. Cimino, J., J.Paris, M.C.Dobson, F.T.Ulaby, J.A.Weber, V.Vanderbilt, S.L.Ustin, J.M.Norman, R.Lang, E.Kasischke and R.Hunt. 1989. Synergism study: Implication of diurnal change in vegetation canopies on Eos orbit selection. IGARSS. Vancouver, B.C., Canada. July 1989. Abstract. -6

LUNIEERSITY (OF C(:ALIFORNIA, DAV'IS ISI RKFl-lE\ * I)\\ 1 iR\1\1 * I >* \\.1 1 1. ~ 1\\ 1 * 1 *.;\ [R\\ -*\\I \ \1\R * * \\i i;: )EP-\RTNIE\T O(F Bo()T\\) D.AV S. C \LI F()R \I. X5 i F\x,\ —:~,4 December 4, 1989 Dr. Fawwaz T. Ulaby Professor and Director Center for Space Terahertz Technology College of Engineering University of Michigan 3228 EECS Building Ann Arbor, MI 48109-2122 Dear Dr. Ulaby: Please accept the technical progress report you requested for our NASA subgrant under the "Extended Ecosystem Signatures with Application to Eos Synergism Requirements" program. Funding for the research effort under this program has been subsidized from other research grants to Dr. Vanderbilt and myself. Analysis to date of the MMR data has required about 8 person-months, analysis of the canopy geometry data and associated programming has required about 24 person-months, about 3 person-months to design and build the optical measurement system, about 2.5 person-months of contract administration (including field station reports to the Kearney Agricultural Center), about 12 person-months for the Kearney field experiment (including subsequent data collection), about 1 personmonth for the Duke and Shasta field experiments, and about 6 person-months in manuscript preparation and abstract/presentation preparation for scientific meetings. We anticipate about 3 more months of work are necessary to analyze the MMR data and about 12 more months to complete the canopy analyses. We do not know how much time will be required to properly analyze the AVIRIS data since it is a relatively new instrument and we will be using new software and new computers for the analyses. Due to the large number of spectral bands the time to fully explore the data could easily exceed the entire budget for this grant. We plan to minimize the time by restricting our approach to a comparison among coregistered scenes having the same solar angles. We will provide some further exploration of the data structure but most of this work will be postponed until FY90-91 funding is available. If you need further clarification of the status of work performed under this grant, please do not hesitate to contact me. Sincerely, Susan L. Ustin Research Botanist

RE: Program objectives and review for U. Michigan subcontract # 204272, "Measurement and modeling of canopy geometry for remote sensing applications," under NASA grant NAGW-ll0l, "Extended ecosystem signatures with application to Eos Synergism Requirements." Susan L. Ustin, Principal Investigator and Vern C. Vanderbilt, COPrincipal Investigator. 1. Objectives. The objective of this subgrant has been to participate in field experiments that measure, test, and validate remotely sensed optical models for the detection of canopy structure and water relations. To date, the emphasis has been on developing an efficient strategy and appropriate techniques for the measurement of forest vegetation data and of optical datasets. These data will be used in optical and microwave inversion models and in field experiments. The focus of current FY 89-90 emphasis is to (1) complete analyses on the orchard experiment and to (2) test the optical models in a forest environment using an aircraft data set and to (3) collaborate with the microwave modelers in the project, in the development of joint algorithms for the extraction of vegetation biophysical characteristics. These research objectives provide two key elements that are necessary to the overall goals of the Eos Simultaneity project: First, provide the canopy architectural and water relations information for testing the predictions of optical and microwave inversion algorithms. Second, provide analysis of the optical data sets from the walnut orchard experiment and the Shasta experiment. The results of these data analyses will be used in collaboration with other PI's to develop joint optical-microwave inversion algorithms. 2. Major Accomplishments. 2.1. Specific Accomplishments. 1. Participated in the design of, and collection of, field datasets for describing canopy architecture and water relations for the walnut experiment. Provided the student crew for the experiment. 2. Designed and built a device for making the optical measurements for the walnut experiment. Participated in the collection of the optical datasets of the walnut experiment and provided 14 student assistants who participated in the field data collection. 3. Provided data entry, quality control checking, and statistical data analysis for all canopy geometry and optical datasets from the walnut experiment. 4. Published one paper on the sampling design and canopy

radiance measurement system, one paper on the canopy water relations, and have submitted one paper for publication describing the canopy geometric properties from the walnut experiment (included in the appendix). 5. Acquired field spectra of loblolly pine branches as part of the Duke Forest Experiment. 6. Acquired fish-eye photos of Mt. Shasta study sites as part of the characterization of the stand structure. Acquired field water relations, Regan radiometer and reflectance calibration data at Mt. Shasta as part of the AVIRIS water relations experiment. 7. Acquired dataset to compare the use of four commercially available optical sensors for indirectly measuring leaf area index and mean leaf angle distributions from the walnut orchard and from a Ponderosa pine forest. 2.2 Detail Statement. The first three publications from this research are included in the appendix, "Canopy radiance measurement system," by W. Caldwell and V.C. Vanderbilt (Opt. Eng. 28: 12271236), and "Measurement and modeling of tree canopy architecture," by S.N. Martens, S.L. Ustin, and J.M. Norman (submitted, Int. J. Remote Sens.), and "Water relations of a walnut orchard: Simultaneous measurement with remote sensing," by J.A. Weber and S.L. Ustin (IGARSS'88, ESA SP-284 Vol. 3: 1749-1751). These three papers, and related abstracts describe the methods of measurement used to describe the optical, geometric, and water relations properties of the walnut orchard. The design and measurement of the orchard was complex due to the fact that the row structure produced a highly non-random periodicity in the datasets. Water relations properties change diurnally and spatially vary with canopy position and sun angle. Adequate statistical sampling of the orchard required obtaining relatively large datasets for these variables, which were acquired over the two-week experiment, and which have since required significant amounts of time to qualitycontrol check all data. Most of the time spent so far has been in validating the data quality, but which now has been completed. More synthetic analyses of the results are presently underway. We began work on the second field experiment last summer. We obtained field reflectance spectra of foliage of loblolly pine by needle age-class and by stand age at the Duke forest. These data will provide a description of the variation in optical spectral signatures as a function of foliage age and stand age (an indirect measure of soil nutrient status). This work is part of a broader investigation of environmental factors controlling spectral signature components in conifer foliage. Initially, this data was to have been used in connection with analysis of AVIRIS data from the site, however, due to technical problems with the aircraft instrument, the AVIRIS data was not obtained at the Duke forest. Eight flightlines of AVIRIS data were acquired at Shasta, CA during September, 1989. We designed the flight so that we could examine

whether diurnal changes in reflectance could be attributed to changes in canopy water status. We obtained field calibration spectra, regan radiometer data, and foliage water potential data at one of the Shasta sites, to be used for calibrating and analyzing AVIRIS data. 3. Planned Activities. 1. We plan to complete the statistical analysis of the optical datasets from the walnut experiment during this budget period. This will include analysis of the spectral signatures of walnut leaves and correlation with leaf biochemistry, and analysis of the canopy reflectance properties as a function of position and time of day. These data will be compared with the results of the geometry study for examining and validating optical inversion models. 2. We will determine the vertical distribution of canopy geometry (branch size class distributions, foliage and fruit distributions as a function of height) for 2 class (trunk, canopy) and 3 class (trunk, low and high canopy) groupings. We will determine the probability density distribution function for the vertical classes. 3. We will complete analysis and submit for publication, our comparison of different methods for indirectly sampling canopy geometric properties. Most of the data analysis has been completed, with the exception of running a 3-d canopy inversion model on the datasets. We have implemented an analysis program for analyzing the data to calculate leaf area index and mean tip angle distribution, assuming a l-d model. We plan to implement Wells and Norman's 3-d model on the computer system this winter. 4. We will obtain AVIRIS data and initiate analyses of the Shasta dataset this year. We have initiated a collaboration with Dr. Youguan Kou (Environmental Systems Research Institute) who will be coregistering the AVIRIS flightlines and will digitize a topographic map of the Shasta study site. We will analyze the AVIRIS datasets for diurnal patterns associated with alterations in the water status of the forest canopy. Field datasets to be used for validation and calibration have been analyzed. 4. Personnel Supported. The following personnel were supported by this project. Scott Martens will receive his Ph.D. in Dec. 1989. Name Status Project Responsibility Principal support: Susan Ustin PI canopy measurements, leaf spectra, AVIRIS data analysis, administration of grant

Vern Vanderbilt PI MMR measurements and analysis, AVIRIS data analysis Rita Pettigrew assistant MMR data analysis Scott Martens Ph.D. Student canopy data analyses expected degree 12/89 Robert Rousseau assistant canopy data analyses, AVIRIS data analysis Guy Cook assistant microwave data analysis Short-term support: Walnut field experiment: Kevin Berger undergrad MMR measurements Jenny Clark undergrad MMR measurements William Caldwell undergrad MMR measurements Peter Collins undergrad canopy measurements Aram Derowetski grad canopy measurements Carol Hotton grad canopy measurements Paul Rich postrdoc canopy analysis software Doug Ryden assistant microwave data analysis Ken Severin grad canopy measurements John Shin undergrad canopy measurements Jatinder Singh undergrad canopy measurements Curtis Smith undergrad canopy measurements; data entry Erik Ustin undergrad canopy measurements Jerome Ward post-doc fish-eye photo analysis Monica Wiley undergrad water relations Nicole Wiley undergrad water relations Collaboration (without support): Dr. Brian Curtiss spectral signatures of conifer U. Colorado foliage Dr. Barry Ganipole MMR data and canopy analysis U. Arizona Dr. Youguan Kou GIS database Environmental Systems Research Institute 5. Publications. The following publications and abstracts have resulted from the research funded under this program. We anticipate several additional publications from this work during the FY-89-90 year that provide additional analyses of the canopy architecture and reflectance properties from the walnut experiment.

PUBLISHED PAPERS AND MANUSCRIPTS 1990 Martens, S.N., S.L. Ustin, and J.M. Norman. Measurement and characterization of tree canopy architecture. (submitted, Int. J. Remote sens.) 1990 J. Way, M. Schier, N. Christensen, C. Dobson, D. Gates, F. Ulaby, E. Kasischke, R. Lang, J. Norman, J. Paris, S.L. Ustin, V.C. Vanderbilt, and J. Weber. Eos Synergism Study: Diurnal Change in trees as observed by optical and microwave sensors. (to be submitted to Trans. Geosci. and Remote Sens.). 1989 William Caldwell and V.C. Vanderbilt. Canopy radiance measurement system. Opt. Eng. 28: 1227-1236. 1989 Curtiss, B. and S.L. Ustin. The remote detection of early stages of air pollution injury in coniferous forests using imaging spectrometer. Int. J. Remote Sensing (in press). 1988 Weber, J.A., and S.L. Ustin. Water relations of a walnut orchard: Simultaneous measurement with remote sensing. IGARSS'88, Edinburgh, Scotland, UK, Sept. 13-16, 1988. ESA-SP284 Vol. III: 1749-1752. PAPERS PRESENTED AT SCIENTIFIC MEETINGS 1990 Ustin, S.L., V.C. Vanderbilt, J. Way, F. Ulaby, C. Dobson, and D.M. Gates. IGARSS'90. Opportunities for Eos instrument synergism in monitoring of forest ecosystems. Geosci. and Remote Sens. Symp. University of Maryland, College Park, MD May 20-24, 1990. 1989 Cimino, J., J. Paris, C. Dobson, F. Ulaby, J. Weber, V. Vanderbilt, S. Ustin, J. Norman, R. Lang, E. Kasischke, and R. Hunt. Synergism Study: Implication of diurnal change in vegetation canopies on Eos orbit selection. IGARSS'89 Geosci. and Remote Sens. Symp. Vancouver, B.C. July, 1989. 1989 Ustin, S.L. and S.N. Martens. Measurement and characterization of tree canopy architecture. Ecol. Soc. Am. 74th Ann. Meeting, Toronto, Canada. August 6-10, 1989. Bull. Ecol Soc. 70: 285. 1988 Ustin, S.L., S. Martens, J. Norman, and D. Goldhamer. Measurement and characterization of tree canopy architecture. IGARSS'88, Edinburgh, Scotland, UK, Sept. 13-16, 1988. ESA SP-284, Vol. 3: 1753. 1988 Weber, J.A., and S.L. Ustin. Water relations of a walnut

orchard: Simultaneous measurement with remote sensing. IGARSS'88, Edinburgh, Scotland, UK, Sept. 13-16, 1988. ESA SP-284, Vol. 3: 1749-1752. 1988 W. Caldwell and V.C. Vanderbilt. Field system to determine reflectance of tree canopies. IGARSS'88, Edinburgh, Scotland, UK, Sept. 13-16, 1988. ESA SP-284, Vol. 3: 1756. 1988 Cimino, J., J. Paris, M.C. Dobson, D. Gates, J.A. Weber, F.T. Ulaby, S.L. Ustin, V. Vanderbilt, J. Norman, R. Lang, and E. Kasischke. Eos Synergism Study: Overview and Objectives. IGARSS'88, Edinburgh, Scotland, UK, Sept. 13-16, 1988. 1988 Curtiss, B. and S.L. Ustin. The remote detection of early stages of air pollution injury in coniferous forests using imaging spectrometry. Proc. European Joint Research Center Remote Sensing of Forests Workshop. Ispara, Italy, Sept. 4-6, 1988.

November 1989 Volume 28 Number 11 ISSN 0091-3286 t' Lt-' The Intemational Socey For Optical Ernegerirng ~.~ -.:- S - -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~. - zx ~~~ ~~~ ~~~ ~~~ ~~~ ~~~ ~~~~~~ ~~~ ~~~ ~~~ ~~~ ~~~ ~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -t; ^- {.r,'; tiri s:,. _ -vd... w-:,3 ~ -—'.: —;..~.''',.-.:'"II...... / —.,... —:'I=I-_.I _G~.~.' -' * -?'"'*"-' ts -"-".:'..' " -',,7 — *._'" -.;_ | "'|'~~~ ~.:-:I~":-::~..~'" I'. -',,,~ ~,' I I —, Polnt w' -- ourcesz~ l7u4 $w - > | H-of r a.'"..O c E e -... Graient InexLeses i~L |.. -4 Optiald Desig Plan A HOloptic pcal Busese -. /) Pattern Synthesis-'.....................,,.' ~ -' More -.-"-'' —' —.-.='=" —" —";-,~,?.:'f: /r It Target Acqulitlon —;F, 3~ b.;.vc fRS l Radince — ~_ DATA RECORING SYSTEM y~~~~~~~~~~, _, Point Spread Function ______________________ Interkrometrkc..... - -Tomo HelmetMIounted Dlsp. /ysL..;.... -':-. s.-'::....':=.~.'.;-is....:;'. -;...'._./.......'w.............-B w o S w 8i SPIE Reports-Book Reviews - - Short Courses-Meetings?::.t:.. - --- -- r - _'-* *:c -—..'":~''T 3 C~~~:- ~~'~" j Hoifu~ a- hl"' cl e n e... I-T Om d ~~~~~~~//~......:~II"~Tc~~~~~~~~~~~~~~~~~:~ ~~~ J~~~' ~'-iS L~j _T-"-7 —,-:::': ~'- ~ -~- --'~'~:::~::?:.-., ~ ~ -~'/;/r truer ~pacs' ".:,.:: *..:-,y,-.::':'~~~~~~~~~~~~~~~~~~~~~~~~ -TT- ~ ~ ~ ~ ~ ~~.:..!1 Rainc k- -': t~~ ~....... I~~~~~... o~-~?.r. -4.. ~-~~~~~~~~~~ ~-~~~~~~~~~~~~~P~~~~~~~i~~~~~~~i~~. _. ~ li~: -L~;' -: -...~ ~~~~~~~~~~~~~~~~~~~~~~~~~, ~~.~L. - s~C)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~i~:~ t~',-' ~

'ree canopy radiance measurement system illiam Caldwell Abstract. A system is described for obtaining both an estimate of the,SA Ames Research Center spatial mean bidirectional reflectance factor (BRF) for a tree canopy (disechanical Systems and Controls Branch playing a horizontally heterogeneous foliage distribution) and the statisS 213-4 tical significance of that estimate. The system includes a manlift supportrffett Field, California 94035-4000 ing a horizontal beam 7 m long on which are mounted four radiometers. These radiometers may be pointed, and radiance data acquired, in any of C. Vanderbilt, MEMBER SPIE 11 view directions in the principal plane of the sun. A total of 80 data iS Tecthnology, Inc. points, acquired in 3 min, were used to estimate the BRF of a walnut S 242o4 orchard 5 m tall and detect true differences of 12% of the mean approxi)ffett Field, California 94035-4000 mately 90% of the time. Subject terms: radiance; bidirectional reflectance factor; bidirectional reflectance distribution function; tree canopy. Optical Engineering 28(11), 1227-1236 (November 1989). NTENTS ments is a sampling problem. It requires estimation of the statistroduction tics of the signal: a radiance L(i;r,x,y) that varies as a function of -ee canopy radiance measurement system both horizontal (x,y) location and sun and view (i:r) directions. 1. Design objectives In remote sensing and plant sciences, such reflectance data 2. Description of tree canopy radiance measurement system are needed for developing and testing mathematical models of 2.2.21. Horizontal cantilever beam the radiation regime in plant canopies.2 These models are used 2.2.2. Support frame in parameter studies to gain understanding of remotely sensed 2.2.3. Control lever images from satellite sensors and thereby to address the more 2.2.4. Sundial detailed problems of discriminating species of plants, determinnopy reflectance ing their areal extent, and assessing their morphological and!. Calibration I. Bidirectional reflectance factor (BRF) of the canopy physiological condition. Such information will support future st and evaluation of system: walnut canopy reflcn efforts in earth system science..4. Data acquisition Part of the variability in a set of radiance data of a plant i. BRF of a periodic structure canopy depends on the horizontal (x,y) location at which each 1. Data analysis measurement was made. The variability provides an indication 4.3.1. Spatial variability of data of how representative the measured portions of the canopy might 4.3.2. Statistical analysis of data be, compared to the entire canopy if it were measured. 4.3.2.1. Estimated BRF biased by track The spatial variability of the radiance decreases as the field 4.3.2.2. Statistics of cstimated BRF radiometric instrument [having generally a field of view (FOV).Vegetation results of 10~ or 150] is elevated higher above the canopy. This is:omnrnendations because the size of the instrument's "footprint" on the canopy cnowledary nns (the area of the canopy within the FOV of the instrument) erences increases with instrument altitude, thereby including at higher altitudes a sample more representative of the entire canopy.' In TRODUCTION practice, the reduction in the variability of the data that comes tree canopy, as for any other plant canopy, determining the with increasing the instrument altitude is limited, since cost and bidirectional reflecance factor (BR) from field measure other factors constrain the maximum height of the ground-based platform supporting the instrumcnt. Companrison of radiance data acquired using a satellite sensor and a field radiometric.659 received Dec. 13. 1988; revised manuscnript received May 30. 1989; instrument requis consideration of their difrin ields of d for publication May 30. 1989. I Soclcy of Photo-Optucl Insurnentaton Engineers. view: fractions of a milliradian for the instantaneous FO' of a OPTICAL ENGINEERING / November 1989 / Vol. 28 No 11 i 1227

CALDWELL, VANDERBILT ~~......J.~~ POLYCORDERS LIFT GANTRY RADIOMETRIC MEASURING AND DATA RECORDING SYSTEM Fig. 2. Tree canopy radiance measurement system. measurement procedures.14 Data values obtained with the method should be directly comparable with published data sets for other plant canopies. Fig. 1. Operator's view of the tree canopy radiance measurement (2) The reflected flux should be estimated in the wavelength system above the walnut orchard. bands of the Thematic Mapper sensor on the Landsat series of satellites. Because several Barnes Engineering MMR model 12-1000 radiometers15 were available, this was the model that satellite sensor compared with as much as 200 for a ground- was used for these measurements. This model has an FOV based sensor. was used for these measurements. This model has an FOV based sensor.15" The second source of radiance variability, sun and view direc- of 15 tions, is canopy dependent.-"- For many canopies (those both (3) The design should allow a data acquisition sequence, horizontally and azimuthally isotropic, for example), measure- repeated at intervals of approximately 0.5 h, that includes ments indicate that both the "hot spot" and the "cold spot" (local measurements of the canopy viewed in the principal plane at maximum and minimum, respectively, in'the BRF2 zenith angles of 0~, + 150, ~ 300 ~ 450 ~ 600 and ~ 75~ found in the principal plane.>" The principal plane is defined The design should allow the sky radiance field to be measured in by the two vectors describing the sun direction and earth ver- 41 directions, the same positive zenith view angles (as used for tical. downward viewing) at azimuth angles forming the eight points Our main goal was to construct an apparatus to provide the of the compass. capability to sample the radiance and determine the mean BRF (4) The mean BRF of the tree canopy should be estimated of a tree canopy measured in essentially one sun direction and within a perod of approximately 3 mi at the a = 0.05 level of one view direction (i;r). statistical significance. We deemed the achievement of this goal a potentially daunting task because the tree canopy to be measured, a walnut 2.2 Descption of tree canopy radiance measurement orchard, displays significant horizontal morphologic variability. system Trees approximately 5 m tall are planted in a rectangular pattern, The tree canopy radiance measurement system (Figs. I to 5) 3.4 m x 6.7 m, which forms rows of foliage separating parallel consists of three parts: the radiometric measuring and data restrips of bare soil. Because the radiance variation was a function cording system, the manlift, and the gantry. The system for of horizontal location, use of the innovative PARABOLA in- measuring and recording the radiometric data (Fig. 2) includes strument, a radiometer quite cleverly designed for rapidly ac- four Barnes MMR radiometers, each capable of measuring the quiring data as a function of view direction, was not con- canopy radiance in seven wavelength bands (Table I). Paired sidered.6 with each radiometer is an Omnidata Polycorder (Omnidata This paper describes the apparatus, the tree canopy radiance Inc., Logan, Utah) for digitizing and recording the data and later measurement system, and provides an analysis of data collected transferring those data directly to a computer. using the system. The manlift (approximately equivalent to model MZ46A, Grove Manufacturing, Shady Grove, Pa.), a mobile aerial plat2. TREE CANOPY RADLANCE MEASUREMENT form, has a maximum vertical platform height of 12 m, a maxiSYSTEM mum horizontal reach of 9.8 m, and a maximum platform load 2.1. Design objectives carrying capacity of 340 kg. When radiance data were acquired, the manlift was operated from the basket and driven back and The rationale for design of the tree canopy radiance measure- forth along a road 25 m long. ment system was to provide the capability to determine the mean The gantry (Fig. 3) mounts to and extends 7 m horizontally BRF of tree canopies measured in as many as 11 view directions from the basket of the manlift. The gantry is designed to support in the principal plane of the sun. Data were to be acquired at both the radiometers and a motor-driven camera and to:'acilitate various times during a day. We had the following specific their rotation to coincident zenith and azimuth view directions. objectives: It includes a horizontal cantilever beam and support frame. Also (1) The measured reflectance values should be obtained by a on the gantry are the control lever (Fig. 4) and sundial (Fig. 5), method that is directly comparable with established field both used to adjust the view direction of the radiometers. 1228 / OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11

TREE CANOPY RADIANCE MEASUREMENT SYSTEM SUPPORT FRAME I HORIZONTAL CANTILEVER BEAM DRIVE RDM SHAFT i SUNDIAL B RADIOMETERS BALANCE i B ARM A A SHAFDRIVE j CONTROL LEVER ~ |I1 SUNDIA~L 1.2 3.4 + i~~~~~0.2;5.0 3 0.1 K 1.05 K 7.0 SgPACERS (12) GUSSETS (12) RADIOMETER BEARING: 12U~~~~~~~~~~~~ SHnOWN) I ~PLATE, (2 SHOWN)~BOLTED ~SEPARATEID BY SHIM STOCK GUSSETS (11) FROM SECOND PLATE.~'WLDED / LiNTLSTEEL ]MOUNTED TO INSTRUMENT RACCK. TUBE, 6061 TB ALUMINUM ROTATES ABOUT CENTER PIN 0.305 m (12 in.) OD S;ELF-ALIGNING..'',- /1.6 mm (0.062 in.) WALL PARALLEL BAR fLL~bLICN)N... \ /o P ARALLEL OAR rS~ 7= "10N~~~~~~~~~~~~~~~~~~~~ABOUT AXIS /.:,~,,,\ ~,.,~~.~,~;~ /TUBE 0302 11.676 i.) OD SHAFT PRESS FITTED INTO TUBE I r(HAFT / K " ~ A'I,:~', R RADIOMETER FRAME ~~/~31.24 BOLTS FIELD OF VIEW The gantry supports four radiometers for measuring the canopy radiance in any of 11 view directions (zenith angles) in the principal of the sun. Dimensions are in meters. PARALLELOGRAM LINKAGE Horizontal cantilever beam ATTACHES TO w-n, -- CONTROL LEVER orizontal cantilever beam (Figs. I to 3) consists of a / CONTROL LEVER trical tube, 7 m in length, with an instrument rack riveted BEANG I / udinally along the outside surface. The tube is constructed ROTATONAL 51 -T6 aluminum, welded longitudinally along the inside, RI-.E AXISO CONROL n outer diameter of 0.30 m, a wall thickness of 1.59 mm, SHAFT LEVER EARING total mass of approximately 29 kg (West Coast Tube and F Agoura Hills, Calif.). To increase the strength of the weld Lv G, L CONTROL LEVER:he instrument rack (Fig. 3, section B-B) is riveted directly XISO o ie weld seam, with one row of rivets on each side of the DRIVE SHAFT AWL Tn OF SUNDIAL,ched to one end of the horizontal cantilever beam is a ing flange. Bolted gussets brace the flange to the wall of Fig. 4. T he control lever, th rough rotation about two axes, points the indricalte s radi ometers in the direction defined by the axis of the pawi. inddcal tube. A short length of aluminum pipe (outer wai l er 30.163 cm x 15 cm lcngth x 1.11 cm), press-fitted cylindrical tube, provides strengtho o the tube in the area Each radiometer (Fig. 3, section B-B) is attached to the;ussets. The end of the 1.11 cm pipe (Fig. 3, section A-A) instrument rack with the aid of a radiometer bearing formed ded to prevent the concentration of stress at that location from two plates that pivot about a center pin and are thinly vail of the cylindrical tube. separated by lubricated shim stock. The radiometer bearings are OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11 / 1229

CALDWELL, VANDERBILT TABLE I. Estimates of the mean and CV of the BRF of the walnut canopy measured in seven wavelength bands having approximately the spectral characteristics of the Landsat Thematic Mapper. There are 180 measurements represented by the CV for all data and 20 by the CV and the nugget for track 5. Each measurement is defined as an average of the data collected simultaneously from the four radiometers at a point along the track traveled by the manlift; thus, 20 9ti i!measurements represent acquisition of 80 radlometric data points. The view direction is nadir, and solar zenith and azimuth angles are approximately 39~ and 238~. 0.4 RADIUS Estimated Coefficient of: mean variation (CV) TO ALIUGN SUNDIAL IN PRINCIPAL PLANE: bidirectional............. Wavelength reflectance All nugget ROTATE SUNDIAL UNTIL EDGE CASTS HNADOW ONJ LINE ~ \ / E Band range, factor (BRF), data, Track 5. track 5, ON ANGLE INDICATOR PLAT gm % % % %.PRINCIPAL PRINCPLA 1 0.45-0.52 2.8 30.7 15.9 10.8 2 0.520.60 4.5 32.2 1 5. 12.2 3 0.63-0.69 4.0 32.5 18.8 8.0 4 0.76-0.90 25.8 35.0 10.7 13.2 5' 1.15-1.30 26.1 30.5 9.9 10.7 i/L///j/ 8' t.61.55-1.75 15.0 30.5 11.2 9.7 7' 2.08-2.35 6.6 32.0 15.5 7.6 P TAWLOn FIT *These band numbers do not correspond to the band numbers of the CONTROL LEVER Landsat Thematic Mapper, which does not have a wavelength band from 1.15 to 1.30 iAm. CD.XIS Of ROTATION shaft and the sundial are attached to the support frame. Its Fig. 5. On the sundial, which is rotated to the solar azimuth, detents height, 1.5 m, allows elevation of the horizontal cantilever beam accept the control lever pawl and thereby allow the optical axes of above the top edge of the basket of the manlift permitting the the radiometers to be pointed in parallel in any one of 11 zenitht o f angular directions in the principal plane of the sun. gantry to be mounted without modification of the basket. The drive shaft, an aluminum pipe with a flange and bracing gussets welded to one end, attaches through a bolt ring to the mounted so that their axes of rotation are parallel, allowing the mating flange of the horizontal cantilever beam. The drive shaft radiometers various view directions toward or away from the attaches to the support frame through two pillow block bearings manlift. that permit the drive shaft/beam assembly to rotate about the The number of radiometers that can be mounted on the in- longitudinal axis of the drive shaft. strument rack is limited by their combined weight. At full load, The longitudinal axes of the drive shaft and horizontal cantithe system is designed to accept six Barnes MMR radiometers, lever beam are parallel but are offset 15 cm. The offset places each 5 kg, spaced equally along the length of the beam and one the center of gravity of the assembly (drive shaft, horizontal motor-driven 35 mm camera mounted at the end of the beam. At cantilever beam, radiometers, and camera) closer to its axis of full load, the system is designed to withstand 4 g dynamic loads rotation (the axis of the drive shaft), and thereby facilitates applied in the vertical direction. At the end of the beam, the rotationally balancing the system with the addition of minimal vertical deflection due to bending in the horizontal cantilever amounts of dead weight. Balancing is accomplished by adjustbeam and in the drive shaft is calculated to be 5 cm at static full ing both the angle of the balance arm (Figs. 1 to 3), clamped to load. project radially from the drive shaft, and the radial position of To prevent secondary illumination (e.g., due to sunlight spec- the balance mass, which is bolted to the balance arm. ularly reflected by the side of the tube) from reaching the target-particularly the calibration panel-a black, coarse mesh of 2.23. Control lever plastic material (medium Weathashade, Weathashade, Inc., All radiometers may be pointed simultaneously to any desired Apoka, Fla.) was sewn to form a sock around the tube. To zenith and azimuth view direction by moving the handle of the further reduce the possibility of unwanted specular reflection, control lever (Figs. 3 and 4) about either or both of its two the mesh on the sides of the tube was sprayed with matte black orthogonal axes of rotation. When the handle is moved about the paint containing a carbon black pigment. first axis, it rotates the drive shaft/horizontal cantilever beam assembly and the attached radiometers about the longitudinal 2.2.2. Supportjrame axis of the drive shaft. When the handle is moved about the The support frame (Fig. 3), a truss of aluminum angle and second axis, that of the control lever bearing, the radiometers channel beams, is anchored with 1.3 cm bolts to steel channel are simultaneously rotated with the aid of a parallelogram linkbeams in the floor of the basket of the manlift. Both the dr.ve age that connects the control lever to parallel bars attached to 1230 / OPTICAL ENGINEERING / November 1989 / Vol. 28 No.

TREE CANOPY RADIANCE MEASUREMENT SYSTEM:h radiometer. The control lever handle always points in the tion (i;r) is the flux from the canopy, measured both for direcection of view of the radiometers. tions (i;r) and within the solid angle of the instrument, divided by the flux that would be received from a hypothetical, perfectly!.4. Sundial white, perfectly diffuse, level calibration surface also measured h sundial (Fig. 5), used with the control lever, permits identi- for (i;r) by the same instrument. We shall assume that both the ition of the desired zenith view directions of the radiometers canopy and the calibration surface are equal in area. The flux he principal plane of the sun. To identify the principal plane, from the canopy is found as an integral of the flux < from each sundial is rotated until its top edge is pointed in the solar &xAy area: nuth direction. When oriented at this angle, the partially ruding top edge casts a sun shadow along a corresponding BRF = c (2a) painted on the angle indicator plate. The plate forms a Dcalibranon -circle centered at the intersection of the three axes of rota-, two of the control lever and one of the sundial. Proper rr ~ation of the sundial requires that the basket of the manlift be 1, ensuring that the rotational axis of the sundial is vertical ='-I (2b) thus that the principal plane is correctly identified. Once this e is identified, the desired zenith view direction is selected j (lrxy)dxdy loving the pawl of the control lever handle into one of the its located at 15~ increments in the angle indicator plate. The flux in the denominator of Eq. (2b) is a constant, indicating that the flux reflected by the calibration panel does not vary as a 4NOPY REFLECTANCE function of the (x,y) position and that it may be taken outside the Calibration integral. Dividing this constant calibration flux into the flux from the canopy [the numerator of Eq. (2b), which does vary ration of all radiance data to determine the BRF was ac- with (x,y) position], we find that the BRF of the canopy is the lished with reference to a field calibration standard, a integral of the value of the BRFy at each point above the canopy )4 painted, 1.3 m x 1.3 m aluminum panel. (The results normalized by the area of the calibration panel. But we have ported as BRF and not as a bidirectional reflectance distri- assumed that the measured areas of both the calibration panel i function, an alternative descriptor of the light scattering and the canopy are identical; thus, rties of the canopy. 16) The BRF of the field standard was nined from comparison with pressed BaSO4 measured in boratory. Nicodemus et al.l6 and Milton"4 describe the ions and procedures for obtaining the reflectance factor. BRF(l;i;r) =. (3) g measurement, the field calibration panel was level and fdxdy,cated on top of a 7 m tower, 2 m higher than the tree y, so that light scattered by foliage would not illuminate gel, thereby affecting the calibration measurements. Durta collection, a recording pyranometer was used to moni- 4. TEST AND EVALUATION OF SYSTEM: WALNUT solar insolation, providing assurance that the quality of CANOPY REFLECTANCE:ctral data was not compromised due to clouds. BRF of the plant canopy measured in each wavelength 4.1. Data acquisition,as approximated as a biconical reflectance factor deter- To evaluate the tree canopy radiance measurement system, data zy using the formula were acquired at the Kearney Agricultural Center (at Parlier, Calif., 360 36' north latitude, 119~ 31' west longitude) of the;i;r,x,y) =.op - BRF, (1) University of California at Davis on Aug. 24, 1987. Data were VJ - Vdwk collected on an irrigated orchard (Fig. 6) of six-year-old black walnut trees (J. hindsi') of the "Chico" J. regia (English) va/canopy, Vcal, and Vd are the voltages of the radiometer riety. The localized spray irrigation system had been operated te orchard, calibration panel, and zero light (instrument during the night prior to data collection, leaving the soil surface;ponse) were measured. In Eq. (1), the letter 1 signifies a mottled mixture of wet and dry, dark and light areas. igth band, (x,y) is a horizontal position relative to the The trees were approximately 5 m tall and planted 3.4 m apart and the i and r correspond to the directions in which in rows 6.7 m wide. The tree crowns were closed (i.e., contigincident and reflected, respectively. BRF<,, is the BRF uous) within the rows but not between rows; the ground cover Vibration panel measured in the laboratory at an incident was estimated to be 65%. tion angle of 100 and radiometer view direction of nadir. Spectral data were collected periodically in time as the manlift: of the panel was not corrected for the incident illumin- with the four radiometers 1 1 m above the soil was driven slowly;le. Each radiometer measurement of the orchard at one back and forth on eight parallel tracks on a short roadway time was calibrated using a linear interpolation pro- adjacent to the walnut canopy (Fig. 6). At each of the along-.pplied to calibration panel measurements acquired at track points at which data were acquired, the outputs of the Fore and after collection of the orchard measurements. radiometers were sampled in each wavelength band. Eighty data points were acquired on each track in each wavelength band, 40 irectional reflectance factor of the canopy as the manlift was driven forward on the track, then 40 addipy BRF for specific directions of irradiation and reflec- tional points as the manlift was driven backward on the same OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11 / 1231

CALDWELL, VANDERBILT formula weighted equally data from four radiometers, located as shown in Fig. 3, and did not require use of a fifth radiometer located at x = 6.7 m, one period. All of this means that the ~ ~. integration, Eq. (3), can be approximated as 3.4m m 20 4 ~' Z Z BRFy(l;i;r;xjpk)A *4,i BRF(1;i;r), 80A (6) 80A -*.~ i >'.& * \\\x\\ \where A, the area of the field of view of each radiometer, may ~~ \0\\\& 1 be canceled from numerator and denominator and the pk are 8TRACKS positions at which data were acquired in the along-track direction traveled by the manlift. In the across-row direction, the xj *." are the positions of the four radiometers measured from the end 6.7m -- / of the horizontal cantilever beam (Fig. 3).;A'/a For purposes of statistical analysis, the two summations in Eq. (6) were performed separately, first yielding 4 Fig. 6. Plan view of walnut orchard overlaid with normalized differ- Z BRFxy(l;i;r;xj,p) ence vegetation index (NDVI,,) data. Black dots represent the loca- BRFy(i ) (7) tions of tree trunks; gray shading represents tree foliage. The radi-'... ance data were acquired in the nadir direction by the four radiometers of te tree canongpy rachiance mhteasur nt sstem as the where BRFy represents an estimate of the mean of the point-bymanlift was driven along each of eight tracks located on a road adjacent to the canopy. The tower used to support the white calibra- point canopy reflectance, BRFxy, averaged along the cross-row tion surface above the canopy was located off the map, 10 m south transect corresponding to position p,. The BRF of the canopy of the eight tracks. On the plot of NDVI,. the black area represents and the coefficient of variation were estimated from the average values from the minimum observed value (0.52) to 0.57, represented of 20 values of BRFy representing the data from one track: by the contour line. Subsequent contours represent values of 0.62, 0.67, 0.72, and 0.77. The white area represents values from 0.77 to the maximum observed value (0.82). The view direction is nadir and 20 illumination zenith and azimuth angles are approximately 39~ and Z BRFy(l;i;r;,) 238~. k- 1 BRF(;i;r) 20 (8) track. Nine data sets totaling 720 measurements were acquired 4.3. Data analysis in each of the seven wavelength bands in the nadir viewing direction between 3:00 and 3:30 p.m. Pacific Daylight Time 43.1. Spatial variability of data (22:00 to 22:30 GMT). To gauge the effect of the changing sun The magnitude of the (x,y) spatial variability of the data was angle during data collection, the last data set was acquired on the examined with the intent of documenting the need for a spatial same track used to acquire the first data set. All spectral data sampling methodology. To do this, the normalized difference were collected in the nadir view direction, one of 11 possible vegetation index (NDVI,,) was plotted against the rows of zenith angle view directions with the tree canopy radiance foliage that form the dominant feature of the canopy architecmeasurement system. ture. Here NDVIXy is defined from the BRF, y in the near infrared (band 4) and red (band 3) spectral regions: 4.2. BRF of a periodic structure A practical objective in the design of the tree canopy radiance NDVII, BRF1r(4;i;r,x,y) - BRF.y(3i;r,x,y) (9) measurement system was to minimize the use of radiometers, BRF,,(4;i;r,x,y) + BRF.y(3;i;r,xy). which are expensive to purchase and difficult to borrow. This goal was addressed by selecting the positions of the radiometers Often employed in remote sensing research, NDVI provides an to aid in sampl sting the oBRFs(lir;xy). Because of the ripro- m s indication of the photosynthetic capacity of a plant canopy. 17.18 to aid in sampling the BRFXy(l;i;rx,y). Because of the pronounced row structure of the canopy, we expected a priori that Typical values of NDVI range from near 1.0 for a dense, green, BREXy should be quasi-periodic in x in the sense that BRFX is actively photosynthesizing plant canopy to approximately 0.1 periodic in x; thus, the integral in the along-row direction y, for sparse vegetation growing under desert conditions. NDVI is most often computed using satellite sensor radiances rather than B.RF1( l; i r;x ) fJB RFx( l;i;rxy y, (4) reflectance factors. Our definition, involving BRF rather than jfOY'''~~~~~~~~ ~radiance, does not affect the results of the spatial analysis, but is periodic in x with a period of 6.7 m. Thus, some NDVIXy values reported here may differ from those that would be obtained from a satellite sensor measuring the walnut fBRFxy(l; i;;x.y)dy - JRF. y1(l;i r;x +6. 7 y)dy. (5) canopy. The NDVI of the canopy is determined by appropriately averaging the per-point NDVIy. Making this assumption of periodicity allowed us to modify the Figure 6 displays the spatial variability of the NDVIXY as a trapezoidal formula for numerical integration. The modified function of position in directions both across and along the rows 1232 / OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11 t

TREE CANOPY RADIANCE MEASUREMENT SYSTEM w.85 ~r.-'. 3~~C~r~'' -. 2.'.-' Z-1: r:.:,.,..x;.. ~:...,.,..,,-)- -2',.o, * 2,2520 N E.55 A, 4 BRF(6) - 13.13t 1667.7 Uj 45 5 OZ>;:47 1 10 E4 DISTANCE ACROSS ROWS2 m o BRF2) - -3.97t 4. 7. Variation of NDVIXY with distance across the rows of walnut X1 is. The viewing direction of the plotted data is toward the north. 3RF)r 4.1>+i$2 view direction of the radiometer is nadir and illumination zenith, the azimuth angles are approximately 39~ and 238~. RF~lI *.2U+ 216.6 2 walnut trees. Figure 7 displays the spatial variability of the 22:00 22:05 22:10 22:15 22:20 22:25 VI,, across the rows. As shown in Figs. 6 and 7, the TIME, hrn GMT est values (0.52) of NDVIY6 correspond to locations be- - 0 1 2 3 4 5 6 7 8 9 10 11 5O en the rows of trees, where the field of view of the radi- 1 2 3 5 6 7 8 1 tDr was more likely to include shadowed soil. The largest 7es (0.84) of NDVI,, correspond to locations on the rows of alnut 4 41 2 s where the field of view of the radiometer was most likely to SOLAR ZENITH ANGLE, dg ude sunlit foliage. Vhese roesults (Figs. 6 and 7) show that NDVIeY varied in a T 236 238 240:ial pattern corresponding to the row structure of the canopy. SOLAR AZIMUTH ANGLE, dog ratio of the largest value (0.84) of NDVIy correspond to the smallest 2) is 1.6; no one of these NDVIXy estimates, chosen ran- Fig. 8. Estimates of the canopy BRF for each track traveled by the fly, can be expected a priori to represent the NDVI of the mlrnlift. Wavelength bands are listed in Table I. Each equation is a enthe rows of tregression ofwhere the eight estimates of thed i canopy BRF for each,py. The results affirm the need for an appropriate measure- wavelength band. The data acquired at 22:26 GMT were excluded It strategy with which to sample this spatial variability. from the analysis. Note that the per-point NDVI, results (Figs. 6 and -r7), if they were spatially averaged, could be compared to a canopy 2. Statistical analysis ofdata NDVI estimated from the results in this figure. each wavelength band, two specific statistical issues were stigated. First, are the estimates of the canopy BRF biased both acquired on track No. 1, suggests that at least part of the Irding to track number? Second, in estimation of the canopy systematic decrease in the estimated BRF (Fig. 8) is due to, what are its statistical properties and how many data changing sun angles. Another factor to consider in understandts should be acquired? ing the decrease is an edge-of-field effect, possibly present in the data acquired where the radiometers viewed the row of trees.al. Estimated BRF biased by track bordeing the orchard. The downward trend is likely not due to a lefinition, the BRF of a plant canopy represents the entire rack-dependnt bias. Such a bias would tend to be oscillatory, py; it does not depend upon the (x,y) location of the sensor with a period fixed by the distance between radiometers, rather to acquire the data. The estimates of the walnut canopy than approximatesly linear as shown in Fig. 8. The results (Fig. do appear to depend upon the sensor location (i.e., the 8) suggest that the overall downward trend is due to both a: that the manlift traveled), decreasing, as Fig. 8 shows, changing sun angle ands an ldge-of-field effect. increasing track number. The downward trend in the BRF Evidence for an oscillatory, track-dependent bias in the estie canopy was statistically modeled for each wavelength mates of the BRE of the canopy was investigated with the aid of with the aid of a linear regression of the data (Fig. 8) as a the residual, the difference between the actual BRF and the BRF ion of time. The data from the revisit of track No. 1 at predicted using the regression equations (Fig. 8). The results S GMT were excluded from the need for analysis. (Fig. 9) shw that for each wavelength band, the residual norterpretation of the downward trend in the estimates of the malized by the mean, (BF - B)BRn F, varies up to 3.5 6 and py BRF is complicated by the fact that the data for all eight about the regression line (Fig. 8). Thc three factors-changes in s were collected during a 0.5 h period, and therefore the sun angle (Fig. 8), an edge-of-fid effect, and a track-depennumber (Fig. 8) corresponds both to a set of data acquisi- dent bias-must be considered for understanding the oscillatory ocations (the portion of thest canopy measured as the manlift pattern (Fig. 9). However, the spatial period of the pattern, of the led that track) and to the sun angles at which the data were approximately the distance 1.67 m between the radiometers on ired. Indeed, the fact that for each wavelength band ther is the horizontal cantilever beam, corresponds to a property of the ernct in the BRF estimated for th first and last data sets, measurement process rather bias would t ecanopy architecture or sun OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 1 1 / 1233

CALDWELL, VANDERBILT 4 T * I 1, | p -. TABLE II. Minimum number of measurements required for detecting WAVELENGTH BAND true differences among treatments using CV = 0.12 and an a = 0.0! 3 O 1 test of significance. Each measurement is defined as an average, b' 0 2 wavelength, of the four data points collected simultaneously fron 9e 2 * 3 the four radiometers of the tree canopy measurement radianci O 4 system. A 6 True difference as percent of mean 0(~~ Ot-~~~ —-— ~~~~s~~r —~~~-Power' 20 10 5.0 2.5 1.0 _-1... I number_ =P2 1- { / 0.80 6 23 91 362 2258 0.85 7 26 104 415 2592 -3 _. 0.90 8 31 122 487 3042 0.95 10 38 150 598 3733 -4 0.98 12 48 189 756 4724 22:00 22:05 22:10 22:15 22:20 0.99 14 55 218 872 5450 TIME, hrs GMT Power relates to the probability that we will detect a difference tha is there. 1 2 3 4 5 6 7 8 TRACK NUMBER remaining V3 of the CV for these data), is the value of impor. tance that determines the statistical significance and statistica. power associated with an estimate of the mean value of the data. DISTANCE, m The size of the nugget should not exceed the CV. Comparisor shows that two values (13.2% of band 4 and 10.7% of band 5,' Fig. 9. For each wavelength band, the residual of a linear regression do. This provides an indication of the potential inaccuracy ir of the data (Fig. 8), normalized by the mean, shows the percent deviation from the downward trend in the estimated BRF of the canopy. If the value 13.2% is disregarded, the largest value of the nuggel is 12.2%. The number of data points required to estimate the canopy angle changes. This supports the hypothesis that the procedure BRF was investigated using the approach to statistical powei for estimating the BRF of the canopy includes a bias dependent found in Howell.21 Table II shows the minimum number of on the track traveled by the manlift. The magnitude of this bias, measurements required for detecting true differences among less than 3.5% of the measured value, is comparatively small treatments as a function of power using an a = 0.05 test of and apparently predictable. If it is not removed from the data, significance. Each measurement is an average of the data from then it will add to the uncertainty in the estimates of canopy the four radiometers. It is assumed that the track-dependent bias BRF. The fact that the bias exists shows that the sampling has been removed from the estimates of BRF. The results are strategy developed for locating the radiometers along the length predicated on the approximate upper bound on the value of the of the horizontal cantilever beam (Fig. 3) may be improved by nugget (12%) in Table I. These numbers represent in effect a taking advantage of a priori knowledge of the BRFy. lower limit to the number of measurements required to estimate a mean for data having the type of "white noise" properties 4.3.2.2. Statistics of estimated BRF represented by the nugget. Values of a and power were selected Table I shows the mean (i.e., the canopy BRF) and coefficient of to obtain a low probability of finding a true difference in the variation (CV) of BRFy [Eq. (7)], both determined from all mean (the canopy BRF) that is not there (therefore a = 0.05) BRFy from all tracks, providing a total of 180 data points. View and a high probability of finding a true difference that is there direction is nadir. Solar zenith and azimuth angles were approxi- (therefore values of power > 0.8). mately 39~ and 238~ (Fig. 8). The mean shows the expected5' 7' The results, Table II, demonstrate that a comparatively large "green vegetation" variation with wavelength, while the CV for data set must be acquired if the reflectances of two similar all data is almost a constant, varying only from 31% to 35%. canopies are to be distinguished. For example, if we acquire 31 Table I shows that the CV for track 5, the track with the largest measurements (each defined as an average of the data of the four individual CV, ranged from 9.9% in band 5 to 18.8% in band 3. radiometers, 4x31 = 124 total data points) on each of two Both periodic and random spatial variation of the canopy treatments for which the true means (the canopy BRF of each reflectance contribute to the magnitude of the CV. This spatial treatment) differ by 10% and use an a = 0.05 test of statistical variation in the data of each track may be characterized with the significance, then 90% of the time we will reject the null hyaid of the semivariogram. 9.20 Table I shows that the value of pothesis and accept correctly the hypothesis that this difference the nugget of the semivariogram for each wavelength band for exists. Or 95% of the time we will accept correctly the hypothtrack 5 was approximately ~ of the value of the CV of the track. esis that a true difference of 5% exists in the means, if we This suggests that approximately ~3 of the CV is attributable to acquire 150 measurements and again use an a = 0.05 test of microscale variability of the data, which represents both scene significance. variability at a higher spatial frequency than the sampling fre- We actually acquired 20 measurements on each track, which quency and measurement noise. The size of this "white noise," means that we can detect true differences of 12% with a power rather than the size of the spatially correlated variability (the of 90%, using an a = 0.05 test of significance. 1234 / OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11

TREE CANOPY RADIANCE MEASUREMENT SYSTEM 4.4. Vegetation results Polycorders were transferred to a laboratory computer. A portaThe estimate of the spectral BRF of the walnut orchard, Table I, bl computer dedicated to storing the radiance data as they are appears typical of the spectral BRF of other healthy, green acquired by the Polycorders is one possible solution to this vegetation canopies.T57'8 The estimated BRF is relatively small problem. in the chlorophyll absorption region (the blue and red regions), We recommend altering the method by which the view direcrelatively large in the green spectral region, larger still in the tion is selected so that the radiometers might be pointed in all middle infrared (bands 6 and 7), and largest in the near infrared view directions, not just those in the principal plane. This could mbands 4 and 5). be accomplished by replacing the sundial with a sun dish, a The large variability in the canopy architecture, due to the portion of a sphere centered at the same point as the angle ronouncvd, wideliy spaced rows of foliage and soil, manifests indicator plate, that is, at the intersection of the axes of the drive roignificant variability spacin the BRF but a comparatively small shaft and the control lever bearing. Like the sundial, the sun dish ignficnt of variabiity in the BRF of but a comparatively small e would be oriented in the solar azimuth direction during opera~sults affirm the validity of the process for estimating the BRF tion. Various radiometer view directions would be identified oefficient of variation of the BREXy of the canopy. These tion. Various radiometer view directions would be identified sults affirm the validityverage of the BRF proessvided by for estimating the BRF with the aid of holes in the sun dish into which the pawl would ~ the canopy-as an average of the BRFxy provided by four be extended axially from the control lever handle. )propriately placed radiometers. There is evidence for track- b e extended axially from that those seeking to compare the:pendent bias in the estimates of the canopy BRF. But the Finally, we recommend that those seeking to compare the mparatively small valuendent bias in th e of estimates ofbias, less than 3.5% ofBut the results reported here with results obtained from satellite sensor mparatively smalled value, affis the validity ofless the sampling strategy the data or using models for the canopy reflectance take into account ncasured value, affirmsce the ralidio meters on the horizontal both the field of view, 150, and the altitude of the measuring nctilerning be am. And because this bias appeaters to bthe horizontal radiometers, a bit more than twice the canopy height. First, tilever beam. And because this bias appears to be predict- rather than integrating within an angular FOV, models and e, it may be removed from the estimated BRF a posteriori. satellite sensors typically represent the canopy reflectance for a results provide the a priori information needed for develop- collimated view. Second, in the results reported here, there is a an unbiased sampling strategy for future data collection interaction between the noncoimated FOV and the altitude an unbiased sampling strategy for future data collection collimated view. Second, in the results reported here, there is an interaction between the noncollimated FOV and the altitude,ns (nvolvmcg t ae walnut orcfara r total of 20 measurements (lvingach an average of the datawalnut orchard. of such that the light received from a scatterer is differentially. total of 20 measurements (each an average of the data of weighted according to its height. Because of these two effects, r radiometers) were acquired in 3 min to estimate the canopy that is, a noncollimated FOV and the interaction between the' and detect true differences of 12% of the mean approxi- that is, a noncollimated FOV and the interaction between the FOV and the sensor altitude, each radiometric measurement will =ly 90% of the time. be unlike those obtained from a satellite sensor, a fact that must be taken into account in the modeling process if valid compariECOMMENDATIONS sons are to be made.:ave three recommendations concerning improvements that d be made in the design of the tree canopy radiance 6. SUMMARY urement system and one recommendation concerning the urement syfstem and Bone recommendation concerning the The tree canopy radiance measurement system is a practical )arison of the estimated BRF shown in Table I with the BRF approach to estimating the BRF of a tree canopy displaying large:ted by canopy reflectance models, heterogeneity as a function of horizontal location. The fact that ~ recommend increasing the strength of the drive shaft to recommend increasring the strength of the drive shaft. to within a short time period high quality spectral data were ac4 g safety margin. The drive shaft, a critical part of the quired over a tree canopy approximately 5 m tall displaying a n, does not meet the design criterion for withstanding a 4 gution provides validating evidence.. force because ofiseatvlheterogenous foliage distribution provides validating evidence ic force because of its relatively small outer diameter (7.6 of the utility of the design and of the use of four radiometers to id the fact that it was machined down to slip fit into the average out much of the spatial variability. For the walnut block bearings. When loaded, the shaft has a stress block bearings. When loaded, the shaft has a stress canopy, a total of 20 measurements (each measurement is deitration at a circumscribed groove located forward of the fined as the average of the data from the four radiometers) was block 30 cm behind the shaft/beam flange connection. acquired in 3 min to estimate the canopy BRF and detect true hallow undercut groove is necessary for machining pur- differences of 12% of the mean approximately 90% of the time.:o ensure the correct shaft diameter at the pillow block In these results only the nadir view direction was used. The,.) Consequently, the safety factor for the current drive system has already been used to measure an additional tree only 2.4. To increase the strength of the drive shaft, we canopy, viewed from 11 directions, and will be employed in the iend increasing its diameter. For example, increasing the future to obtain better understanding of the BRF of tree iameter to 9.5 cm, while maintaining the same wall ss, increases the safety factor to 4.1, with a weight:of only 4 kg. In addition, the increase in shaft diameter:s the amount of deflection at the beam's free end by 7. ACKNOWLEDGMENTS highly desirable design objective for measurement ac- This work was performed for the Jet Propulsion Laboratory, California Institute of Technology, and was supported by the zcommend increasing the memory capacity of the data National Aeronautics and Space Administration program on Eos on system of the tree canopy radiance measurement Synergism. o permnnit system operation for an extended period of We thank Barry D. Ganapol of the University of Arizona, eferably for all day without interruption. The system Tucson, for his good humored assistance in the design and cached memory capacity after acquiring approximately construction of the gantry, a task performed in the summer heat a points during a period of 0.5 h and required frequent of Fresno, Calif., for which success, just as Thomas A. Edison during which the contents of the memories of the said, was more than 98% perspiration. OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 11 / 1235

CALDWELL VANDERBILT We thank the manager and staff of the Kearney Agricultural 13. S. A. W. Gerst and C. Simmer, "Radiation physics and modeling 1 Center of the University of California at Davis for access to the off-nadir satellite sensing of non-Lambenian surfaces," Remote Set Center of the University of California at Davis for access to the Environ. 20, 1-29 (1986). machine shop and facilities of the center and Dave Goldhamer of 14. E. J. Milton, "Principles of field spectroscopy," Int. J. Remote Sens. the University of California at Davis for use of the walnut 1807-1827 (1988). orchard for data collection. 15. B. F. Robinson, M. E. Bauer, D. P. DeWitt, L. F. Silva, and V. Vanderbilt, "Multiband radiometer for field research," in Measurements We thank Harry Brining of Grove Manufacturing, Shady Optical Radiations,Proc. SPIE 196, 8-15 (1979). Grove, Pa., for advice and information about the capabilities of 16. F. E. Nicodemus, J. C. Richmond, J. J. Hsia, I. W. Ginsberg, and Limperis, "Geometrical considerations and nomenclature for reflectance manlifts. NBS monograph 160, Nat. Bur. of Stand., Washington, D.C. (1977). Finally, we thank the team that made the effort possible: 17. P. J. Sellers, "Canopy reflectance, photosynthesis, and transpiration," L JDobson, Jack Paris, Susan Ustin, Guy. Remote Sens. 6, 1335-1372 (1985). JoBea Cimino, Craig Dobson, Jack Paris, Susan Usin, Guy 18. P.. Sellers, "Canopy reflectance, photosynthesis, and transpiration n. T Cook, Jenny Clark, Eric Kasischke, and Kevin Berger. role of biophysics in the linearity of their interdependence," Remote Set Environ. 21, 143-183 (1987). 19. A. G. Journel and C. J. Huijbregts, Mining Geostatistics, Academic Prer 8. REFERENCES New York (1978). 20. P. J. Curran, "The semivariogram in remote sensing: an introductior 1. C. S. T. Daughtry, V. C. Vanderbilt, and V. J. Pollara, "Variability of Remote Sens Environ. 24ri493-507 (1988). reflectance measurements with sensor altitude and canopy type," Agron. J. 21. D. C. Howell, Statistical Mcthodsfor Psychology, Duxbury Press, Bosto 74, 744-751 (1982). 2. J. A. Smith, "Mantter-energy interaction in the optical region," in Manual ofass. (1982). Remote Sensing, Second Edition, R. N. Colwell, editor-in-chief, pp. 61-113, Am. Soc. of Photogram., Falls Church,Va. (1983). 3. "Earth system science, a program for global change," NASA, Washington, D.C. (May 1986). D.C. (Ntay 1986). __ William F. Caldwell received the BS degrn 4. "Earth system science, a closer view," Rept. of Earth System Sciences in mechanical engineering in 1988 from tt Committee, NASA Advisory Council, NASA, Washington, D.C. (Jan. University of California at Davis. He is cu 1988). rently a mechanical design engineer fi S. K. T. Kriebel, "Measured spectral bidirectional reflection properties of four NASA Ames Research Center and is workir vegetated surfaces," Appl. Opt. 17(2), 253-259 (1978). toward the MS degree in mechanical ens 6. N. S. Goel and D. W. Deering, "Evaluation of a canopy reflectance mode ner t Staf Uiesal A for LAI estimation through its inversion," IEEE Trans. Geosci. RemotePal Al Sens. GE-23, 674684 (1985). Ca if. 7. K. J. Ranson, C. S. T. Daughtry, L. L. Biehi, and M. E. Bauer, "Sun-view angle effects on reflectance factors of corn canopies," Remote Seas. En- i =;4 viron. 18, 147-161 (1985). 8. K. i. Ranson, L. L. Biehl, and M. E. Bauer, "Variation in spectral response of soybeans with respect to illumination, view, and canopy geometry," Int. Ven C. Vanderbilt is a senior research sciei J. Remote Sewi. 6, 1827-1842 (1985).'tist at the Ames Operations of TGS Technc 9. K. J. Ranson, C. S. T. Daughtry, and L. L. Biehl, "Sun angle, view angle tist at the Amesearch interations in the light-sca and background effects on response of simulated balsam fir canopies," terinogy, with research interistics of plant canopies an Photogram. Eng. Remote Sens. 52, 649-658 (1986). tering characteristics of plant canopies an 10. D. S. Kimes, W. W. Newcomb, R. F. Nelson, and 1. B. Schutt, "Direc- leaves and also in the development of ti tional scattering properties of a wintering deciduous hardwood canopy," sensors associated with measuring suc IEEE Trans. Geosci. Remote Sens. GE-24, 281-293 (1986). characteristics. Dr. Vanderbilt earned his B! 11. D. S. Kimes and W. W. Newcomb, "Directional scattering properties of a MS,and Ph.D. degrees in electrical enginee wintering deciduous hardwood canopy," IEEE Trans. Geosci. Remote ing from Purdue University, where he serve Seas. GE-25, 510-515 (1987). 12. S. A. W. Gerstl and A. Zardecki, "Coupled atmosphere/canopy model for first ctoral fellow and then as remote sensing of plant reflectance features," Appl. Opt. 24, 94-103 research engineer in the Laboratory for A; (1985). plications of Remote Sensing. 1236 / OPTICAL ENGINEERING / November 1989 / Vol. 28 No. 1 1

1749 WATER RELATIONS OF A WALNUT ORCHARD: SIMULTANEOUS MEASUREMENT WITH REMOTE SENSING J A Weber S L Ustin Biological Station Department of Botany The University of Michigan University of California Ann Arbor, Ml 48109-1048, U.S.A Davis, CA 95616, U.S.A ABSTRACT continuum is affected by a variety of factors, including: Water relations of trees vary over time scales of hours soil water potential, conductances between the varinous to weeks. The effect of changes in water status on compartments, and storage. The primary driving force microwave backscatter measurements from tree for movement of water is evaporation from the leaves canopies was determined. Of the several parameters (transpiration). Water status may be measured in measured, leaf xylem water potential showed the several ways. Water content is the easiest parameter to measured, leaf xylem water potential showed the greatest diurnal variation and appeared to co-vary in measure; however, it is the least reliable. Leaves are time with C-, L-, and X-band microwave measurements. especially variable in both amount of water and dry In addition data collected with L-band dielectric probes mass. Water content can vary at full hydration from less embedded in the trunk of selected trees showed than 70% to nearly 90% depending on species (Ref. 6). Within a species or treatment, a large number of parallel temporal variations. It is unlikely that water potential, per se, is being sensed; however, water samples are needed in order to reduce inherent potential of the leaves is a sensitive measure of overall variability. Relative water content (i.e., in situ water plant water status and is a useful parameter for mass per water mass at full hydration) reduces the connecting remotely sensed data to plant activity. apparently stochastic variability. However, comparison among different species is questionable. Keywords: water potential, water content, relative water Keywords: water potential, water content, relative water The concept of water potential (Ref. 7) is derived from steady-state thermodynamics using the chemical potential of pure water as a reference. Water potential 1. INTRODUCTION is the sum of turgor pressure (the mechanical pressure of the cell sap on the cell wall), osmotic pressure (effect of solutes), and matrix pressure (effect of various Study of water relations of encompasses not only water of solutes, and matrx pressure (effect of vaious status but also the dynamic processes of transpiration more consistent measure of plant water status and and water flux through the plant. Water status (any more consistent measure of plan water status and measure of the general state of water, (Ref. 7)) of a connection to plant function than water content. As plant is a primary determinant of growth and transpiration increases, flw rate to the leaves plant is a primary determinant of growth and increases and the amount of stored water decreases development (Ref. 4). The components of water status (Ref. 9). Leaf water potential responds by decreasing. (e.g., water potential, water content, and transpiration) can vary over periods of hours to weeks. Because of its 2 MATERIALS AND METHODS significance to plant functioning, estimation of plant water relations parameters using remotely sensed data Water status was measured on walnut trees growng in would be useful. Several studies have suggested the an experimental orchard (Kearney Agricuttural Center utility of optical data to determine water relations Fresno Co., California, USA) irrigated at 100% or 33% properties (Refs. 5, 8, 10). Since microwaves are of potential evapotranspiration. Several parameters sensitive to the water content of objects (Ref. 11), they were measured to provide a broad picture of water may be useful for detection of short- to long-term changes in water status, while for other applications of status for comparison wth concurrent acquired microwaves, e.g., estimates of biomass, such changes microwave and optical data. will add to the noise in the signals. Fresh and dry weight, water and osmotic potential ot leaves were measured periodically throughout the 1.2 Water relations experimentalperiod. Single leaflets were placed in plastic bags, excess air removed and sealed Samples Water movement through the soil-plant-atmosphere were stored on ice for fresh weight determinations Dry Prr,,,Hinoc nf~T IC;ARC'gR e id. 1,i,...l,,,- 1" of.,. o. n'-e.'-''.'.o -..

1750 J A WEBER &AL weight was measured after drying at 70 C for 36-48 hr. altered water transport processes. Coupling of water Water potential of leaflets was measured with a flow through the soil-plant-atmosphere continuum pressure chamber. Relative water content was makes measurement of leaf water potential a sensitive determined by hydrating leaves at 0-4 C. Osmotic indicator of plant water status for trees. potential was measured using pressure-volume curves or frozen samples for thermocouple psychrometry (6). The most striking observation is the parallel diurnal Transpiration rate and leaf conductance was measured variation of dielectric constant and water potential. during the day in conjunction with leaf water potential Both parameters declined from dawn until early measurements. Fresh and dry weights of fruit (husks afternoon then recovered overnight. The spatial and nuts) were measured over diurnal periods. A few variation in dielectric constant (Ref. 2) indicates a measurements of stem water content were taken. complex process, but more information needs to be gathered to characterize the response of dielectric Optical and microwave measurement are presented in probes to changes in water status in tree trunks. (Refs. 1, 3). L-band dielectric probes were used to measure the dielectric constant at various positions and C-, X-, and L-bands showed diurnal changes in depths in trunks (Ref. 2). backscatter features that paralleled changes in water potential (Ref. 3). While it is highly unlikely that 3. RESULTS microwave backscatter is directly sensing water potential, the fact that backscatter and water potential Water content as a function of dry or fresh weight, or co-vary provides a potential opportunity to monitor leaf area varied so much from leaf to leaf that no diurnal indirectly water potentials of canopies. However, this pattern was observed. On average water content per correlation may also introduce noise into remotely unit dry weight was lower for sun leaves than shade sensed microwave data unless analyses include water leaves. Relative water content of leaves was between status models. 0.8 and 1.0 with little diurnal change. Over a two week period fruit water content declined slightly as they 5. REFERENCES mature. 1. Caldwell, W. and V. C. Vanderbilt, 1988, "Field Unlike the other measures of water status, water system to determine reflectance of tree canopies," potential showed very strong diurnal variation with to appear in IGARSS'88 Digest, Edinburgh, maximum water potential of about -2 bars occurring Scotland, September 13-16, 1988. before sunrise and the minimum of -15 to -20 bars in the early afternoon, when transpirational water loss is 2. Dobson, M. C., 1988, "Diurnal and seasonal maximum. Osmotic pressure did not vary significantly variations in the microwave dielectric constant between early morning and afternoon. Shaded leaves of selected trees," to appear in IGARSS'88 Digest, in the 100% irrigation treatment generally had higher Edinburgh, Scotland, September 13-16, 1988. water potentials than sun leaves, reflecting difference in transpiration. However, shade and sun leaves in the 3. Dobson, M. C., K. McDonald, F. T. Ulaby, and 33% treatment had essentially the same water J. F. Paris, 1988, wDiurnal patterns in multipotentials, probably reflecting both more open canopy frequency, multipolarization backscattering by a and lower water availability. The minimum turgor, walnut orchard," to appear in IGARSS'88 Digest, calculated as the difference between osmotic potential Edinburgh, Scotland, September 13-16, 1988. and water potential, occurs in the earty afternoon, with minimum values for both irrigation treatment of 0-5 4. Bradford, K.J. and T.C. Hsiao, 1982, "Physiological bars. Responses to Moderate Water Stress," In Lange, O.R., P.S. Nobe, C.B. Osmond, and Diurnal changes in backscatter of microwaves H. Ziegler (Eds.) Ency. Plant Phys. NS 12B. paralleled changes in water potential (Ref. 3). Data Physiological Plant Ecology II, Springer-Verlag, from the dielectric probe (Ref. 2) show that the dielectric pp. 264-324. constant of the trunk (and presumably the water content) declines markedly during the day and 5. Hunt, E.R., Jr., B. N. Rock, and P. S. Nobel, 1987, rebounds at night. "Measurement of leaf relative water content by infrared reflectance," Remote Sens. Environ., 22: 429-435. 4. DISCUSSION 6. Kramer, Paul J., 1983, Water Relations of The basic diurnal pattern in leaf water potential is Plants, Academic Press, Inc., New York. typical of plants growing in a region of high evaporative demand (i.e., high temperature and low humidity). 7. Passioura, J. B., 1982, "Water in the soil-plantHowever, at no time did any of the trees on either 33% atmosphere continuum." In: O. L. Lange et al and 100%o potential evapotranspiration treatments have Eds., Physiological Plant Ecology II. Encyclo water potentials less than -20 bars, although a few Plant Physiol. NS, Vol. 12 B. pp 5-33. leaves reached this level and had wilted. These data indicate that irrigation treatment effects were manifested by altered canopy development rather than

WATER RELATIONS OF A WALNUT ORCHARD 1751 8. Ripple, W. J. and B. J. Schrumpt, 1987, "Remote sensing of plant water status," Proc. Int. Conf. Measurement of Soil and Plant Water Status, Logan, UT. July-6-10, 1987, V2: 103-109. 9. Schulze, E-D., J. Cermak, R. Matyssek, M. Penka, R. Zimmermann, F. Vasicek, W. Gries, and J. Kucera, 1985, "Canopy transpiration and water fluxes in the xylem of the trunk of Larix and Picea trees -- A comparison of xylem flow, porometer and cuvette measurements," Oecologia, 66: 475-483. 10. Sellers, P. J., 1985, "Canopy reflectance, photosynthesis and transpiration," Int. J. Remote Sens, 6:1335-1372. 11. Ulaby, F. T., R. K. Moore, and A. K. Fung, Microwave Remote Sensing: Active and Passive, Vol. 1. -- Microwave Remote Sensing Fundamentals and Radiometery, Addison-Wesley, Reading, MA, 1981, 456 pages.

MEASUREMENT OF TREE CANOPY ARCHITECTURE BY S.N. Martens and S.L. Ustin, Department of Botany, University of California, Davis, CA 95616, USA and J.M. Norman, Department of Soil Science, University of Wisconsin, Madison, WI 53706, USA SUBMITTED TO: INTERNATIONAL JOURNAL OF REMOTE SENSING ABSTRACT. The lack of accurate, extensive geometric data on tree canopies has retarded development and validation of radiative transfer models. We devised a stratified sampling method to measure the three-dimensional geometry of 16 walnut trees which received irrigation treatments of either 100% or 33% of evapotranspirational demand for the previous two years. Graphic reconstructions of the 3D geometry were verified by 58 independent measurements. The distributions of stem and leaf size classes, lengths, and angle classes were determined and used to calculate leaf area index (LAI), stem area, and biomass. Reduced irrigation trees have lower biomass of stems, leaves and fruit, lower LAI, steeper leaf angles and altered biomass allocation to large stems. These data can be used in ecological models that link canopy processes with remotely sensed measurements. Address correspondence to: Scott N. Martens, Department of Botany, University of California, Davis, CA 95616; Phone (916) 752-2956 1

1. Introduction Radiative transfer models have wide application for meteorological and ecological studies of canopy energy exchange processes and for studies of plant productivity. Improved knowledge of the geometric structure of plant canopies is essential to the further development and validation of radiative transfer models of canopy reflectance. The lack of accurate, extensive geometric data has retarded development and testing of canopy reflectance models (Vanderbilt 1985). Existing models for canopy reflectance use minimal geometric input perhaps in part because of the difficulty of obtaining necessary data (Kimes and Kirchner 1982; Norman and Welles 1983; Goel and Grier 1988). The fundamental description of canopy geometry includes the inclination, azimuth, surface area, and location of individual plant parts. Data gathering requirements can be eased by applying some reasonable assumptions. Azimuthal angle distributions of stems and leaves can be assumed to be symmetric, which is usually the case for most nonheliotropic plant species. Further, the location of individual canopy elements can be assumed to be random, usually in a horizontal monolayer. These are reasonable assumptions for many herbaceous canopies of nearly full 2

cover (Norman 1979) so that the remaining parameter, inclination angle, is all that need be documented. Further refinements are needed for canopies having widely spaced plants with non-homogeneous horizontal or vertical dispersion patterns. Agricultural examples include orchard trees with dense crowns or some widely spaced row crops such as grapes. In natural communities, individual plants may be aggregated due to microsite differences or to competitive interactions. For these complex cases, higher dimension shapes, such as ellipsoids (Norman and Welles 1983), have been used to decompose canopies into subunits. Ellipsoids may represent individual plants within the canopy or may depict a cylinder of biomass as might occur along a row of closely spaced plants. Canopy elements are usually assumed to be randomly located within ellipsoids though aggregation can be introduced if more information on the geometric structure of individual plants is available. The location of canopy elements is obviously non-random in canopies of some large plants, e.g., trees and some shrubs, even though inclination and azimuth angles may be random. Smaller stems are completely dependent upon larger stems with respect to location. Because branching angle can be genetically fixed in a species the inclination and azimuth angles of smaller stems are also dependent on those of larger stems to some degree. Leaves, in turn, are usually 3

attached only to the smaller stems and are dependent on them for location. However, because of phyllotaxy or twisting of petioles among other reasons, leaf angular distributions can be relatively independent of the stem angular distributions (Fisher 1986). Because of these location dependencies statistical sampling of canopy geometry for extrapolation to the stand or field level can be difficult. 1.B. Geometric assumptions of canopy radiation transfer models A considerable literature on canopy radiation transfer models exists and several reviews are available (e.g., Ross 1981; Goel 1988). An early use of a one-dimensional canopy radiation model was made by de Wit (1965) to estimate canopy photosynthesis. His model, based on work by Monsi and Saeki (1953), assumed a canopy of randomly positioned foliage elements. Several studies have shown that canopy photosynthesis or carbon gain can be estimated from knowledge of the photosynthetically active radiation (PAR) intercepted by a canopy (Monteith 1981; Jarvis and Leverenz 1983; Linder 1986; Russell et al. 1989). Much literature now exists on this type of one-dimensional model for predicting light penetration in vegetation and for its use in estimating canopy photosynthesis (Lemeur and Blad 1974; Norman 1975; Ross 1981; Sellers 1985, 1987).

The simplest one-dimensional models of radiative exchange in vegetation canopies, e.g. Verhoef (1984), assume that leaves are randomly located Lambertian scatterers, symmetrically distributed about the azimuth and inclined to the horizontal with some reasonable distribution. The essential information required for input to such models is leaf spectral properties, leaf area index (LAI), and leaf inclination distribution (LID). The LAI can be measured by a number of direct and indirect techniques (Norman and Campbell 1988). The LID can be characterized by a twoparameter beta distribution (Goel and Strebel 1984) or a single parameter ellipsoidal distribution (Campbell 1986). Suits (1972) one-dimensional model, which assumed random leaf positioning, a symmetric azimuthal distribution and leaves inclined in only vertical or horizontal directions, was the first to predict canopy bidirectional reflectance. Verhoef (1984) generalized Suits (1972) approach to include any leaf inclination distribution with azimuthal symmetry. Several newer models have been proposed that are based on Lambertian leaf spectral properties, random leaf positioning, azimuthal symmetry, and any leaf inclination angle distributions (e.g. Cooper et al. 1982; Camillo 1987; Choudhury 1987). Hapke (1984) published a similar model for soil surfaces. Norman et al. (1985) describe a numerical model of canopy bidirectional reflectance similar in assumptions but which specifically considers the soil

bidirectional reflectance and non-Lambertian leaf characteristics. A slight variation on the random-leaf-positioning concept was proposed by Nilson (1971) to accommodate clumping or regular-leaf-spacing tendencies with empirical coefficients. However, obtaining the necessary empirical coefficients depends more on fitting the modified quasi-random model to radiation penetration measurements than on appropriate canopy structural measurements. A different concept of clumping was proposed by Norman and Jarvis (1975) for spruce trees, and the radiation penetration model was linked to measurements of projected needle area of shoots and the organization of shoots onto branches, with the model including the woody portions of the canopy. This might be referred to as a weighted-random approach. Other clumping models have been proposed for predicting light penetration in agricultural crops and some are reviewed in Norman (1974). Models that incorporate clumping require some measurements of the horizontal distribution of foliage and these are difficult to obtain. The three-dimensional radiation penetration model of Norman and Welles (1983), which is similar to the models of Whitfield and Conners (1981) and Welles (1976), has attempted to simplify the measurement input requirements by requiring only the dimensions of an 6

ellipsoidal envelope within which all the leaves are contained and assumed randomly distributed. Thus, for these envelope models, the only additional measurements beyond those required for a one-dimensional model are the overall crown dimensions of individual trees. The light penetration model of Norman and Welles (1983) has been recently extended to include the canopy bidirectional reflectance distribution function (BRDF) by Welles (1988). Incorporating the three-dimensional structure of the canopy into radiation models permits closer approximation of the complexity of real canopies. The model of Kimes and Kirchner (1982), which considers the general threedimensional distribution of elements requires even more detailed spatial data on the distribution of canopy elements than envelope models. This is also true for the model of Wang (1988). The most general three-dimensional canopy radiative transfer models are based on ray tracing (Myneni et al. 1987) and Monte Carlo techniques (Ross and Marshak 1985). An additional general method termed radiosity has been used in graphics applications (Greenberg 1986) but no published literature is yet available for its application in canopy studies. In principle, these radiative transfer approaches can accommodate almost any architecture. The Monte Carlo model of Ross and Marshak (1985, 1988) is a good example of 7

a model capable of predicting the canopy BRDF, including the "hot spot", for generalized canopy architectural descriptions even though they considered a random canopy. The latter model inputs considerable canopy structural information, including the number of leaves per canopy and the distance between leaves -- information which is not generally available. The greatest obstacle to the application of threedimensional radiative transfer models is that suitable canopy architectural information is scarce because of the difficulty in obtaining such measurements. Although recent advances in L-systems (Lindenmayer 1987) and fractal geometry (Mandelbrot 1982) can provide quantitative tools for representing some vegetation structures, detailed structural characteristics such as branching patterns, leaf size, leaf shape, and the distribution of flowers and fruits must still be measured directly. The study described in this paper provides the essential information required to fully characterize a threedimensional canopy with L-systems or fractals, but that task will be the subject of another paper. Here we describe the methodology for obtaining the essential information and summarize the results of the fundamental distribution of canopy elements from row-spaced trees with complex crown architecture. The canopy geometry of walnut trees which had

received irrigation treatments of 100% and 33% of calculated evapotranspiration for two years was measured. The angular and spatial distributions of stems were sampled in such a manner as to readily allow a verifiable three-dimensional reconstruction of the canopy stem components. Combined with other data (e.g., leaf angular distributions, tallies of branches, leaves, and fruits, wood and leaf specific weights) we can provide estimates of important canopy architectural characteristics such as leaf area index and biomass. 2. Methods 2.A. Description of Study Site The walnut orchard is at the University of California Kearney Agricultural Center, Parlier, CA, USA (36.60~ N 119.50~ W). The trees (Juglans regia cv.'Chico') were 6 years old at the time of sampling in August 1987. Tree spacing is nominally 6.7m across and 3.35m along the north - south oriented rows. Average maximum height of the crown outline was 4.8 m (n=24). Treatment blocks of 8 trees (2 rows by 4 trees) are surrounded by one row of border trees which receive the same irrigation treatment. Irrigation treatments of 33%, 66% (not sampled in our study) and 100% of calculated evapotranspirational use have been applied since 1986 (Goldhamer et al. 1988). Pruning is of the 9

"hedgerow" type where alternate row-facing sides of the trees are pruned each year. There is also pruning of some excessively heavy upper canopy branches as well as branches in the lower (< lm) canopy. Walnut trees have a longshoot/short-shoot morphology. The short shoots ("spurs"), which are usually less than 1 cm diameter, frequently occur at more or less uniform intervals along the long-shoots. Leaves and fruits occur at the apices of these short-shoots or at dominant terminal apices on larger shoots. 2.B. Sampling Rationale For the purpose of relating the geometry data set to spectral data we need to make inferences about the population of leaves and stems and their angular and spatial distribution in the orchard. The obvious natural sampling unit, the tree, is not appropriate because sampling a tree would undersample the least frequent components (e.g. large stems, which are few per tree, but many per orchard) and oversample the most abundant canopy components (i.e. small stems and leaves). By regarding the orchard as several populations of canopy components (i.e. stems in 5 diameter size classes, leaves, fruits), we could sample each with the appropriate sampling effort. However, the spatial and angular dependencies among canopy components complicates this simple stratified population scheme. Sampling to reveal the dependencies (e.g. location, branching angles) 10

among the populations, however, must be done on a natural unit, the tree. Therefore, we applied the stratified sampling scheme at the scale of orchard, tree, and branch levels to yield a data set which can be synthesized to provide angular distributions and spatial locations of the canopy components at the orchard level. 2.C. Sampling Methods 2.C.1. Stems Stem segments were divided into 5 size classes based on diameter at the midpoint along the length: Class 1, <= 1.0 cm; Class 2, 1.1 to 2.0 cm; Class 3, 2.1 to 3.0 cm; Class 4, 3.1 to 4.0 cm; Class 5, > 4.0 cm. Stems were sampled as segments. A segment (the definition of which differs slightly depending on the sampling level) was a section of stem, uniform along its length in azimuth and elevation angle, which is terminated by either 1) a break in zenith angle, 2) a break in azimuth angle, 3) an apex, or 4) branching into a stem of diameter equal to or less than the critical size class. The critical size class stem depended upon the stratified sampling level as described below. 11

For each segment we measured length, diameter, zenith angle (from vertical), and azimuth. Figure 1 presents a diagram of the stem and leaf angles measured. Lengths were measured to the nearest cm. Zenith angle (eb, Figure 1A) was measured by placing a draftsmen's protractor with a plumbbob along the segment. Zenith angle is the angle between a vertical line and the line the branch segment forms when its' base intersects the vertical line. A branch pointing straight up would have a zenith of 0~; horizontally, 90~; straight down, 180~. Zenith angles were translated to elevation angles (0~ = horizontal; 90~ = vertical) for use in the beta distribution calculations described below. Azimuth angles (ea, Figure 1A) were measured with a magnetic compass and later corrected to true north, or estimated with the protractor (using the true north - south row direction for orientation) where accurate magnetic readings were precluded. All angles were recorded to the nearest 5~. Data for each segment were stored in a doubly-linked list, i.e. the segment number from which the current segment came was recorded as well as the numbers of the segments into which the current segment branched. A doubly-linked list was used, even though a singly-linked list would suffice, to safeguard against errors which would break the linkage. This data structure allowed us to examine the relations among the segments and permitted the three-dimensional reconstruction of the data set. 12

2.C.l.a. Orchard-level Sampling Eight trees, in a 2 tree by 4 tree block, were sampled for each irrigation treatment (33% ET and 100% ET). For the orchard-level sampling, the critical size class was Class 4, so that only segments greater than 4 cm diameter (229 in all) were measured on all sixteen trees. All segments greater than 4 cm diameter are referred to collectively as Class 5 segments. All segments of size Classes 1 through 4 which branched directly from the Class 5 segments were tallied. There were no leaves or fruits directly attached to any of the Class 5 segments. 2.C.l.b. Tree-level Sampling Two trees in each of the two irrigation treatments were sampled. Each Class 3 or Class 4 segment tallied in the orchard-level sampling was measured down to and including Class 2 segments. Here, the critical size class was Class 1, so no Class 1 segments were measured on these Class 3 or Class 4 segments except when 1) a Class 1 segment terminated a long-shoot or 2) was greater than 25 cm in length (a "sucker" shoot). A total of 553 segments were measured. The number of Class 1 segments occurring on each segment measured were tallied. The number of leaves and fruits on the Class 1 segments measured were tallied. Notice that 13

this sampling scheme does not include Class 2 segments which branch directly from Class 5 segments. A relatively small number of these branches exist. Most of these Class 2 segments were located in the interior of the canopy and, for the purposes of calculations described below, are estimated to be the equivalent of three Class 1 branches. 2. C. 1. c. Branch-level Sampling Five branches, each on a different tree, were sampled down to and including all Class 1 segments. The diameters of the basal segment of these branches were 1.9, 1.9, 2.0, 2.0, and 3.4 cm. They included upper, lower, east- and west-side canopy positions. All leaves and fruits occurring on Class 1 segments were tallied. Of the 256 segments sampled, 126 were Class 1 segments which contained 378 leaves and 209 fruits. These data are used here primarily for establishing leaf and fruit numbers per Class 1 branch for extrapolations to the whole tree level. 2.C.2. Leaves The odd-pinnate compound leaves show a bilateral symmetry about a plane along the rachis such that the elevation angle of the lateral leaflets from this plane is about the same on both sides. The leaflets also show a similar symmetry of laminar folding about the midrib of the leaflet (Figure 1C) 14

We measured 147 leaves on 20 vertical transects of 8 trees in the 100% ET treatment and 72 leaves on 10 vertical transects of 6 trees in the 33% ET treatment. The transect locations were randomly selected below the canopy and the leaf nearest the vertical transect at 0.5 m intervals from the ground was measured. Measurements were made to fully describe the leaf in three-dimensions. For each leaf we measured the height, number of leaflets, the length of the petiole plus rachis, and the zenith (er) and azimuth (ea) of the rachis (Figure 1A). The terminal and lateral leaflets were obviously different in several respects so we measured the terminal leaflet and the adjacent left-lateral leaflet for the following: midrib zenith (e1) and azimuth angles, the azimuth of the normal to the leaflet lamina (or the zenith angle of the normal if necessary), and the width at the widest part of the leaflet lamina when naturally folded and when flattened with the ruler. The last two measurements allowed calculation of the folding angle of the leaflet lamina about the leaflet midrib (em, Figure 1C). 2.C.3. Biomass measurement Stem volumes for 110 stem segments were determined by displacement of water and subsequently divided by the stem dry weight to determine the wood density for stems in size 15

classes 1 through 3. Stems in Classes 4 and 5 were assumed to have the same density as those in Class 3. The dry weight of 100 fruits in the 33% ET treatment and 80 fruits in the 100% ET treatment was measured. Both dry weight and leaf area (LICOR LI-3000 Leaf Area Meter) were measured on each of 50 freshly excised leaves to determine leaf specific weight. Leaf area was measured for each leaflet on 100 leaves from each irrigation treatment. These data were used in the calculation of the area-weighted average of leaf zenith angles. 2.C.4. Plumb line Measurements To verify the three-dimensional reconstruction based on the angle and length measurements for stems we hung plumb lines at identified points in each of the trees sampled for treelevel measurements. The measured vertical height and distance from the trunk of each of the 58 plumb lines were compared with calculated values from the reconstruction. 3. Results and Discussion 3.A. Orchard and Tree Reconstructions 1G

To visualize the degree of sampling at the orchard- and tree-levels, diagrammatic representations of the canopy reconstructions are presented in Figure 2. The two irrigation treatment blocks of 8 trees each are shown in realistic proximity to one another. The 4 trees sampled at tree-level are shown with the applicable level of detail, which includes all measured branches down to Class 2 size and some Class 1 branches. The other trees are represented by a reconstruction of all segments greater than 4 cm diameter (Class 5 branches). Figure 3 illustrates the treelevel sampling using data for tree R3T09 from the 100% ET treatment. The line segments are color coded with respect to diameter size class. The figures demonstrate the realistic reconstruction of this orchard and how the complex spatial structure is retained in the data set. Once verified, this data set will be useful for evaluating spatial distributions of canopy components and for validating model predictions of canopy properties based on remotely sensed optical and microwave data. The three-dimensional reconstructions are subject to cumulative positional errors. These errors could be a function of the number of preceding segments (greater chance for errors) or a function of the length of the preceding segments (incorrect projection of segments due to errors in angle measurements). To evaluate the fidelity of the three17

dimensional reconstructions we compared the calculated positions in space of 58 points on the 4 trees used in the tree-level sampling with independent measurements of location determined by plumb lines. There is close agreement between the two data sets for both vertical height (Figure 4) and horizontal distance from the trunk (Figure 5). The regression equation in each case is highly significant (p <.001). Intercepts are not significantly different from 0 and slopes are not significantly different from 1.0 (standard error of slope for each regression is 0.025). Examination of residuals shows no deviation from the linear model. Close inspection of the statistical patterns provides further support for the validity of the reconstructions. Correlations of residuals from regressions in Figures 2 and 3 to the number of preceding segments, or their summed length were significant only for distance residuals versus number of preceding segments (r = 0.474, p =.001). However, when the number of preceding segments was included as a term in the regression model for distance the R2 increased only slightly from 0.9773 to 0.9801. Thus, the accumulation of measurement errors, leading to large deviations from the expected measurements in height or distance, was not found to be significant in the data set. Hence, this data set can be used with some confidence for 18

parameters which depend on the three-dimensional location of canopy components. 3.B. Stem Size Classes The stem length distributions of stem diameter sizes is presented for the orchard-level (Figure 6) and tree-level (Figure 7) sampling. The orchard-level data is the summed length of stems of each diameter size class for the 8 trees in each treatment. Summed length of stems was used because the number of segments (stems) measured is an artifactual value due to the manner in which segments were defined and sampled. Smaller diameter stems are more frequent overall but there is a rise in frequency in the 13 to 16 cm size classes which corresponds to the sizes of most of the trunks on the trees. The largest stem diameters (17 cm and 18 cm classes) are exclusively among the 100% ET treatment trees while the largest stem diameters in the 33% ET treatments are 13 cm to 16 cm. Similarly, stem size categories from 5 cm to 10 cm have greater length in the 100% ET treatment than the 33% ET treatment (Figure 6). T-tests between the treatments for each size class category indicate a significant difference for only the 7 cm class (t = 2.2681, p <.04). However, the total length of stems > 4 cm diameter per tree is significantly less in the 33% ET treatment (t = 2.8353, p <.02). The difference could result from reduced allocation of photosynthate to the growth of large stems 19

during the two seasons of reduced irrigation. The observation that the largest diameter stems are in the 100% ET treatment is consistent with this idea. The 100% ET treatment also has a greater calculated biomass of Class 5 stems (x = 26.63 kg/tree) than the 33% ET treatment (x 20.44 kg/tree; t = 2.265, p <.04). The stem length distributions for the tree-level sampling (Figure 7) are expressed for the two irrigation treatments (two trees per treatment). The data for Class 1 and Class 2 stems includes extrapolated values for the tallied Class 1 and Class 2 stems which occurred on the Class 5 stems. Extrapolations were based on the following assumptions. A mean Class 1 branch length of 6.72 cm (calculated from the branch-level sampling) was assumed for the tallied Class 1 branches. A further assumption was made for the Class 2 segments tallied on Class 5 segments. These small branches are mostly in the interior of the canopy. Examination of branch-level data most similar to them indicates that the Class 2 branches are equivalent in length to 3 to 6 Class 1 segments. We assumed for our calculations that one Class 2 branch is equivalent to three Class 1 segments. An assumption at the upper extreme of 6 Class 1 segments per Class 2 would increase the summed length of Class 2 segments by 9.0% in the 100% ET treatment and 6.4% in the 33% ET treatment. The summed length of all stem size classes would increase 2.9% in the 100% ET treatment and 2.1% in the 33% 20

ET treatment. Thus, the worst case error indicates the assumptions provide reasonable values for further modelling. The results (Figure 7) for the two irrigation treatments are very similar for each size class category and similar in the summed length of all stem size classes (27.5 m and 27.0 m in the 100% and 33% ET treatments respectively). This pattern is consistent with the maintenance of the carbon allocation to smaller stems under water-stress and contrasts with the observed pattern of reduced allocation to larger stems. 3.C. Stem Angle Distributions The azimuthal distribution of stems of Classes 2 through 5 for both treatments is presented in Figure 8. The distribution of each size class is uniform with respect to azimuth. The smallest size class, Class 2, appears to have an increased frequency in the north to northeast quadrant (X2 = 12.68, p <.01). The relatively symmetric azimuthal distribution of stems we found, while expected for opengrown trees, might not be expected in this situation where the trees are grown with compact spacing along the rows (north - south) and receive hedgerow pruning on alternate sides (east - west) every year. The zenith angle of the 4 stem size classes for both treatments is shown in Figure 9. Stem zenith angles range 21

from 0~ (vertical) to 90~ (horizontal) to 1800 (stem pointing downward). Our definition of zenith angle for stems differs from the conventional because the angle is expressed in reference to the origin of the stem and therefore extends over a 180~ arc. There is a clear trend for larger stems to be more upright as shown by the rapid rise of the curve for the Class 5 ( > 4.0 cm diameter) branches in the 0~ to 900 region. In contrast, the Class 2 stems are nearly uniformly distributed throughout the 180~ range. 3.D. Leaf Angle Distributions The azimuthal angle distribution of lateral and terminal leaflet types by irrigation treatment is presented in Figure 10. All four leaflet types appear to be uniformly distributed with respect to azimuth. The zenith angles of the four leaflet types are shown in Figure 11. Zenith angles are expressed relative to the origin of the petiole and the top surface of the leaf. Leaf zenith angles may exceed 1800 in cases of extreme petiole twisting. Terminal leaflets are more steeply angled than lateral leaflets within a treatment. The mean angles for the 100% ET leaves are 151.4" and 131.0~, terminal and lateral leaflets respectively; 156.4" and 139.8~ for the 33% ET treatment. The leaflets of the 100% treatment are more horizontal than the same type of leaflet in the 33%

treatment as shown by the shift to lower zenith angles in Figure 11 for the 100% ET treatment leaflets. The area-weighted average leaflet zenith angle was calculated for each treatment. The 100% ET treatment leaves have 6.77 leaflets per leaf and the 33% ET treatment leaves have 6.49 leaflets per leaf. Average terminal leaflet area is 83.0 and 77.8 cm2, lateral leaflet area is 36.3 and 36.2 cm2, for 100% and 33% ET treatments. Therefore, the areaweighted average leaflet angle for the 100% ET treatment is 136.8~ from vertical and is 144.5~ for the 33% ET treatment. 3.E. Beta Distributions of Elevation Angles The angular distributions of the canopy components can also be expressed using the beta distribution (Goel and Strebel 1984). The two parameters of the beta distribution, mu and nu, can be calculated from the mean and variance of the leaf and stem elevation angle distributions, and together can be used to characterize distributions relative to ideal types (de Wit 1965; Ross 1981). Zenith angles were translated to the 0~ (horizontal) to 90~ (vertical) quadrant for these calculations and are referred to as elevation angles. All angles referred to concerning the beta distributions are relative to this expression of zenith angles. 23

The beta distribution parameters for the five size classes of branches are plotted in Figure 12 for each treatment. The smallest diameter branches in both treatments show a uniform distribution tending slightly towards planophilic (most angles horizontal). With increasing diameter the distribution shifts past uniform toward spherical or spherical - erectophilic (most angles vertical). Branches of Classes 3 and 4 in the 33% ET treatment tend to be more towards plagiotropic (most angles about 45~) than the corresponding 100% ET treatment branches. Class 5 branches of both treatments tend toward the erectophilic distribution but the 33% ET branches show a lower variance. Leaflet angle beta distributions are plotted in Figure 13. Terminal leaflets in both treatments have an approximately spherical distribution of elevation angles. The distribution of lateral leaflets differs between the treatments. The 33% ET lateral leaflets have a plagiotropic distribution. The 100% ET lateral leaflets have a distribution between uniform and plagiotropic due to a greater variance (x = 41.0~, s2 = 381.2) than the 33% ET lateral leaflets (x = 48.7~, s2 = 259.8). The tendency for the 33% ET leaflets to be more erect than the 100% ET leaflets is indicated by their position more towards the erectophile point than the uniform - plagiotropic line. 3.F. Leaf Geometry Changes with Height 24

Several parameters of leaf geometry were noted to vary with height in the canopy (Table 1). The leaflet folding angle describes the folding of the lamina about the midrib (Figure 1C). Leaflet folding angles decreased with height in the canopy for all four leaflet types indicating increased folding. The folding angle was more pronounced for terminal leaflets (146.8~ and 148.4~) than for lateral leaflets (156.30 and 160.1~, 100% ET and 33% ET treatments respectively) when averaged over all heights. Folding angles differed significantly between leaflet types (ttests, p <.002) within the same treatment but neither leaflet type differed significantly between treatments (ttests, p ~.24). The zenith angle of the leaflet midrib showed a significant increase with height for lateral leaflets in both treatments but only weakly for terminal leaflets in the 100% ET treatment. The zenith angle of the rachis, to which the leaflets are attached, is significantly negatively correlated with height only in the 33% ET treatment. The angle becomes more horizontal toward the top of the canopy. The width of the terminal leaflets is negatively correlated with height for both treatments. Lateral leaflets show a significant negative correlation with height only in the 25

100% ET treatment. The number of leaflets per leaf significantly increases with height in both treatments Despite the statistically significant correlations of some leaf geometry parameters with height it is important to note that the R2 values are generally low and the slopes are often small. 3.G. Tree Descriptors A number of descriptors for each of the four trees sampled in the tree-level sampling are presented in Table 2. We assumed that each tallied Class 2 branch, as we did for length (see 3B above), was equivalent to three Class 1 branches with respect to weight, area and numbers of fruits and leaves. If the assumption of 6 Class 1 branches per tallied Class 2 branch were used instead (see 3B above) the range of percent increase in the values would be as follows: branch biomass,.28% to.37%; branch area,.70% to 1.05%; leaf area and biomass, 4.81% to 6.68%; nut mass, 5.13% to 8.77%. The 100% ET trees (R2T09 and R3T09) show greater increases within the ranges given because they have more tallied Class 2 segments than the 33% ET treatment trees (R2T14 and R3T17). The same leaf specific weight (7.409 mg per cm2) was used for calculations for both irrigation treatments. Wood density was measured as 347.2, 388.6, 447.6, and 449.6 mg/cm3 for stems of Classes 1 through 4. 26

Class 5 stems were assumed to have the same wood density as Class 4 stems. Individual fruit weight was 11.10 g and 8.29 g for the 100% ET and 33% ET treatments respectively. Leaf area per leaf was measured at 297.04 cm2 in the 100% ET treatment and 285.43 cm2 for the 33% ET treatment. The average height of the 100% ET trees is greater than for the 33% ET trees but there is near overlap in values. However, the maximum height of the Class 5 branches shows significant separation of the two types with the 100% ET trees taller than the 33% ET trees. This is consistent with the results for Class 5 segments on all 16 trees measured at the orchard-level where the 100% ET trees have a maximum height of 411 cm which is significantly greater than the 335 cm height of the 33% ET trees (t = 5.245, p = 0.0001). The 33% ET trees may reach maximum heights as great as the 100% ET trees but have fewer large diameter branches in the upper canopy. The reduced irrigation treatment appears to have decreased radial growth of stems more than extension growth. The total biomass of the 100% ET trees is greater than the 33% ET trees which are about 86% of the biomass of the 100% ET trees. The apportionment of biomass among organs, however, seems to be about the same between the treatments except for fruits. Notably, the 33% ET trees are 89% of the stem weight and 94% of the leaf weight, but they have only 76% of the fruit weight of the 100% ET trees. 27

Stem areas were calculated as the summation of length times diameter for all stems. Stem areas and leaf areas are greater for the 100% ET trees. LAI (m2 leaf area per m2 ground area) averages 3.40 for the two 100% ET trees and 3.18 for the two 33% ET trees presented in Table 2. However, projected LAI exhibits a greater difference between the treatments due to the steeper leaflet angles of the 33% ET trees. The number of sucker shoots is greater on the 100% ET trees (7 and 9 shoots) compared to the 33% ET trees (0 and 4 shoots) which contributes to the dissimilar appearances of the trees in the field. The projected surface outline of the upper tree canopy is the primary visual difference between the treatments and leads to this perception. 4. Future Work The potential exists for deriving more detailed descriptions of some canopy components from this data set. In addition, the synthesis of the stem and leaf data into a threedimensional representation of the orchard canopy would provide an unparalleled data set for radiation transfer model development and validation. Because of the high level of detail supplied it may also aid in clarifying the degree of abstraction permissible in models before physical fidelity is lost. Specifically, the assumption of random

versus clumped distributions of canopy elements in subcanopy volumes, an area much in need of further research (Goel and Grier, 1988), can be addressed. Further, the data set could be used in models of canopy growth, development, or physiology (e.g. Caldwell et al. 1988; Myneni et al. 1989; Sellers 1985, 1987) and so can support the link between canopy processes and remote sensing. 5. Conclusion The sampling procedure we used allowed us to collect detailed data on many aspects of canopy components and faithfully, and verifiably, represent those components in three-dimensional space. Some of the basic data needed for input into models of canopy reflectance are presented. The three-dimensional geometry was reconstructed for each of the 16 trees measured for all branch segments with diameter greater than 4 cm, and in 4 trees, all branch segments having diameters greater than 2 cm. Branch and leaf size classes, lengths and angle class distributions were presented for two irrigation treatments, one receiving 100% of potential evapotranspiration and the other receiving 33% evapotranspiration for two years prior to measurement. These data were used with specific weights to determine the distributions of canopy biomass. The beta distributions were determined for each irrigation treatment. 29

Our data set provides a good picture of the consequences of two years reduced irrigation on walnut tree morphology: less biomass of stems, fruits and leaves, lower LAI, less allocation to large stems, fewer sucker shoots, and more vertically inclined leaf angles. These data can be used to test and validate canopy photosynthesis and carbon gain models. 6. Acknowledgements We wish to thank the University of California Kearney Agricultural Center fbr their assistance and cooperation during this experiment, and to Dr. David Goldhamer for the use of his experimental orchard. We wish to thank Narinder Chauhan, Peter Collins, Jatinder Singh, Curtis Smith, and Erik Ustin for field assistance. This research was supported under NASA grant NAGWl101 subcontract from the University of Michigan (#204272). 7. References Caldwell, M.M., Meister, H.-P., Tenhunen, J.D., and Lange, O.L., 1986, Canopy structure, light microclimate and leaf gas exchange of Quercus coccifera L. in a Portuguese macchia: measurements in different canopy 30

layers and simulations with a canopy model. Trees, 1, 25-41. Campbell, G.S., 1986, Extinction coefficients for radiation in plant canopies calculated using ellipsoidal inclination angle distribution. Agricultural and Forest Meteorology, 36, 317-321. Campbell, G.S., and Norman, J.M., 1989, The description and measurement of plant canopy structure. In Plant Canopies: Their Growth, Form, and Function, edited by G. Russell, B. Marshall, and P.G. Jarvis. (New York: Cambridge University Press, Society for Experimental Biology, Series 31). Camillo, P., 1987, A canopy reflectance model based on an analytical solution to the multiple scattering equation. Remote Sensing of Environment, 23, 453-477. Choudhury, B.J., 1987, Relationships between vegetation indices, radiation absorption and net photosynthesis evaluated by a sensitivity analysis. Remote Sensing of Environment, 22, 209-233. Cooper, K., Smith, J.A., and Pitts, D., 1982, Reflectance of vegetation canopies using the adding method. Applied Optics, 21, 4112-4117. 31

de Wit, C.T., 1965, Photosynthesis of Leaf Canopies. Agricultural Research Report, 663. (Wageningen: Centre for Agric. Pub. and Doc.). - Fisher, J.B., 1986, Branching patterns and angles in trees. In On the Economy of Plant Form and Function, edited by T.J. Givnish (Cambridge: Cambridge University Press). Goel, N.S., 1988, Models of vegetation canopy reflectance and their use in estimation of biophysical parameters from reflectance data. Remote Sensing Reviews, 4, 1212. Goel, N.S., and Strebel, D.E., 1984, Simple beta distribution representation of leaf orientation in vegetation canopies. Agronomy Journal, 76, 800-802. Goel, N.S., and Grier, T., 1988, Estimation of canopy parameters for inhomogeneous vegetation canopies from reflectance data. III. TRIM: A model for radiative transfer in heterogeneous three-dimensional canopies. Remote Sensing of Environment, 25, 255-293. Goldhamer, D.A., Beede, R., Sibbett, S., DeJong, T.M., Ramos, D., Phene, R.C., and Doyle, J., 1988, Third 32

year effects of deficit irrigation on walnut tree performance. Walnut Research Reports. Walnut Marketing Board, Sacramento, CA. pp. 42-52. Greenberg, D.P., Cohen, M.F., and Torrance, K.E., 1986, Radiosity: A method for computing global illumination. The Visual Computer, 2, 291-297. Hapke, B., 1984, Bidirectional reflectance spectroscopy. 3. Correction for macroscopic roughness. Icarus, 59, 41-59. Jarvis, P.G., and Leverenz, J.W., 1983, Productivity of temperate deciduous and evergreen forests. In Encyclopedia of Plant Physiology, Physiological Plant Ecoloqgy IV, NS. Ecosystem Processes: Mineral Cycling. Productivity and Man's Influence, edited by O.L. Lange, P.S. Nobel, C.B. Osmond, and H. Ziegler (New York: Springer Verlag). Kimes, D.S., and Kirchner, J.A., 1982, Radiative transfer model for heterogeneous 3D scenes. Applied Optics, 21, 4119-4129. Lemeur, R., and Blad, B.L., 1974, A critical review of light models for estimating the shortwave radiation 33

regime of plant canopies. Agricultural Meteorology, 14, 255-286. Lindenmayer, A., 1987, Models for multicellular development: characterization, inference and complexity in L-systems. In Trends, Techniques and Problems in Theoretical Computer Science. Lecture Notes in Computer Science 281, edited by A. Kelmenova and J. Kelmer (Berlin, Springer-Verlag). Linder, S., 1986, Potential and actual production in Australian forest stands. In Research for Forest Management, edited by J.J. Landsberg, and W. Parsons (Melbourne, Australia: CSIRO). Mandelbrot, B.B., 1982, The Fractal Geometry of Nature (New York: W.H. Freeman and Co.) Monsi, M. and Saeki, T., 1953, Uber den lichtfaktor in den Pflanzengesellschaften und seine Bedeutung fur die Stoffproduktion. Japanese Journal of Botany, 14, 2252. Monteith, J.L., 1981, Does light limit crop production? In Physiological Processes Limiting Plant Productivity, edited by C.B. Johnson (London: Butterworths).

Myneni, R.B., Asrar, G. and Kanemasu, E.T., 1987, Light scattering in plant canopies: The method of successive orders of scattering approximations (SOSA). Agricultural and Forest Meteorology, 39, 1-12. Myneni, R.B., Ross, J. and Asrar, G., 1989, A review on the theory of photon transport in leaf canopies. Agricultural and Forest Meteorology, 45, 1-153. Nilson, T., 1971, A theoretical analysis of the frequency of gaps in plant stands. Agricultural Meteorology, 8, 2538. Norman, J.M., 1975, Radiative transfer in vegetation. In Heat and Mass Transfer in the Biosphere, edited by D.A. deVries and N.H. Afgan (Washington, D.C.: Scripta Book Co.). Norman, J.M., 1979, Modeling the complete crop canopy. In Modification of the Aerial Environment of Plants, edited by B.J. Barfield and J.F. Gerber (St. Joseph: Amer. Soc. Agric. Eng. No. 2). Norman, J.M., and Campbell, G.S., 1989, Canopy Structure. In Plant Physioloqical Ecology: Field Methods and Instrumentation, edited by R.W. Pearcy, J.R. 35

Ehleringer, H.A. Mooney and P.W. Rundel (London, New York: Chapman and Hall). Norman, J.M., and Jarvis, P.G., 1975, Photosynthesis in Sitka spruce (Picea sitchensis (Bong.) Carr.). V. Radiation penetration theory and a test case. Journal of Applied Ecology, 12, 839-878. Norman, J.M., and Welles, J.M., 1983, Radiative transfer in an array of canopies. Agronomy Journal, 75, 481-488. Norman, J.M., Welles, J.M., and Walter, E.A., 1985, Contrasts among bidirectional reflectance of leaves, canopies and soils. IEEE Transactions in Geoscience and Remote Sensing, GE-23, 659-667. Ross, J., 1981, The Radiation Regime and Architecture of Plant Stands (The Hague: W. Junk, Publishers). Ross, J., and Marshak, A.L., 1985, A Monte Carlo procedure for calculating the scattering of solar radiation by plant canopies. Soviet Journal of Remote Sensing, (English translation), 4, 783-801. Ross, J., and Marshak, A.L., 1988, Calculation of the canopy bidirectional reflectance using the Monte Carlo method. Remote Sensing of Environment, 24, 213-225. 36

Russell, G., Jarvis, P.G., and Monteith, J.L., 1989, Absorption of radiation by canopies and stand growth. In Plant Canopies: Their Growth, Form and Function, edited by G. Russell, B. Marshall and P.G. Jarvis (New York: Cambridge University Press, Society for Experimental Biology, Series 31). Sellers, P.J., 1985, Canopy reflectance, photosynthesis and transpiration. International Journal of Remote Sensing, 6, 1335-1372. Sellers, P.J., 1987, Canopy reflectance, photosynthesis and transpiration. II. The role of biophysics in the linearity of their interdependence. Remote Sensing of Environment, 21, 143-183. Suits, G.H., 1972, The calculation of the directional reflectance of a vegetative canopy. Remote Sensing of Environment, 2, 117-125. Vanderbilt, V.C., 1985, Measuring Plant Canopy Structure. Remote Sensing of Environment, 18, 281-294. Verhoef, W., 1984, Light scattering by leaf layers with application to canopy reflectance modeling: The SAIL model. Remote Sensing of Environment, 16, 125-141. 37

Wang, L-P., 1988, Crown structure, radiation absorption, photosynthesis and transpiration. (University of Edinborough: Ph.D. Dissertation), 188 p. Welles, J.M., 1978, A model of foliage temperatures for a heated orchard. (Pennsylvania State University: M.S. Thesis), 88 p. Welles, J.M., 1988, A Bidirectional Reflectance model for non-random canopies. (University of Nebraska: Ph.D. Dissertation), 121 p. Whitfield, D.M., and Conners, D.J., 1980, Penetration of photosynthetically active radiation into tobacco crops. Australian Journal of Plant Physiology, 7, 449-461. 38

Table 1. Regressions of leaf geometry variables versus height (m) in the canopy. Irrigation Treatment 100% ET 33% ET Variable Intcp Slope R2 Intcp Slope R2 Terminal Lflt Folding Angle 174.60 -10.6.2013*** 168.55 -7.76.1306** Lateral Lflt Folding Angle 170.47 -5.42.0671** 181.63 -8.32.1785*** Terminal Midrib Zenith Angle 143.04 3.18.0269* 154.66 0.70.0023 Lateral Midrib Zenith Angle 112.15 6.66.0900*** 125.23 5.64.1436** Rachis Zenith Angle 137.46 -1.94.0059 154.14 -7.95.0980** Terminal Lflt Width (cm) 8.27 -0.59.1910*** 7.94 -0.38.0809* Lateral Lflt Width (cm) 5.33 -0.19.0581** 5.37 -0.13.0211 Number of Leaflets per Leaf 4.97 0.69.2694*** 5.37 0.43.1685*** *= p <.05; ** = p <.01; *** = p <.001 39

Table 2. Descriptors for the trees sampled in the tree-level sampling. Tree Descriptor R2T09 R3T09 R2T14 R3T17 Irrigation treatment 100 100 33 33 Max ht of Cls 5 stems (cm) 440 447 306 316 Max ht of tree (cm) 712 585 549 558 Biomass (kg) Stems 30.47 34.97 30.19 28.08 Stems, > 4 cm 18.15 25.04 17.18 17.38 Leaves 6.14 5.17 5.45 5.14 Fruits 16.00 13.23 11.74 10.46 Total 52.61 53.37 47.38 43.68 Area (m2) Stems (Length * Diam) 2.52 2.49 2.41 2.33 Leaves 82.87 69.83 73.61 69.42 LAI 3.69 3.11 3.27 3.09 LAI, projected 2.53 2.13 1.90 1.79 Sucker shoots Count 7 9 0 4 Leaves per shoot 12.1 13.9 0 9.7 Leaf area, % of total 3.2 5.8 0 1.7 4O

Captions to Illustrations Figure 1. Illustration of the angles measured on stems and leaves. A. Zenith angle of the rachis (Or), and zenith angle of the branch (eb). The azimuth direction (8,) is measured from true North. B. The zenith angle of a leaflet (el) is measured from the vertical to the top surface of the leaflet. C. The folding angle about the leaflet midrib (em) decreases with increased folding. Figure 2. Graphic representation of the three-dimensional structure of stems in the orchard. Orchard-level sampling stems ( Class 5, > 4 cm diameter) are shown in white for the sixteen trees measured. The eight 100% ET treatment trees are to the upper left; 33% ET trees are to the lower right. Tree-level sampling stems are shown on four trees sampled at this level. Diameter class is indicated by color: Class 4, green; Class 3, blue; Class 2, yellow; Class 1, pink. The gap between the two blocks of trees represents the space occupied by the unmeasured border trees. Figure 3. Graphic depiction of tree R3T09 stems sampled in the tree-level sampling and branch-level sampling. Diameter class of tree-level sampling stems is indicated by color as in Figure 2. Stems of the branch-level sampling are shown in orange at the right-center of the tree. Not all Class 1 41

and Class 2 stems are depicted as explained in the text (see section 2.C.l.b. Tree-level Sampling). Figure 4. Comparison of the heights of 58 points in the canopy measured with a plumb line (Measured Height) or calculated from tree-level geometric sampling data (Calculated height). Figure 5. Comparison of the horizontal distance from the trunk of 58 points in the canopy measured with a plumb line (Measured Distance) or calculated from tree-level geometric sampling data (Calculated Distance). Figure 6. The cumulative length of stem segments sampled in the orchard-level sampling (eight trees per treatment) shown by diameter size class. Figure 7. The cumulative length of stem segments sampled in the tree-level sampling (two trees per treatment) shown by diameter size class. Figure 8. The cumulative frequency of azimuth angles by stem size class for the tree-level sampling. Figure 9. The cumulative frequency of zenith angles by stem size class for the tree-level sampling. 42

Figure 10. The cumulative frequency of azimuth angles by leaf type. Terminal leaflets, "T", and lateral leaflets, "L", in the 100% ET treatment, "100", and 33% ET treatment, "33", are denoted. Figure 11. The cumulative frequency of zenith angles by leaf type. Terminal leaflets, "T", and lateral leaflets, "L", in the 100% ET treatment, "100", and 33% ET treatment, "33", are denoted. Zenith angle can exceed 1800 if the top surface of the leaf faces downward. Figure 12. Plot of the two parameters of the beta distribution, mu and nu, for the five stem diameter size classes in the 100% ET treatment (x) and the 33% ET treatment (o). The positions of idealized distributions (*) are labelled on the plot. Figure 13. Plot of the two parameters of the beta distribution, mu and nu, for terminal leaflets, "T", and lateral leaflets, "L", in the 100% ET treatment, "100", and 33% ET treatment, "33". The positions of idealized distributions (*) are labelled on the plot. 43

C) o.. 0~~~~~~~~ Oo 0~~~~~~~~~~~ 0 /a 0 0~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~..,,~ I~~~~~~~~~~~~~~~C~~ ) ~' ~: 3 CX)~~~~~~~~~~~~~~~~~~~~~~, ".'~~~~~~~~~~.."..' 00~001~ O0 0

Figure 2. Martens et al., "Measurement of Tree Canopy Architecture"'

Figure 3. Martens et al., "Measurement of Tree Canopy Architecture"

600 500 + 400- + + I 200+ Y=5.19+0.96X ( 200- +r=0984 100 100- i 100 200 300 400 500 600 Calculated Height (cm) Figure 4. Martens et al., "Measurement of Tree Canopy Architecture"

400 S 300 0) + + + cd VI 200 100 +0 YO.82+O.99X OrO.986 ++ 0I 0 100 200 300 400 Calculated Distance crn Figure 5. Martens et al., "Measurement of Tree Canopy Architecture"

3000 2500 L 250ORCHARD -LEVEL E N 2000 G DI~I~I I O 100% T 1500 ET H W~l~l~n ~ 33% 1000 ET 500 5 6 7 8 9 1011 12 13 14 15 16 17 18 DIAMETER CLASS crm Figure 6. Martens et al., "Measurement of Tree Canopy Architedture"

12000 10000 L 1000 REE LEVEL E N 8000 G Lii 100% T sooI6000 ET H II~II ~ 33% 4000 ET C 2000 im 1 2 3 4 5 6 7 8 9 101112131415 DIAMETER CLASS cmn F igure 7. Martens et al., "Measurement of Tree Canopy Architecture"

100 I -- t /, 0 80 0) =mI / / ~ 60 4 C~~~~~4~..,_,/'"/" Q) /t"' > 40.,' ~r,. - r /-~. IC / ~~.I., J 2ff I S~ 240 6-/ I:, C) ~~III I I, 0 90 180 270 360 Azimuth Figure 8. Martens et al., "Measurement of Tree Canopy Architecture"

100 0-0 — 8 0~~~~~~~~~~~~~~~~~1 - J# I~~~~~~ 4/i / / 1 I 408 050 / / /. QI /,I -r 0) 60 5 k ~~~~4/ /I A-t II /0 I S 20 1~~~~~~I ~ ~ 20~~~~0 0 45 90 135 180 Zenith Angle Figure 9. Martens et al., "Measurement of Tree Canopy Architecture"

100 J 0 80 1 T100 /), /I j -(L33 =f~~~~~~~~~~~~~~ 6 0.9-440 / - I ~~~~~~~~~~~~I.4 ) Cd T33 —' S 20 L100 0 90 180 270 360 Azirnuth Figure 10. Martens et al., "Measurement of Tree Canopy Architecture"

100- / 0 80 1 I >40 L33 -T33 4' - L100i - -1-0 -T100 O-1._, - I' E 20 0 30 60 90 120 150 180 210 Zenith Angle Figure 11. Martens et al., "Measurement of Tree Canopy Architecture"

4 x 100 ET o 33 ET *PLAG 3 EREC 05 2 SPH* o4 x5 03 x 1 x 4 O I*PLAN 1- UNIF *2 x2 EXTR 0. I I 0 1 2 3 4 Mu Figure 12. Martens et al., "Measurement of Tree Canopy Architecture"

4 +L33 *PLA 3 * EREC +T33 z2-* ~SPH +L100 +T100 * PLAN 19 * UNIF * EXTR 0 0 1 2 3 4 Mu Figure 13. Martens et al., "Measurement of Tree Canopy Architecture'

s)ERIM 1. OBJECTIVES The overall objectives of the Eos Synergism Study was to determine the optimum orbital tracks required for the Eos visible/IR and SAR instruments to be used in monitoring terrestrial ecosystems and other vegetated surfaces. In meeting these objectives, one principal focus of the studies conducted by the Eos Synergism Team was on determining the degree of the diurnal variations in microwave signatures from vegetated terrains. The summer field program of 1987 clearly showed that some vegetation canopies have a significant (1 to 3 dB) diurnal variation in the intensity of the radar return. This observation led to a further set of questions which need to be addressed, specifically: (1) do such diurnal variations occur in other terrestrial ecosystems, and if they do, what are the magnitudes of these variations? (2) can diurnal and seasonal variations be detected on SAR imagery collected over vegetated regions? and (3) what are the implications of these diurnal and seasonal variations on the use of information derived from SAR data as inputs into forest ecosystem models. The research conducted under our subcontract focused on addressing these issues. Specifically, the objectives of the research conducted by ERIM using the FY88 funding were to: (1) assist in the designing and conducting of an airborne SAR data collection program over a forested test site where seasonal and diurnal variations in the microwave signature are expected; (2) further develop and evaluate techniques required for calibration of airborne SAR imagery over vegetated regions; and (3) develop a methodology whereby the information derived from SAR data can be linked to forest ecosystem models.

ERIM 2. MAJOR ACCOMPLISHMENTS During the past year, the following items were accomplished: -A protocol for connecting forest ecosystem and microwave backscattering models was developed. -Techniques for polarimetric calibration of airborne SAR data using a combination of calibrated targets and background clutter was developed and evaluated. -Airborne SAR data collections over the Duke University Research Forest were planned and conducted. Part of the research efforts over the past year have focused on the connections between forest ecosystem and microwave backscattering models. A paper was written to explore these connections (Kasischke and Christensen, 1990; see section 5 below). The connectivity between these models is a two-way street. On the one hand, the desired objective is to utilize remotely sensed data to provide inputs to the ecosystem models. However, in order to fully develop techniques to extract the required inputs from remote sensing data, it is often necessary to develop and validate theoretical remote sensing models which predict a signature based on the composition and structure of vegetation canopy. Once such models have been validated, they can be exercised to determine what canopy components are primarily responsible for the remote sensing signature, and hence assist in defining the information content of that signature. Developing a set of inputs for these theoretical models is often not possible due to the number of different scattering elements which comprise a forest canopy, as well as the dynamic nature of tree canopies. However, using existing models of canopy structure and physiology may provide the means to develop the required input data set for the theoretical scattering models. The paper by Kasischke and Christensen explores the connections between forest ecosystem and microwave scattering models. Forest ecosystem processes themselves can be organized on a hierarchical basis, e.g., processes can occur on an individual tree basis, at a stand level, at a community level, on up to ecosystem and biosphere levels. In utilizing remote sensing data as inputs to these ecosystem process

ERIM models, several limitations must be recognized. First, remote sensing data represents a snap shot of the ecosystem being processed; thus, in order to study a process, imagery from different dates has to be combined. Second, the parameters required as inputs to various ecosystem process models are not necessarily those parameters which are estimated from the remote sensing data. Often, these parameters have to be inferred from the remote sensing parameters. In the paper, the connection between the different models are explored using old-field loblolly pine forest ecosystems as an examples. Specifically, the potential of SAR to estimate aboveground biomass is explored. There are two types of forest ecosystem models which can be used to assist the theoretical microwave modeler in his validation activities. First, there are those models which deal with describing tree geometric parameters, which are referred to as static condition models. Next there are models which deal with daily and seasonal changes of tree growth and physiology. These are referred to as dynamic models. Examples of how these models are used in the overall model validation process are presented in the paper. In developing the data sets necessary for studies which will eventually lead to an Eos-type system, it is highly likely that data sets from various SAR satellites will be used (e.g., SIR-C, ERS-1, JERS1, Radarsat, and the Eos SARs themselves). In order to maximize the time-period overwhich observations from these different SARs, calibration procedures will have to be developed and validated. In addition, one of the potential new advances in SAR remote sensing of vegetation canopies is the polarimetric capabilities of SARs. Techniques needed for calibration of these new polarimetric must also be developed and validated. As will be discussed below, the next phase for the Eos Simultenaity Studies will involve utilizing aircraft SAR data. In all likelihood, data from two different systems will be used: the DC-8 JPL SAR and the P-3 ERIM/NADC SAR. Thus, during the present studies, techniques to cross-calibrate these two systems were explored. These studies focused on exploring the similarities and differences between these two systems, and determining whether similar approaches

@ERIM were suitable for calibration of these systems. This study, in essence, is the first step in cross-calibrating these two SARs. A paper (Sheen et al., 1989) resulted from this study. This paper points out the major differences between these two systems, which lie in the antenna design and some differences in the radar receivers. How these differences alter the transmit and receive distortion matrices needed in polarimetric calibration are discussed. It is demonstrated that despite these differences, a similar approach can be used to calibrate the two different SARs. This technique involves using a combination of calibrated targets (corner reflectors and active radar calibrators) as well as clutter within the scene itself. This clutter calibration techniques involves exploiting the expected phase characteristics of the clutter, primarily the fact that one can assume that the two cross-polarized channels are completely uncorrelated and that the.two like polarized channels are expected to be correlated. Finally, using FY87 funding, an airborne SAR data collection was planned and carried out. This data collection occurred at the Duke University Research Forest in August/September 1989. This experiment was intended to determine was degree of diurnal and day-to-day variation exists in SAR signatures over a natural forest ecosystem. On 31 August 1989, approximately 15 passes of data were collected with the P-3 SAR System. On 2 September, approximately 10 passes of data were collected with the DC-8 SAR and on 3 September, an additional 3 passes of data were collected. Under the Simultaneity funding, calibrated radar targets (trihedrals and ARCs) were deployed to support these flights.

ERIM 3. PLANNED ACTIVITIES During the remainder of 1989 and during 1990, we plan to: -In conjunction with Duke, U of Mich, and JPL scientists, select test sites within the Duke University Research Forest from which radar signatures will be extracted. -Process and calibrate P-3 SAR data collected during the August mission. -Compare data from all three frequencies (X, C and L-bands) collected in the early morning and mid-afternoon to determine if any diurnal trends can be detected. -Assist in the calibration of data collected by the DC-8 SAR by acting as a consultant to JPL during their processing activities. -In conjunction with JPL and U of Mich scientists, compare the various signatures extracted from the two radar data sets to determine if any day-to-day and diurnal variations are detected by the SARs. -In conjunction with JPL, Duke and U of Mich scientists, compare any detected changes in radar signature to the ground truth collected during the experiment to determine the causes of those changes.

ERIM 4. PERSONNEL SUPPORTED The following personnel received support under this year's funding: Eric S. Kasischke of ERIM was the Principal Investigator for this program. His responsibilities included development of techniques to connect forest ecosystem and microwave backscatter models, evaluation of SAR calibration techniques, and planning of airborne SAR data collections, and planning and conducting of ground-truth activities associated with the airborne SAR data collection programs. Dan R. Sheen of ERIM worked on calibration of the P-3 SAR data used in this study. Finally, Norman Christensen of Duke University worked as a consultant to this program, and assisted in the model development activities as well as assisted in the deployment of calibration targets during the airborne SAR experiments. 5. PUBLICATIONS Two papers were published using funding from our subcontract: Kasischke, E.S. and N. L. Christensen, Jr., Connecting Forest Ecosystem and Microwave Backscatter Models, Int. J. Remote Sens., in press, 1990. Sheen, D.R., A. Freeman and E.S. Kasischke, Phase Calibration of Polarimetric Radar Images, IEEE Trans. Geosci. Remote Sens., 27, pp. 719-731, 1989.

ON C;GEOSCIENCE AND REINIOTE SENSING. VOL. 27. NO 6. NOVEMBER 1989 711 phase Calibration of Polarimetnc Radar Images DAN R. SHEEN, MEMBER, IEEE, ANTHONY FREEMAN, MEMBER, IEEE, AND ERIC S. KASISCHKE, MEMBER, IEEE Abstract-"The problem of phase calibration between polarization Other topics which are quite important in calibrating channels of an imaging radar is addressed in this paper. The causes of polarimetric radar and which will not be focused on in th various types of phase errors due to the radar system architecture and paper include the amplitude calibration, registration b system imperfections are examined. A simple model is introduced which paper include the amplitude calibration, registration b csa explain the spatial variation in phase error as being due to a dis- tween channels, signal-to-noise ratio (SNR), and chann placement between the phase centers of the vertical and horizontal an- coupling. It is possible to decouple the phase calibratic tennas. It is also illustrated that channel leakage can cause a spatial from these topics to a certain extent. Of course, one shou variation in phase error. Phase calibration using both point and dis- consider all of the different performance parameters as tributed ground targets is discussed and a method for calibrating phase part of designing or totally calibrating a radar [8] T using only distributed targets is verified, subject to certain constraints. Experimental measurements using the NADC/ERIM P-3 SAR system radar parameters should be specified in a consistent ma and NASA/JPL DC-8 SAR are presented. Both of these systems are ner. A sufficient signal-to-noise ratio (SNR) is especia. multifrequency, polarimetric, airborne, Synthetic Aperture Radar important in the calibration of the radar. For example, (SAR) systems. Good polarimetric calibration is required to fully ex- the SNR were less than 12 dB, a RMS phase error of 1 ploit the promising remote-sensing capabilities of these instruments would result. Other aspects of calibration such as cc pling or noise will only be discussed when they imp. I. INTRODUCTION the relative phase measurements described in this pape IMAGE data from Synthetic Aperture Radars (SAR's), In Section II a brief description of the two radars which measure the complete polarimetric scattering given, with emphasis on the parameters important to ph, matrix on a pixel-by-pixel basis, have recently become calibration and to end users of data. In Section III we available within the microwave remote-sensing commu- view the basic theory of the scattering and covariai nity. The Jet Propulsion Laboratory (JPL) has developed matrices and how they are observed with a polarimei and flown the NASA/JPL DC-8 SAR, which operates at radar. Then calibration procedures using point targets I C-, L-, and P-bands and is a prototype for the next Shuttle clutter are discussed. In Section IV a simple model to Imaging Radar (SIR-C) [1]. In conjunction with the Naval scribe a spatially varying phase error caused by an ofi Air Development Center (NADC), the Environmental in the phase centers of the antennas is developed. In S Research Institute of Michigan (ERIM) has developed and tion V we give examples of phase distortions from b flown the NADC/ERIM P-3 SAR, which is a high-reso- the NADC/ERIM P-3 and the NASA/JPL DC-8 pol lution system operating at X-, C-, and L-bands [2], [3]. metric imaging radars. Phase calibration results using 1: In this paper the experimental results collected by these point and distributed targets are presented. systems and analyzed by both ERIM and JPL will be pre- II. RDAR SYSTEM OVERVIEW sented. To date, most uses of imaging radar have not required In Table I some fundamental specifications for the accurate phase information in the image. Exceptions to and DC-8 SAR are listed for side-by-side comparison. this include applications such as phase differencing [4], systems are quite similar, with both being able to ga SAR interferometry and phase unwrapping [5], and anal- polarimetric data. Both systems can operate at the Cysis of ocean wave spectra from complex images [6], [7]. L-bands. The P-3 SAR has an additional capability foi This paper is concerned with the problem of calibrating X-band and has a higher resolution than the DC-8 syst the relative phases between polarization channels of a po- However, the P-3 system cannot gather polarimetric larimetric imaging radar. The problem of absolute phase simultaneously for multiple frequencies, while the E calibration does not concern us here. system can. The DC-8 system also has a P-band sy, and the P-3 does not. Manuscript received February 16, 1989; revised July 6, 1989. This work The systems also have different system architectn was supported by the U.S. Navy under Contract No. N00014-87-C-0726 In Fig. 1 a simplified diagram of the DC-8 system is g and by the National Aeronautics and Space Administration under Grant [9], which shows the path lengths for transmit horiz NAGW-1101. [9] which shows the path lengths for transmit honi'z D. R. Sheen and E. S. Kasischke are with the Environmental Research ( T H), transmit vertical ( T, V), receive horizontal Institute of Michigan, Radar Science Laboratory, P.O. Box 8618, Ann Ar- H), and receive vertical (R, V). In Fig. 2 a simpl b, MI 48 107. diagram of the P-3 system is given. One of the fu A. Freeman is with the Jet Propulsion Laboratory, California Institute dir the system is the of Technology, 4800 Oak Grove Drive, Pasadena, CA 91103. mental differences between the two systems Is that the 8 system has two receivers, while the P-3 system ha, 0196-2892/89/1100-0719$01.00 ~ 1989 IEEE

720 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. 27. NO. 6. NOVEMBER 1989 Horizontal Horizontal Antenna Antenna R. H Horizontal Receiver v Vertical T cVertical I Transmitter awitch 777 Antenna Antenna Receiver -----......l R V Receiver Fig. 1. Simplified diagram of the DC-8 SAR system illustrating the path lengths through the transmit and receive channels. Taken from reference pulse - to - pulse va [9]. pulse - to - pulse vanable attenuator [9]. Fig. 2. Simplified diagram of the P-3 SAR system illustrating path lengths through the transmit and receive channels. TABLE I A LIST OF OPERATING PARAMETERS FOR THE NADC/ERIM P-3 SAR AND NASA/JPL SAR in phase calibration at this time. Also, the simplified diaNADC/ERIM NADSA /JP grams do not show any details of how the different carrier frequencies or chirps are generated. For more detailed deCenter frequency z scriptions of the radars, see [1]-[3]. X-band 9.375 GHz - C-band 5.26 GHz 5.3 GHz L-band 1.25 GHz 1.25 GHz P-band - 440 Mz III. BACKGROUND THEORY Wavelength A. Measurement of the Scattering Matrix X-band 3.2 cm -I C-band 5.7 cm 5.65 cm L-band 24 cm 24 cml The goal of a polarimetric imaging radar is to measure P-band _ 68.1 cm the scattering matrix from an area of the earth's surface. Polarization isolation (bolaresaton oaght) The scattering matrix S relates the incident electric field X-band 23 dB - C-band 23 dB >25 dB Ei to the scattered field Eu by L-band 23 dB >15 dB P-band - >20 dB ESu ed[SB S12E ] Range resolution 3.0 ~ or 1.6 11 LS21 S2 = - r I I I(1) lsngle look) r S21 S22 L Ei2 Azimuth resolution 2.1 m 4 m (single look) where subscripts 1 and 2 usually refer to horizontal (H):Range pixel spacing 2.4 a or 1.2 a 6.67 a and vertical (V) polarization, k is the wavenumber, and Azimuth pixel spacing 1.62. 3.03 r is the range. For a reciprocal scattering medium, S12 = Image size in pixels (az x ra) 4096 x 4096 4096 x 750 s21 [10] Incidence angle range 20 to 65 deg* 20 to 70 deg We define a matrix M, which is composed of the complex value of each channel for each pixel of the uncaliIncidence angle is adjustable beyond this range for special situations. Because of path length differ-...... brated polarimetric image. Because of path length differences, gain differences, coupling between channels, and noise, the measured (complex valued) matrix M is not receiver which is switched between the V and H antennas equal to the scattering matrix. Zebker et al. [9] have deon successive pulses. The pulse-to-pulse variable atten- scribed the DC-8 radar system architecture shown in Fig. scribed the DC-8 radar system architecture shown in Fig. uator in the P-3 system can be set to different values for 1 and have used the following formula (after [111) to reall of the different transmit and receive polarizations. The different settings of the attenuator cause different phase shifts and gains for each of the combinations of transmit M = R'ST + N and receive polarizations. This has important implications R R S S12 in the theoretical development presented in the next sec- M = Rt R21 LS. S12 T TI T12 + N., N12 tion. In this paper we will be primarily concerned with LR2 R22 LSt1 S22LT21 TR22.N1 N2,j calibrating the end-to-end system and will not separate the (2) radar from the SAR processor. Of course, the radar architecture can affect the theoretical formulations. where R and Tare the receive and transmit distortion maThe simplified diagrams of the systems do not show the trices, respectively, superscript t denotes the transpose, calibration signal generator that both systems have. This and the matrix N represents the noise present in each signal generator injects a tone or synthetic signal into the channel. For an ideal radar system. R and T are both idenreceiver channel immediately after the antennas, and can tity matrices multiplied by complex constants The offbe used to radiometrically or amplitude calibrate the dif- diagonal terms of R and T represent leakage or coupling ferent receive channels. The signal generators are not used between channels and ideally would be ver, small.

e a' PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 721 71W P-3 system collects data for HH (horizontal trans- also need to phase calibrate the cross-polarized channels mit. honzontal receive), HV (horizontal transmit, vertical relative to the like-polarized channels. receive). VH (vertical transmit, horizontal receive), and Ignoring the effects of noise and channel coupling, it yV (vertical transmit, vertical receive) individually on has been shown by [9] that successive pulses. The measured matrix M cannot be expressed as in (2) because the variable attenuator setting is arg (MM1122) = arg (SS22) + do + Or changed from pulse to pulse. The phase shift through the attenuator can vary by as much as 10~ for different set- arg (M12M~l) = t (5) tings of attenuation ranging from 0 to 64 dB. The pulse- where it is the phase offset between the H and V transmit to-pulse variable attenuator creates four independent re- channels, and'r is the phase offset between the H and V ceiver channels. To model these independent gains and receive channels. With this model, polarimetric phase phases the R'ST product in (2) would have to have each calibration can be carried out by using any scene where element multiplied by a complex constant. For the P-3 the cross-polarized measurements M12 and M21 are signifSAR, we will not separate the effects of the R matrix, the icant with respect to the like-polarized (e.g., I M12 12 > Tmatrix, and these complex constant factors and only deal 10-1 I Mll 12) to estimate (Ot - 4r). To find the sum of with the total phase shifts. the transmit and receive phase differences, the authors recommend using an area in the image, where (arg B. The Covariance Matrix of Distributed Targets B. he Covariance Matrix S1122) ) = 0, given a single-bounce scattering model. For polarimetric clutter or scatter from distributed tar- Hence (, t ~ r) are easily calculated, and the measuregets, the scattering statistics can be described by the co- ment matrix M can be phase calibrated. We shall show variance matrix C [12], which is defined as from our results that this approach is justified in many circumstances, and shall examine further the range of (j~ | 5 |) 21' 22 ) clutter backgrounds over which it can be used. C= (S2lS*) (IS2112) (S21S2*2) (3) L(S22 S ) (S2S2 S2) (1S22 2 ) C. Calibration Using Trihedral Corner Reflectors The goal of polarimetric calibration is to determine the where the <'s indicate expected values, the I's indicate relative values of the elements of the R and T matrices in absolute value, and * indicates the complex conjugate. (2) to allow the determination of the scattering matrix S The expected covariance matrix for the clutter from dis- from the measured matrix M. Also, the statistics of the elements of the noise matrix N should be determined. The tributed targets can be simplified somewhat from the gen- elements of the noise matrix N should be determined. The eral expression. Under the second-order Born approxi- general procedure described by [11] is to deploy targets mation for a layer of randomly positioned particles, the with known scattering matrices and form enough indepenlike- and cross-polarized elements of the scattering matrix dent equations to determine the elements of the R and T are completely uncorrelated [12]. This is because the matrices. For example consider the scattering matrix a cross-polarized terms come entirely from higher-order scattering (two-bounce or greater), while the like-polar- - 01 ized terms usually come predominately from first-order STRI = ~ (6) scattering (single-bounce). For randomly positioned scat- 0 1 terers the higher-order scattering paths are independent of the first-order scattering paths and thus, the cross- and Dependng upon the radar system, it may be valid to like-polarized terms of the scattering matrix are uncorre- express the radar measurement of the trihedral in terms of the R and T matrices alone. The associated radar mealated. The final resultant covariance matrix as a function of the scattering matrix S has the following form: surement neglecting noise is (l2* \ (S,1522)n [T, 1R11 + T21R21 T12R1R + T22R21R (IC = SI | ) IS 0 S2 112M T11R12 + T21R22 T12R12 + T22R22i C=L2S22S 1) 0 (1S22 ( T12R11 R21 T11rllRl/T22R22 + The only phase difference which is critical is that be- = T22R2 T22R22 (7) tween the HH and VW channels. This would indicate that T1l R12 T2 1 a polarimetric radar would only have to be phase cali- L T22R22 T22 brated between the HH and W channels to collect good data on different types of ground clutter. A trihedral cor- to a good approximation if the off-diagonal terms of the ner reflector would be sufficient to do this phase calibra- R and T matrices are small compared to the diagonal tion. Of course, to examine other targets or clutter where terms. Cross- and like-polarized scatter are correlated, one would For phase calibration, we find that if the off-diagonal

722 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. 27, NO. 6. NOVEMBER 1989 terms of T and R are small, then and for phase calibration we find that arg (MlIM2)TRI = arg (TtlRl/T22R22) arg (MllM22) = dO + Or =:L (12a) ='t + Or = OrL (8) if where, = arg (Tll/T22), Or = arg (Rll/R22), and IL arg (ac*) = 0 is the total phase difference between the like-polarized channels in the radar. As was pointed out in Section III-for a single-bounce terrain, and also, A, for the P-3 SAR, OL cannot be broken down in terms arg (M21M2) = Xt - Or (12b) of 0, and Or without first removing the effect of the variable attenuator. A trihedral could be used to find (,t + In order for the above expressions to be valid the chan&r) or 4L. Additional targets such as dihedrals at a variety nel coupling must be negligible. For this to be true, the of orientations, gridded trihedrals, or Active Radar Cali- ratio of cross- to like-polarized backscatter must be much brators (ARC's) [13] could be used to solve for other un- greater than the cross-polarization isolation terms. For exknowns in the R and T matrices. For example, the pro- ample, cedure using dihedrals at a variety of orientations and b 12 T12 12 trihedrals is given in [11]. ARC's could also be used in a 2 etc. (13) similar fashion. |a IT221 D. Calibration Using Clutter A channel isolation greater than 20 dB is probably sufficient for most clutter. If these conditions apply, then The fact that clutter has a correlation between the two M M*M >, (M22M2), and (M22M>) like-polarized channels and between the two cross-polar- should approximately equal zero, since (S S* >, ized channels can be used as an aid in phase calibration ( SllS2 ), ( S22 S2, and S2221 > should theoretically [9]. This promises to be very useful, since calibration tar- be zero according to some clutter models [12] gets require careful deployment and some degree of ground support. Also, the phase calibration can change IV. SPATIAL VARIATION OF PHASE CALIBRATION spatially in the image, and many calibration targets would In Fig. 3 we present some imaging geometry charachave to be deployed to examine the spatial variation ofThe the phase. As outlined previously, it is not expected that antenna may have slightly different phase centers for verthe like- and cross-polarized channels would be correlated tical (V) and horizontal (H) polarization. This difference with each other. Thus it will not be possible to determine from H to V is given by the vector J and is exaggerated this relative phase shift using in-scene clutter. The method in the illustration. From Fig. 3 it canbe seen that one would only allow HH to be phase calibrated with respect expects the pathlengths H to V and V to H to be the same. to VV, and likewise for VH to be phase calibrated with Thus phase calibration of the HV channel relative to the HV. In the JPL data the average HV and VH phase differ- VH channel should be spatially constant in the image. ence is corrected for, and the HV and CVH channels are However, this is not the case with the HH versus VV chanaveraged together. The average phase difference between nel because of the offset d. The phase vaation across the HV and VH is included in the header of the JPL com- image can be ttenvariation across the pressed data format. Using the clutter phase statistics, one could partially phase calibrate an image and could also 4 examine how the relative phases shift spatially in the im- (L(R) X Ed -j + (14) age. R I Given a cell of background "clutter" with the follow- where XL is the phase of VV relative to HH, X is the radar ing scattering matrix: wavelength, d is the offset from H to V antenna phase a b- centers, R is the vector from the radar to the pixel of inS = a, b, c 0 O (9) terest, and Oc is a constant phase offset resulting from difb c ferent path lengths in the transmitter and receiver for HH and neglecting the effects of noise, the associated radar and W. In general, R and a will have along-track as well measurement can be expressed in terms of the R and T as cross-track components. However, in a strip-map SAR, matrices as follows: r aT11Rl + bT21R11 + bT21R11 + CT21R21 aTi2RI2 + bT22R + bT2R21 + T22R21 (10) aTiIR12 + bT21R12 + bT21R22 + CT21R22 aTi2R12 + bT12R22 + bT22R12 + CT22R22 1 Neglecting channel coupling (i.e., assuming that offdiagonal terms of Tand R are negligible) results in the along-track components should be negligible. Of r~aT1OR~ bT2~Rl course, offsets in the along-track direction that are large Mc= l 1 II IR b2211 | (11) can cause registration problems which are a potential L~bTIR22 cT22R22 jsource of error. Neglecting any along-track component,

SHEEEN a/l.: PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 723 * V antenna The terms o,0 and yro can be related to the path length * d y differences Ax, and Ax, between transmitters and receivers, via: 2w 20r A lt0 &-' kXt (O)rO X = i Xr. (20) X ~ From (20), path length differences of the order of 1/36th of a wavelength can cause phase errors on the order of 10~. In practice, it is very difficult to calibrate the entire radar transmit and receive paths to this kind of preFig. 3. SAR geometry. Used to solve for the W-HH phase difference as a function of slant range R. V and H antenna separation exaggerated and cision, particularly at the shorter wavelengths, without given by the vector d. using ground targets. Hence the necessity of calibrating phase differences between channels by using properties of R can be written in terms of the incidence angle 0, as fol- known targets. Large path-length differences can also lows: cause misregistration between polarization channels depending upon the wavelength and resolution of the syssin xi - cos iy (15) tem. R si Phase noise 0,, may be caused by the presence of adand ditive noise in the system, as given in (2), in which case [14],= cos [-] (16) E[kn] = 0 rad (21a) where A is the altitude of the SAR platform above the and ground. Using (15) and (16), (14) can be written in the SNR-1 SNR-2 1/2 following form: S.D.) = SNR rad 21b 4T r; 1' rAl -1/2 8 1 %= - -dx I [R -dy[A +Oc. (17) = dl IJ -L-r where the SNR is the (power) signal-to-noise ratio. According to this model, an SNR of only 12 dB will cause Equation expresses the phase offset L in terms of an rms phase error of 10~. Another possible cause of phase five variables: A, R, d,, dy, and O,. Of these variables, noise is the decorrelation between the measured returns the only known quantities are altitude (A) and slant range due to misregistration between polarization channels. Care (R). The approach taken in analyzing experimental data from the P-3/SAR will be to examine kL as a function of polarimeters to ensure that the image channels are in fact R, with A set equal to the actual altitude of the SAR. Then registered, to avoid excessive phase noise caused by not this curve will be determined by using a nonlinear least- comparing returns from exactly the same image cell or squares curve-fitting algorithm which can solve for d,, dY, pixel on the ground. In practice, a registration accuracy and OC, resulting in the least-square error. Because of of 0.2 times the resolution is usually sufficient phase wrap-around this solution will not be unique for Oc, Adding or subtracting (18) and (19), we obtain: but does indicate if spatial phase variations in the SAR image can be modeled as a simple H and V phase center 2w R offset. Rt[o ~ Oro] +t ~ dr) More generally, the phase offset caused by displaced Ri antenna phase centers could be written in terms of the + (Ot, ~ Om). (22) transmitter and receiver phase offsets 4, and X,. In this case the system model for phase error is, If phase noise terms in individual receiver and transmitter channels are uncorrelated, then adding or subtracting gives 2n, = o +-[d, ~ + r,. the following results: L t r 2 a(1 8 ) rX~~~ |R|E[k, ~ rn] = 0 (23a)'k = X,0k~ + X IRI $ + Fr (19) assuming a -- to r range for phase, and W~her S.D. (~ ~,)S.D.(, + = [Var (O,f) + Var (r,)]/2 (23b) re ~k,0 0ro are phase differences due to path length dafeences between receivers, the terms involving d, and Hence the RMS phase error in the relative phase beC,. caused by displacement in the cross-range direction tween HH and W or between HV and VH is greater than cie, ~uective antenna phase centers on transmit and re- the RMS phase errors in either the transmit or receive and, and ~rn are zero mean phase noise terms. channels individually.

724 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. 27. NO. 6. NOVEMBER 1989 Fig. 4. X-band, polarimetric image phase referenced to the HH channel. The image is of a forested region in Alaska containing trihedrals and dihedrals. V. RESULTS image corresponds to the phase of the data. The color bar In this section we present results from the initial exper- at the bottom of the image goes from - 1800 on the left iments designed to study the phase characteristics of the to + 180 on the right. The image has been phase referNADC/ERIM P-3 and the NASA/JPL DC-8 SAR's. Be- enced to the HH channel. cause of the different architectures for these two polari- In Fig. 5, histograms of the relative phase for the formetric SAR's (which result in the slightly different cali- ested region are presented. In Fig. 5(a) the histogram of bration approaches), we present these results separately. the phase of VV versus HH is shown, and one can see that it appears to be Gaussian, with a width which is confined in the interval -ir to ir and with a mean of 0. This is A. X-Band Calibration of the P-3/SAR fortunate, since a wider distribution of phases would exIn Fig. 4 a P-3/SAR X-band image from a forested re- hibit 2ir ambiguities and one could not use the clutter for gion of approximately 1-km square is shown. This image phase calibration. The phase distribution of HV versus VH was initially calibrated using in-scene trihedral and di- is given in Fig. 5(b). In theory, for an ideal. calibrated. hedral corner reflectors [11]. This is one approach of monostatic radar this should be a delta function at 0~. As many, and now we shall investigate the utility of clutter illustrated in this figure, the HV versus VH phase is not for phase calibration using this image. The image is com- always 0, but is approximately a Gaussian centered at 0. posed of four smaller images which (clockwise from the with a standard deviation much less than that of TV versus upper left corner) correspond to the HH, VH, HV, and W HH. In Fig. 5(c), a histogram of the phase between VV polarized channels. The intensity of the image corre- and VH is shown, which indicates that these two polarisponds to the amplitude of the data, while the color of the zations have a uniformly distributed phase difference and

SHEEN er al PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 725 that the VV and HH reflections come predominately from single-bounce scattering and thus tend to be in-phase. Fig. 5(a) is a strong indication that for X-band scattering from trees, this is the case. Note that the standard deviation of the phase is significantly less than 27r. The V and HH channels may not always tend to be in-phase. For example, at L band the ground-trunk interface may act like a horizontal dihedral and be a significant source of scattering. In this case VV and HH would be 180~ out of phase. For this type of clutter one could not assume an average phase difference of 0~ between VV and HH. An exhaustive study of clutter from different terrains has not yet been performed, but it is hoped that clutter from thick vegetation with random orientation would usually aid in-phase calibration. The average phase difference as a function of range for the X-band tree data is presented in Figs. 6 and 7. The data has a pixel size of 2.4-m in range, and 1.6-m in azimuth. A subset of the image, which was 4096 pixels in range and 100 pixels in azimuth, was extracted. The average phase difference for a given range was computed by averaging together the phases for 100 azimuth pixels. The HV-VH phase is plotted in Fig. 6 and seen to be zero across most of the image, with the exception of some nearrange points. The near-range points are in the proximity of the nadir reflection (at range 143) and should be excluded because the pre-nadir response is dominated by,....... o' noise and the antenna response at these angles is uncer(b) tain. The HV-VH phase result is anticipated since the average phase of HV versus VH is not expected to vary across the image. The HV versus VH was calibrated by using some 450 dihedrals in the image, so that is why it is a constant 0 rather than being some other constant. The reflection from a 45 dihedral is completely cross polarized, and the HV and VH channels should be corrected so that they are in-phase. In Fig. 7 the average phase of VV-HH is plotted as a function of range. This results in a curve which could be fitted to the theoretical curve given by (17) by using a nonlinear least-square curve-fitting algorithm. The results are plotted with a dotted line in Fig. 7. The parameters used in (17) are A = 2299 pixels, and R = 2156 + range record number, and the fitted parameter i- o n values are d, = 0.26X, dy = -0.017X, and 0~= -63.3~. (c) The fitted value of c, is not unique because of phase wrapFig. 5. Histograms of phase differences for X-band tree-clutter data: around, but it does give the expected variation in phase (a) VV-HH, (b) HV-VH, and (c) W-VH. across the image quite accurately. The spatial variation (in range) of the phase of the image could easily be corare uncorrelated. The other combinations of like- and rected by multiplying by a spatially varying, complex calcross-polarizations were also examined and seen to be un- ibration factor. correlated. Fig. 5 indicates that the mean value of the This section has illustrated that radar clutter from disphase difference between VV and HH, as well as between tributed trees has prove to be useful for in-phase calibraVH and HV, is zero for clutter. In uncalibrated data these tion of VH relative to HV, and HH relative to W for the two mean phase differences would not necessarily be zero, P-3/SAR. Also, this section has demonstrated that the relbut could be corrected by phase shifting one of the like- ative phase variation across the P-3/SAR image could be polarized channels and one of the cross-polarized chan- modeled as a simple H and V phase-center offset. This is nels. indicated by the fact that the least-squares fit worked well. One important assumption in using the clutter to cali- To calibrate the like-polarized channels relative to the brate the W channel with respect to the HH channel is cross-polarized channels, a dihedral or ARC would have

726 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL. 27. NO. 6. NOVEMBER 1989 180~ - I Fig. 9 shows a plot of the average HH/ VV phase dif1640 i ference over the whole swath, calculated by averaging 140 L 420 -L i MllM2 coherently over 1024 points in each range gate,,00 centered in azimuth on the dry lake bed. This measure of 80 ~ J the L-band HH/ VVW phase difference is uniform across the Q) 60 t' swath, with a mean value of 158.0~ and standard devia$ 20 | tion of 4.6~. From this plot we infer that the H and V X O,,J tantennas have common phase centers (see (22)). The DC0 -20 8 and P-3 SAR's do have different-type antennas, so the' -,60-~ fact that the DC-8 phase centers are co-located is not sur-80 prising. Indeed, this is the ideal. -0,o - Figs. 10 and 11 show plots of the phase difference be-120 tween the HV and HH channels, and the HV and VH chan-140 - nels, respectively. The slight phase variation visible in -160 Fig. 11 across the swath could be due to a misregistration Re512 102 62048 2560 3072 3584 4096 between transmit and receive phase centers for the crossRange record number polarized channels, but a more likely cause of the variaFig. 6. Average phase differences (HV- VH ) as a function of the slant range Fg6.Aefor X-band tree-clutter data. tion across the swath is the contribution of the cross-polarized contamination, which is typically - 15 dB down 180 at L-band but varies across the swath. The cross-polarized 160- contamination is probably also contributing to the phase 140 it j plot of arg (M21M > in Fig. 10, which varies signifi120- 1 -oo: ---— i cantly across the swath. In this case, since we expect the 100 8o HV and HH backscatter to be uncorrelated and the phase s 60 H i'' i to be uniformly distributed, we infer that Fig. 10 can be ~> 20 4 1 explained by the dominance of the term involving l a'2 in a. -- 20 ination terms across the swath which is also probably the do -6g-20 i.. cause of the phase slope between the HV and VH channel -8~0'~, i measurements, visible in Fig. 11. The slope is small, just -,Oo - 0.4~/degree of incidence angle, or 8~ across the total -120- swath. -1450 - Figs. 12 and 13 show the HH/W and HV/VH phase -:60 - -~ -180 2I differences for the C-band DC-8 SAR image of Gold0 512 1024 1536 2048 2560 3072 3584 4096 stone, which was obtained at the same time as the L-band Range record number image. Both curves are constant across the swath, which Fig. 7. Average phase difference ( VV-HH ) as a function of the slant range for X-band tree-clutter data. The dotted line is a fitted curve assuming suggests that the receive and transmit H and V antenna that.the H and V antennas have different phase centers. phase centers are co-located. Fig. 14 shows a plot of the HV/HH phase difference, estimated from arg ( MM l ) to be used which had an appropriate scattering matrix [11], The phase difference is fairly uniformly distributed over [13]. the first half of the swath. Then the channels appear to become correlated, and the relative phase becomes conB. C-Band and L-Band Calibration of NASA/JPL SAR stant at around 60~ for the last half. This is probably due Fig. 8 shows an L-band image of the Goldstone Cali- to the fall-off of the cross-polarized backscatter return as bration site produced by the NASA/JPL DC-8 Airborne the C band over the dry lake bed, which, estimated from SAR. The image on the left is a color overlap of LHH, LW, Me, is typically - 10 dB down from the like-polanzed in and LHV polarizations, color-coded red, green, and blue, the first half of the swath where the ground is vegetated respectively. In the image on the right the grey scale rep- versus -30 dB down over the dry lake bed. This s,ormresents total power ( I LHH I2 + I LW12 + 2 1 LHV 2), while parable with the level of cross-polarization isolation at the the color indicates the phase difference between the HH C band, which varies between -23 and -30 dB Thus and W channels. The phase distribution is centered the fall-off in backscatter over the lake bed wIll all (' tIhe around Xr and is uniform across the image. This phase dis- like-polarized returns to dominate the calculatlwn,t tribution has a nonzero mean because it is uncalibrated, M21M', and (13) is not valid. but the image can be calibrated. The dark patch near the Figs. 15 and 16 show the relative phase plots derl cd center of both images is the dry lake bed at Goldstone, from the array of trihedral corner reflectors running acro,, California, where an array of corner reflectors and other the swath (see Fig. 8) for the L- and C-bands. The error calibration devices were deployed. The corner reflectors bars on the plots are derived by assuming that any error appear as a line of brieht points in the image nn the mpr irprvon,,t,.,f. -. C......

SHEEN el al.: PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 727 Fig. 8. L-band image of the Goldstone calibration site produced by the NASA/JPL DC-8 SAR. d er _7n 1e764 -' 7 s -- as'OM in 30 a2 352 Lan ie 422 452 522 S 55 2 b t ited om- Fig. 9. Plot of the average HH/ VV phase difference over the whole swath Fig. 10. Plot of the average HV'HH phase difference over the whole s, l lhus:2 the by averaging over a small cell 64 x 64 pixels in extent, itselfl or imperfections in the coer reflector onst rIved located near the reflector array. Worst-case phase errors tion. We suggest the l atter cause as the most likely-,- -gi o Were assumed, in which P o the clutter is orthogonal to the the fairly smoowath behavFig. 10. Plotor of the relativerage HV/HH phase difference over the ahole s the frror oint target return in each polarization. It is clear that ments derived2.) forom the imL band. (LHVHH b Mack 3ro 1988 Run d. ) rse is Sonmething other than the clutter background is contribut- Results from the corner reflector array and image b;

728 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING. VOL.'- NO b. NOVEMBER 989 14 3 - 80 f 6-0- i,. X i:' - t+ + t11 A - - - = 1 I x x -! 2 t I-13 O i, e I i 30 0 35 a 40 0 45 0 50 0 55 0 60 0 30 0 350 4 0 0 45 a 50 0 55 0a 6 0 theta (eg) th Iet c(deg) Fig. 12. Plot of the average HVIH/ VH phase difference over the whole swath Fig. 14. Plot of the average HV/FIH phase difference over the whole swath for the C band. (CHHVV, May 23, 1988, Run 2.) for the C band. (CHVHH, May 23, 1988, Run 2.) calibration to give 3 similar results. Another point to note nels were of fixed lengths. For the P- SAR only the total from Table II is that none of the HV/HH phase compar- path lengths, or sums of the transmit and receive paths, difference4rasestimatedfromtheHH/VandHV/VH combinations. Path lengths outside the radar can also -6 3 6 -8 3 i 2 6 30 o 35 e 48 e 45 0 50 0 55 e 60 0 30 0 35 e 40 0 45 0 50 a 55 a 60 a thet( coer reflectors, since polarized antennas are offs thetai(de can cause a spaFig. 12. Plot of the average HHm i/ phase difference over the whole swath ig. 14. Plot of the a verag e HVphase difference over the whole swath for the Cblike- and. (CHHVVreflection. Also, the backun 2.) for the C band. (CHVHH, May 23, 1988, Run 2.) ground probably has urements ar colleted like-together for This error is small in both systems when the SNR is sufcomparison in Table II. Note the close correspondence ficient. Additional deterministics. This is errors are caused by path between all the comer reflector and background measure- length differences in the total transmit to receive channels ments, at both the L- and C-bands. This suggests that for the various polarization combinations. For the DC-8 either the comer reflector or image background ap- SAR, it could be assumed that the two transmit channels preaches, or combinations the reof, can be used for phase were of a fixed length, and also that the two receive chancalibration to give s imilar results. Another point to note nels were of fixed lengths. For the P-3 SAR only the total from Table I is that none of the HVsources ofHH phase coerropar- path lengths, or sums of the transmit and receive paths, isons appear to give a good estimate of the receiver phase were considered to be constant for the four polarization differeence, as estimated from the HHoretical stand HVI point H combinations. Path lengths outside the radar can also phase terms, at least for this radar system. We would not cause a phase error if the phase centers of the V and H expect a good ag reement for the comer reflectors, since polarized antennas are offset. This offs et can cause a spathe cross-polarized measurement is due to the leakage tially varying phase shift in the image, and the P-3 X-band from the like-polarized reflection. Also, the background data was used to illustrate this. Secondly, channel couclutter probably has uncorrelated like- and cross-polar- pling can cause spatially varying phase shifts in measured ized scatter so that the average phase difference HVHH clutter statistics. This is due to the incidence angle codeis also determined by leakaget p endence of channel coupling. The NASA/JPL DC-8 SAR C- and L-bands data were used to illustrate some of the effects of channel coupling on the measurement of clutter We have examined some of the sources of phase errors statistics. receiver and transmitter noise and is completely random. polarized components of the scatterng d uere correlated

SHEEN et al.: PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 729, HH/VV phase H1s phse 0 VH/HV phase 98 8 -9 0 + 4 -9 0 + + + + + -18 e -18 4 4 9 4 5 5 0 5 5 6 08 6 5 4- 4 45 9 45 5 0 55 6 6 5 LOOK angle LooK angle HV/HH phase VH/HH phase 18 8B I I I i _ 18 8' i' I I I 99 +49 + + + + 9 _ vlB + l +' l0 + -9 4 + +4 -9 + + -9 e+ -18 0 t -18 0 4 4 4 9 5 5 0 5 6 6 5 44 45 9 49 5 5 8 5 5 6 8 6 LOOK angle Loo ang I e Fig. 15. L-band phase plots-derived from trihedral corner reflectors. HH/VV phase VH/HV phase 188 I I l I 188 I I I 9 04.+ + + + + + 0 _ _ X d -18 0 -18 4 4 4 6 48 9 521 1 5 3 5 5 61 7 4 4 4 6 48 9 52 1 55 3 5 5 61 7 LOOK ongle LOOK angle HV/HH phase VH/HH phase 18 I I I 18I I 90 - + 9 0 -. + + * -8 0 4+ 4.+ + + 88 4 S. S. -98 e + + + + 4 -9 -18 -18 42 "4 45 6 48 9 521 3 5 5 61 -18 0 4 4 46 49 1 5 3 5 5 6 LOO, 5~ le 5 6?9 I 5; 3 54 5 7 LOOK angle LOOK angle Fig. 16. C-band phase plots-derived from trihedral corner reflectors. ( ( SlS 22 > 0 ), that there was significant cross-polar- L-bands using the DC-8 SAR. Examples of where the apized scattering (( S12S 2 > 0), and that the like- and proach isn't valid would be: Scattering from cornstalks at cross-polarized scattering were uncorrelated ( ( S i S2 ) = the L-band [4], scattering from urban areas. and other O0). These assumptions are only valid for certain limited similar predominately two-bounce scattering processes. In types of clutter. So far we have verified the approach for these cases (S1iS 12 > 0. Also, scenes with low-cross X-band scattering from a tree canopy using the P-3 SAR polarized scatter such as ocean scenes would be difficult and surface scattering from a dry-lake bed at the C- and to use because system leakage could predominate the

730 IEEE TRANSACTIONS ()N GE()SCIENCE AND REM()TE SENSING. VOL 2'. N() 6. NOVEMBER 1989 TABLE II P-3/SAR for their efforts in collecting the data used in this COMPARISON OF AVERAGE RELATIVE PHASE (IN DEGREES) ACROSS THE study. SWATH DERIVED FROM TRIHEDRAL ARRAY AND BACKGROUND CLUTTER FROM GOLDSTONE IMAGE REFERENCES HH/VV H'/;'VH HV/HH Or * [1] D. N. Held, W. E. Brown, and T. W. Miller, "Preliminary results L-Band from the NASA/JPL multifrequency, multipolarization synthetic aperture radar,"' in Proc. IEEE Nat. Radar Conf. (Ann Arbor, MI), Corner Reflectors 172.2(, 5.0)* 65.7(,12.0) 32.4(,18.7) 53.3r. 1988, pp. 7-8. Apr. 1988, pp. 7-8. Background 158.0(a 4.6) 57.5(* 6.2) 22.9(*18.7) 50.3 [2] A. Kozma et al.,'"Multifrequency-polarimetric SAR for remote sensing," in Proc. IGARSS'86 (Zurich, Switzerland), Sept. 1986, vol. 1, pp. 715-719. C-Band [3] R. J. Sullivan et al., "Polarimetric X/L/C-band SAR," in Proc. Corner Reflectors -38.0(*11.1) -114.4(I10.2) -62.14(*11.4) 38.2 IEEE Nat. Radar Conf. (Ann Arbor, MI), Apr. 1988, pp. 9-14. Background -38.4(t 5.9) -106.3(* 4.8) -78.8(*59.5) 34.0 [4] F. T. Ulaby, D. Held, M. C. Dobson, K. C. McDonald, and T. B. A. Senior, "Relating polarization phase difference of SAR signals to scene properties," IEEE Trans. Geosci. Remote Sensing, vol. GE* Numbers in brackets are standard deviations taken across the swath. 25, pp. 83-92, Jan. 1987. [5] R. M. Goldstein, H. A. Zebker. and C. L. Werner, "Satellite radar interferometry: Two-dimensional phase unwrapping," Radio Sci., ** Estimates from HH/VV and HV/VH in each row. vol. 23, pp. 713-720, July-Aug. 1988. [6] I. J. LaHaie, A. R. Dias, and G. D. Darling, "Digital processing considerations for extraction of ocean wave image spectra from raw synthetic aperture radar data,' IEEE J. Oceanic Eng., vol. OE-9, pp. cross-polarized scattering and not be negligible. Experi- 114-120, 1984. [7] R. A. Cordey and J. T. Macklin, "Complex SAR imagery and mental data also indicates that the pre-nadir and nadir re- speckle filtering for ERS-I wave mode," in Proc. IGARSS'88 gions in images are not usable. (Edinburgh, UK), Sept. 1988, vol. 1, pp. 387-390. Results using calibration reflectors appear to be quite [8] A. Freeman, J. C. Curlander, P. Dubois, and J. D. Klein, "Shuttle imaging radar-C calibration workshop report," Jet Propulsion Lab., consistent with clutter results for both the P-3 and DC-8 Pasadena, CA, Tech. Rep. 88-003, Nov. 8, 1988. systems. One interesting coincidence (we believe) for the [9] H. A. Zebker, J. J. van Zyl, and D. N. Held, "Imaging radar polarDC-8 system is that trihedrals HV/ VH phase difference is imetry from wave synthesis," J. Geophys. Res., vol. 92, pp. 683701, Jan. 1987. in agreement with the phase difference calculated by using [10] J. A. Kong, Electromagnetic Wave Theory. New York: Wiley, 1986. clutter statistics. This is interesting, since it implies that [11] R. M. Barnes, "Antenna polarization calibration using in-scene rethe channel coupling into the cross-polarized receive flectors," Lincoln Lab., Lexington, MA, Proj. Rep. TT-65, Sept. 1986. channel is in-phase with the actual cross-polarized re- [12] M. Borgeaud, R. T. Shin, and J. A. Kong, "Theoretical models for ceived signal. This is not in general true and we do not polarimetric radar clutter," J. Electromagn. Waves Appl., vol. 1, pp. advise using trihedrals to calibrate the HV channel relative 73-89, Apr. 1987. to the VH channel. [13] D. R. Brunfeldt and F. T. Ulaby, "Active reflector for radar calibrato the r/H channe. tion," IEEE Trans. Geosci. Remote Sensing, vol. GE-22, pp. 165In this paper the main focus has been on the use of clut- 169, Mar. 1984. ter statistics to phase calibrate, as well as on the possible [14] R. B. Dybdal and R. H. Ott, "Coherent RF error statistics," Trans. variation of phase calibration with range. Phase calibra- Microwave Theory Tech., vol. MTT-34, pp. 1413-1420, Dec. 1986. variation of phase calibration with range. Phase calibration using reference reflectors is fairly well understood and has been discussed previously [11], and an example procedure using trihedrals was briefly discussed in Section III-C. A complete procedure to derive all of the elements of the T and R matrices would require at least three different types of reference reflectors, including trihedrals and dihedrals at two orientations [11], or three ARC's in different configurations. The expense and difficulty in de- Dan R. Sheen (S'82-M'86) received the B.S. deploying many reference targets necessitates the use of gree (1981) in electrical engineenng from Washington State University, Pullman, and the M.S. clutter statistics as a supplementary tool in phase calibra- 1983) and Ph.D. (1987) degrees In electncalention. With both systems, the deterministic phase errors gineering from the University of Illinois at Urdescribed in this paper could be removed through proper bana-Champaign. Currently, he is a Research Engineer In the Rcalibration experiments using clutter and reference reflec- dar Science Laboratory at the Environmental Ra1 tors. The calibration procedures for both systems are search Institute of Michigan (ERIM). Ann Arbor. slightly different because the system architectures are not ince joining ERIM in 198b7 ha s pnrlftS workb a been on the calibration and remoc-senslng a PI I cations of polarimetric synthetic aperture radar He has patnilpled iD I' merous airborne data collection experiments and subscquentl data aalyll; ACKNOWLEDGMENT activities. During the period from 1981 to 1987 he worted i a Rg d Assistant in the Ionosphere Radio Laboratory at the L'nlvefnlY of The research presented in this paper was carried out by at Urbana-Champaign. His research work at the laboratory f tt; both the Environmental Research Institute of Michigan perimental study and modeling of wave propagain'h ppa (ERIM) and the Jet Propulsion Laboratory (JPL). The au- atDr Sheen is a member of the American Geoph cal t no. thors would like to thank J. Lydeu and the crew of the Phi, Eta Kappa Nu, and Tau Beta Pi.

SHEEN; r a/ PHASE CALIBRATION OF POLARIMETRIC RADAR IMAGES 731 Anthony Freeman (M'83) received the B.Sc. Eric S. Kasischke (A'85-M'88) received the B.S. (Hons.) degree in mathematics in 1979 and the degree in natural resources in 1974 and the M.S. Ph.D. degree in astrophysics in 1982, both from degree in remote sensing in 1980 from the Unithe University of Manchester Institute of Science versity of Michigan, Ann Arbor. He is presently and Technology, Manchester, England. a Ph.D. degree candidate at the University of Between 1982 and 1987 he worked at the Mar- Michigan. coni Research Centre, Chelmsford, England. on Since 1976 he has worked as a Research Scimoving target imaging with SAR, aircraft SAR entist in the Radar Science Laboratory of the motion compensation, SAR design studies and Environmental Research Institute of Michigan image quality assessment. Since March 1987 he (ERIM), Ann Arbor. His research activities have has been employed by the Jet Propulsion Labo- concentrated on developing techniques to exploit rator. Pasadena. CA, as a radar systems specialist and group leader. His information collected by imaging radars for both oceanic ac. terrestrial current area of research is in the field of multifrequency, multipolarization applications. He has directed numerous airborne data colle;ilon experiSAR calibration. ments and has been involved in the calibration of imaging radars for the Dr Freeman is Chairman of the Committee in Earth Observing Sensor past six years. Working Group (CEOS) on SAR-Calibration. Mr. Kasischke is a member of Sigma Xi and the American Geophysical Union.

Phase Calibration of Polarimetric Radar Images DAN R. SHEEN, MEMBER, IEEE, ANTHONY FREEMAN, MEMBER, IEEE, AND ERIC S. KASISCHKE, MEMBER, IEEE Reprinted from IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING Vol. GE-27, No. 6, November 1989

CONNECTING FOREST ECOSYSTEM AND MICROWAVE BACKSCATTER MODELS Eric S. Kasischke Radar Science Laboratory Advanced Concept Division Environmental Research Institute of Michigan Ann Arbor, Michigan USA Norman L. Christensen, Jr. Department of Botany Duke University Durham, North Carolina USA Accepted for Publication in International Journal of Remote Sensing, February 1989.

Kasischke and Christensen 1 1.0 INTRODUCTION The variety of papers presented in this issue is indicative of the substantial research effort underway to evaluate and develop techniques which utilize active microwave remote sensing systems to monitor the earth's forest ecosystems. Hopefully, these efforts, along with similar research programs focusing on visible and infrared remote sensing systems, will lead to a set of algorithms which will constitute the forest ecosystem element of the Earth Observing System (Eos). It is clear that the eventual development of these algorithms will require the close collaboration of forest ecologists, remote sensing modelers and other specialists. The focus of this paper will be on a proposed methodology to connect the forest ecosystem models, which are the desired end product of the Eos process, with the active microwave remote sensing signatures provided by imaging spaceborne synthetic aperture radar (SAR) data. In doing so, it is hoped that a clearer understanding of the rationale behind the development of the theoretical scattering models (see, e.g., Richards et al. 1987; Sun and Simonett 1988; Blanchard this issue; Durden et al. this issue; Ulaby et al. this issue) and the conducting of extensive field measurement programs (see, e.g., Cimino et al. this issue; Dobson et al. this issue) will evolve. Methods to utilize the quantitative signatures within a SAR data set as inputs for forest ecosystem models are at a very early stage of development. This relative immaturity stems from two facts: (1) the paucity of calibrated, multifrequency, polarimetric SAR data of forest test sites for use in technique development; and (2) the lack of validated backscatter models which explain the microwave signature from complex forest canopies. The microwave remote sensing community has been without a source of SAR data since 1985, due to the unfortunate demise of the NASA CV-990 aircraft and the JPL SAR, and the decision of the U.S. Navy to re-fabricate the ERIM/CCRS CV-580 SAR System. However, both NASA/JPL and Navy/ERIM have recently finished reconstruction of their airborne SAR systems (Kozma et al. 1986; Sullivan et al. 1988; Held et al. 1988) and data sets are now becoming available to the

Kasischke and Christensen 2 forestry/remote sensing community. In addition, significant strides have been made over the past several years in the development of theoretical models of microwave backscatter from forested landscapes (Richards et al. 1987; Sun and Simonett 1988; Ulaby et al. this issue). Thus, the necessary instrumentation and theoretical tools exist for development of techniques necessary to utilize SAR data as inputs into forest ecosystem models. In order to accomplish this goal, however, the proper protocol for linking forest ecosystem and microwave backscatter models must be established. 2.0 REMOTE SENSING/FOREST ECOSYSTEM MODEL STRUCTURE Figure 1 presents a conceptual flow chart of the relationships between SAR-derived remote sensing signatures and forest ecosystem models. For a given forest stand, there are several dimensions to the radar signature which are available to generate different channels of data to use as inputs into the connecting models, including: (1) radar frequency; (2) the relative phase between the transmitted and received signature (i.e., the polarization of the signature); (3) the radar incidence and azimuth angles; and (4) time. If the forest stand is larger than the resolution cell size of the SAR system, then there will be spatial variability in the radar signature outside of image speckle (caused by spatial variation in stand geometry and physiology), resulting in another dimension to the radar signature, image texture (Kasischke et al. 1987) The connecting models in Figure 1 represent a set of algorithms which convert the remotely-sensed signatures into parameters used as inputs into the forest ecosystem models. It is important to recognize the connection between the remote sensing signature and the forest ecosystem is being made at the individual tree and forest stand (population) level. Thus, even though higher order forest ecosystem models (e.g., element cycling models, successional models, etc.) may be the desired output, it is through an understanding of the lower order forest ecosystem models that the initial linkages with the remote

Kasischke and Christensen 3 sensing signatures will be made (see section 3.0). The development of connecting models is highly dependent on the microwave backscatter models: these predict the SAR signature as a function of the radar system parameters and the geometry and physiology of the forest canopy being imaged. For a forest ecosystem consisting of a single species of tree, it may be possible to derive an empirically-based model to derive forest parameters from radar imagery (see, e.g., Wu and Sader 1987). However, because of the variations In tree canopy geometry among different tree species, as well as the physiological dynamics of Individual trees, such an empirically-derived connecting model may only be applicable for a very narrow range of tree geometries and environmental conditions. Because of this variability, generalized connecting models can only be developed through the use of a validated set of backscatter models. The proper use of existing tree geometry and tree and population (stand) growth models can provide the information needed for the validation of the backscatter model. This validation process is discussed further In section 4. The formulation of such connecting models will not be accomplished through a straightforward inversion of the backscatter model. Although exercising of the backscatter model for a variety of tree/stand characteristics may provide the basis for the initial format of the connecting model, the final coefficients of the connecting model will most likely be derived from actual radar data, as Is discussed in section 5. 3.0 FOREST ECOSYSTEM MODEL HIERARCHY A great deal of emphasis has been placed on the potential use of SAR data to predict ecosystem properties of interest (NASA, 1983; National Academy of Sciences, 1985). However, the specific protocols by which such data will be successfully linked to ecosystem models are not explicitly enunciated. In the development of these linkages, it is important to recognize that ecosystem models can be classified in a

Kasischke and Christensen 4 hierarchical fashion, ranging from models dealing with the properties of and processes occurring within a single tree to those simulating processes over the entire biosphere (Figure 1). This is by no means a hierarchy of model complexity, rather, these modelling levels differ in the spatial and temporal scales of the inputs and outputs. For programmatic reasons, governmental agencies responsible for the development of satellite remote sensing systems generally discuss ecosystem models at the biosphere level. However, operational linkages between the remote sensing data and the ecosystem models will in most cases be made at the community, population and individual tree levels. In addition to understanding the forest properties that can be directly measured using remotely sensed data, linkages among the various levels in the ecosystem model hierarchy must also be made in order to use remote sensing data to drive the higher-order models presented in Figure 1. For example, leaf area and soil water characteristics are inputs to simulation models of whole plant photosynthesis and evapotranspiration (Running et al. 1986, Running et al. 1988). Such models predict patterns of net photosynthesis and water use over watershed-size areas and short time intervals in relatively simple ecosystems. Population and stand growth models (e.g. Daniels and Burkhart 1975; Mitchell 1969) are driven by empirically or experimentally determined estimates of growth rate and mortality rate, often in relation to stand density. Changes in composition and structure of groups of species populations can be modelled based on assumptions regarding the component populations and the nature of their interactions (Shugart 1984). The complexity of such models varies in terms of the extent to which they simulate competitive interactions, disturbance cycles, and variations among populations in age structure. Most such models can be initialized based on species population data (e.g., density and size) and parameterized from equations of varying complexity that predict tree establishment, growth, and mortality. Changes in forest tree populations are often major determinants of ecosystem-level processes such as carbon and nitrogen cycling. Thus, simulation models of such processes may be directly connected to

Kasischke and Christensen 5 community-level models (e.g. Pastor and Post 1985, 1986, Pastor and Huston 1986). Regional and global scale forest models have been used to predict the impact of forests on very large scale processes such as atmospheric carbon flux (Moore et al. 1981, Houghton et al. 1983). Such a model classification is, of course, arbitrary, but it focuses our attention on four important facts: (1) remotely-sensed data are most directly relevant to forest properties that are inputs to lower-level (i.e., tree, population, and community) models; (2) the forest characteristics which can be directly estimated from remotely-sensed data are often a small subset of those needed for the forest ecosystem models we wish to parameterize; (3) to successfully utilize remotely-sensed data to drive higher-level models, linkages must be established among models that take advantage of the parameters actually provided by the remote sensing systems via the connecting models; and (4) given the complexity of forest ecosystems, no single remote sensing system (e.g., passive optical or active microwave) will likely be adequate to provide all of the inputs the necessary inputs to forest ecosystem models. To illustrate the problems posed here, assume that our goal is to estimate carbon flux from a large forested landscape such as the piedmont and coastal plain of the southeastern United States. Given the importance of natural and human-caused disturbance on such landscapes, variations among forest stands in carbon flux are to a large degree related to variations in successional stage. Trajectories of change in potentially directly-sensible forest features such as green leaf area index (GLAI), stem density, biomass or tree size are often non-linear or even multi-modal (see Figure 2). Thus, successional change cannot be Inferred from such features individually. For example, from Figure 2 it is evident that variation in GLAI early in succession is highly correlated with tree growth and forest carbon flux. The ability of optical remote sensing systems to detect variations in GLAI has been clearly demonstrated. However, GLAI peaks early in succession and subsequently is uncorrelated with carbon flux. Although major changes in forest biomass and carbon flux may occur

Kaslschke and Christensen 6 subsequent to this peak in GLAI, such changes will not be detectable using optical sensors. However, utilizing population-level models, biomass/density data generated from active microwave remote sensing instruments may be used to infer structure in the middle to old age stands. Such structural information, in conjunction with a firm understanding of the dynamics of the tree populations, can be used to drive community-level models of successional change. These may, in turn, provide the inputs for models of patterns of carbon cycling and carbon flux. The succession/carbon flux mode'ls developed for forests to date require some inputs that we will not likely be able to estimate using remote sensing systems. These include site hydrologic characteristics and biochemical characteristics of the forest floor (see Running et al. 1988). In order to. utilize remotely-sensed data to drive carbon forest flux models for an entire landscape, we must be able to infer these characteristics from those data, or supplement our data set utilizing other sources of information (such as geographic information systems). 4.0 BACKSCATTER MODEL DEVELOPMENT To develop models which can explain microwave scattering from forest canopies, it is necessary to determine what canopy parts are responsible for the backscattering of the microwave energy transmitted by a SAR system. The size, shape, distribution, and dielectric properties of the individual tree constituents (i.e., the leaves, reproductive organs, branches, stems and boles) are primary determinants of the radar signature. To a first approximation, those structures whose dimensions are less than a radar wavelength act primarily as attenuators of microwave energy, those structures whose dimensions are larger than a radar wavelength act as scatterers of microwave energy, and the attenuation/scattering of the structure will be directly proportional to the dielectric constant of the plant material. The plant material within a forest canopy can be divided into two parts: plant tissue and plant fluid. From measurements at a frequency

Kasischke and Christensen 7 of 1.2 GHz (Ulaby et al. 1986; Dobson 1988) it is known that: ev = 3.0 - jO ewv = 18.0 - j6 and ef = 72 - J25 where ev is the dielectric constant of the dry vegetation matter, ewv is the dielectric constant of wet vegetation, and ef is the dielectric constant of the plant fluid. Thus, the major source of scattering and attenuation from a forest canopy is the plant fluids which are suspended above the ground by the plant tissues. In other words, the radar signature is a function of the distribution of wet aboveground biomass, and it is the size, shape and distribution of the aboveground biomass elements of a tree, and the dielectric properties of these elements which define the primary inputs for the backscatter model. This fact is illustrated in Figure 3, which presents a schematic diagram of the inputs for the Michigan Microwave Canopy Scattering (MIMICS) model (Ulaby et al. this issue). Validation of a microwave backscatter model can be thought of as a two-step process. First, the model needs to be validated for a stationary condition, i.e., the case where there are no temporal variations (such as diurnal changes in tree fluid content due to moisture stress, seasonal variations in plant biomass, etc.) in the forest canopy. This initial validation step is used to study changes in the microwave signature owing to variation in the overall canopy structure. The second step is validation of the backscatter model for canopy dynamics, i.e., temporal variations in canopy biomass and fluid content. To provide the data for validation of the backscatter models for the stationary case, a set of test sites must be identified which vary in aboveground biomass distribution. To study the dynamic case, several of the test sites have to be monitored over a period of time. If the tree parameters in Figure 3 must be measured individually over an entire growing season, then validation of the backscatter models

Kasischke and Christensen 8 using actual forest ecosystems represents a truly formidable task. However, if test sites are selected where the forest ecosystems have been studied and modeled, then existing tree and stand growth models may provide the input parameters needed for exercising and validation of the backscatter models. In the following sections, we will first discuss a prototype forest ecosystem that can be utilized for studying the connections between backscatter and forest ecosystem models. This will be followed by a discussion of how tree and stand growth models for this prototype forest ecosystem can be used in validation of the backscatter model for the stationary condition. Finally, we will discuss how a different set of tree and stand growth models can be used for validation of the backscatter model for the dynamic scene condition. 4.1 OLD-FIELD LOBLOLLY PINE FORESTS In this paper we will utilize old-field loblolly pine (Pinus taeda L.) forests found in the southeastern United States as a prototype forest ecosystem. Christensen and Kasischke (1987) discuss the important characteristics of the old-field loblolly pine forest ecosystems with respect to microwave remote sensing studies. In summary, loblolly pine stands offer an ideal prototype for use in studies connecting microwave backscatter and forest ecosystem models because: (1) they have been well studied, and numerous tree and stand growth models and forest ecosystem models exist; (2) the pure stands of loblolly pine represent both a simple tree structure and a simple stand structure, which minimizes the complexity in deriving the inputs for the backscatter models; (3) such ecosystems are widespread, offering numerous test sites for use in remote sensing studies; and (4) numerous stands at different stages of the successional sequence exist, providing a wide variety of forest stand densities, tree diameters, and tree heights for evaluation of the backscatter models. Therefore, tree and stand growth models exist for loblolly pine forests, or can be developed based upon existing data or data which can be readily obtained due to the widespread availability of test sites.

Kasischke and Christensen 9 In turn, these test sites can be used to validate the microwave backscatter model for both the static and dynamic scene cases. 4.2 THE STATIONARY CASE Because different loblolly pine stands found over a regional landscape were established at widely varying times, these different stands represent the successional chronosequence for loblolly pine. Although there are potential pitfalls in assuming differences among stands of varying age are a simple consequence of succession (Pickett 1988), abundant long-term studies in these ecosystems provide confidence in such comparisons (Peet and Christensen 1987). Figure 2 presents the general trend of changes in the major tree characteristics which influence the total -aboveground biomass during secondary forest succession for old-field loblolly pine, and hence the radar remote sensing signature. Both axes on this graph represent only relative scales, which vary both as a function of site conditions and initial stocking densities. Since individual stands established at the same point in time have different site conditions, as well as different initial stocking densities, loblolly pine stands found over a large region typically have considerable variation in the amounts and distribution of aboveground biomass (i.e., in the stem density, height, and diameter, and GLAI etc. within the stand). Thus, a radar data set gathered over a number of these stands offers the diversity in stand biomass and geometry required to validate the backscatter models. A general list of forest parameters needed to validate such backscatter models is given in Figure 3, and Table 1 lists all the loblolly pine characteristics which are needed to validate a backscatter model such as MIMICS. We can divide the tree characteristics in Table 1 into three general categories: (1) those characteristics that are easily measured from the ground or that are assumed to be fairly constant throughout the stand; (2) those characteristics measured for a specified number of trees within the stand, which form the basis for equations

Kasischke and Chrtstensen 10 that are applied to the rest of the trees within the stand; and (3) those characteristics estimated using existing tree/stand growth models. For the first category, we assume that the characteristics are being sampled sufficiently to provide a statistical description of the characteristic for the entire stand. The second category of tree characteristics represents those which are more difficult to measure directly from the ground or those for which no tree/stand growth models exist. These parameters will have to be estimated using allometric equations that predict parameter values based on more easily measured variables. For example, tree height can be estimated based on tree diameter and stand density. The third category of variables are those for which tree/stand growth models already exist. These models primarily deal with the amount and vertical distribution of biomass within the tree.crown (Christensen 1988; Kinerson et al. 1974; Hepp and Bristler 1982; Nehmeth 1971; Peet and Council 1982; Clark and Taras 1976; Labyak and Schumacher 1954) and the shape or taper of the tree bole (Burkhart and Walton 1985; Byrne and Reed 1986). Loblolly pines exhibit a high degree of self-pruning, with the living branches being limited to the upper portion of the canopy, and the dead branches below this region falling off the tree. As illustrated in Figure 4, the canopy depth for loblolly pine stands follows a predictable trajectory based upon initial stocking density and stand age. The lines in Figure 4 represent different stands with different initial stocking densities. Each stand started out with a high stem density, which decreases with age as stem mortality occurs. As density decreases, canopy depth decreases in a predictable manner. Thus, a model can be developed which predicts crown depth based upon stem density. If the tree height (h) and crown depth (dc) are known, then it is possible to estimate the crown ratio (Cr) as Cr = [(h - dc) / hi x 100. (1) Models exist (Kinerson et al. 1974; Hepp and Bristler 1982; Nehmeth 1971; Peet and Council 1982) which predict total crown weight, total

Kasischke and Christensen 11 branch weight and total needle weight as a function of tree diameter and crown ratio. Figure 5 (after Hepp and Bristler 1982) shows the relationship of crown weight, branch weight and needle weight as a function of crown ratio for a fixed tree diameter. Figure 6 Illustrates the trend in the proportion of branch weight and needle weight as a function of tree diameter. It is also Important to determine the distribution of the total biomass In the crown. Although some of this information will be provided via direct measurements, some can also be provided from the growth models (Kinerson et al. 1974; Hepp and Bristler 1982), as illustrated in Figure 7. In summary, models exist or can be readily developed for loblolly pine stands to estimate difficult to measure parameters based upon other characteristics which can be easily obtained. Once a set of these models has been derived, it should be possible to generate an input data set for a large number of different stands for validation of the backscatter model. 4.3 THE DYNAMIC CASE If the backscatter models are going to be useful for predicting signatures for other geographic regions at different times of the year, then changes in the loblolly pine stand characteristics due to regional (i.e., site) or temporal events must be accounted for. There are subtle physiological variations in tree structure due to genetic responses to climate variations. For example, Zobel and Rhodes (1956) illustrated that the specific gravity for loblolly pine bole wood ranged from 0.358 to 0.486, and the specific gravity for branch wood ranged from 0.328 to 0.692. Thames (1963) showed clear differences in needle structure and composition between pines located in drought-prone regions and those from non-drought areas. These variations need to be expressed in terms of the characteristics used as inputs into the backscatter models in order to determine their effects on the radar signature from the loblolly pine stands.

Kasischke and Christensen 12 There are four types of temporal, physiological changes within a loblolly pine forest that may influence the radar signature: (1) diurnal or seasonal variations in the relative water content of the vegetation; (2) variations in total needle biomass; (3) growth of new stems; and (4) growth of reproductive organs. Development of a generalized connecting model requires a backscatter model which can predict change in the microwave signature due to these temporal variations. Recent experimental work by Dobson et al. (this issue) has shown there can be substantial diurnal variation in the radar intensity signature of a single tree stand (3 to 5 dB). The most likely explanation for these observed changes was diurnal variation in the state of the water in the leaves, branches, and boles of the trees in response to moisture stress. Loblolly pine trees are known to have similar diurnal variations in plant moisture (Figure 8, Hodges and Lorio 1971). It may be possible to develop a generalized model which predicts diurnal changes in both relative water content of the needles and water potential within the trunk for loblolly pines based upon meteorological data (i.e., rainfall and temperature). Such models have been developed for other pine species (Running 1984, Running et al. 1988). Several models have been developed to predict seasonal variations in total needle biomass (Kinerson et al. 1974; Clark and Taras 1976; Labyack and Schumacher 1954). Figure 9 (after Kinerson et al. 1974) depicts seasonal variation in total needle biomass over an entire growing season, while Figure 10 (also after Kinerson et al. 1984) presents the vertical distribution of this biomass as a function of time based on such a model. As can be seen from Figure 9, there is almost two-fold variation in total needle biomass over a single growing season. New stem growth or branch flushing should have a significant impact on the radar signature. For loblolly pine trees less than 20 years old, new stem height growth averages between 0.5 and 1.2 meters per year, depending on site conditions. For loblolly pine trees between 20 and 50 years old, new stem growth averages between 0.15 to 0.45 meters per year (Gaiser 1950). This growth occurs in 2 to 5 flushes (i.e., growth

Kasischke and Christensen 13 spurts) per growing season, depending on site conditions and climatic conditions. The final temporally-varying growth event which may influence the radar signature from a loblolly pine forest is the growth of the pine cones on the trees. Mature loblolly pine cones are distributed on the end of the pine branches and are up to 9 cm in length. However, these cones take three years for development and maturation, and the number of cones reaching maturity during the third year is highly variable (Dewers and Moehring 1971; Wenger 1957; Grano 1957). To develop a backscatter model which can adequately account for these temporal variations, it will be necessary to conduct experiments in which radar data are collected over time periods sufficiently long to encompass these events at a sufficient frequency to track the trajectory of change. For example, diurnal measurements will be necessary to monitor moisture stress, whereas measurements will need to be made over entire growing seasons to determine the Impact of changes in new stem, cone, or needle biomass on SAR signatures. Once these data have been collected and used to validate the backscatter models, then the models can be exercised over the expected range of conditions (as predicted by the loblolly pine growth models) to determine the Influence of such temporal variations on radar signatures. The influence of such temporal changes can then be factored into the connecting models developed for the loblolly pine forest ecosystems. 5.0 DEVELOPMENT OF CONNECTING MODELS After their validation, a sensitivity analysis can be performed using the backscatter models to determine which forest stand characteristics significantly influence the radar signatures. The results of one such partial sensitivity analysis using the MIMICS model are presented in Figures 11 and 12. These were generated using the characteristics of pure stands of bigtooth aspen (Populus grandidentata) located in the northern portion of the lower peninsula of Michigan (average stand characteristics: tree height: 8 m; density; 1110

Kasischke and ChrIstensen 14 stems/acre; dbh: 24 cm; trunk dielectric constant: 15 - j5; leaf area index: 4.7). Developing a set of algorithms to generate the required inputs for the ecosystem models from the remote sensing signature requires careful consideration of the results of the sensitivity analyses performed using the backscatter model as well as the requirements of ecosystem models. The parametric exercising of the backscatter models should be conducted to consider those forest characteristics which are potential inputs to the ecosystem models of interest. If sensitivity analyses indicate that a variation in a required ecosystem model input results in a discernable trend within the radar signature, then an algorithm to extract that forest characteristic from the radar data can be devised. The modeling results indicate that VV-polarized channels of the radar are more sensitive to changes in the tree characteristics (Figures 11 and 12). In the absence of any corresponding SAR or radar data to compare with these model results, we cannot make any definitive statement as to the validity of the model. However, if we assume the MIMICS model has been validated, then the results from the VV channel of data can be used in the development of a connecting model between the radar observations and a forest ecosystem model. In order to explore this connecting model development, let us consider the problem of estimating total aboveground biomass for bigtooth aspen example (see data in Figures 11 and 12). Total aboveground biomass for an individual tree, Bm, can be expressed as Bm = Bt + Bb + B1 (2) where Bt is the biomass of the tree trunk, Bb is the biomass of the tree branches, and B1 is the biomass of the leaves. The biomass of an individual tree may be related to the biomass of a stand of trees from which a signature is derived. To calculate trunk biomass for an entire stand of trees it may be necessary to derive a biomass term such that Bt = f(density, diameter, height) (3)

Kaslschke and Christensen 15 and develop a separate equation for each of the terms in Eq. (3). The sensitivity analysis using the backscatter model allows determination of which radar channels are most sensitive to each of the parameters In Eqs. (2) and (3). As is illustrated in Figure 12, it appears that variations In the bole density, diameter and height characteristics are best detected in the L-VV and C-VV channels, whereas the X-VV channel is most sensitive to changes in the leaf biomass (which is directly correlated to leaf area index). Based on the shape of the curve (i.e., linear, exponential, quadratic, etc.), an expression which relates the biomass signature of interest to the specific radar channels may then be derived. Using Figure 12 as an example, a relationship between (L-VV) and tree height, h, might be constructed as h = a + (L-VV)b (4) where the terms a and b are determined via least squares regression techniques. In the example in Figure 11, a = -5 and b = 0.9. However, it should be recognized that the total L-VV signature will vary as a function of other forest characteristics, such as leaf area index and tree density. Therefore, in order to account for these variations, additional terms will have to be added to Eq. (4), and the final version of Eq. (4) may be of the form h = a + (L-VV)b + (C-VV)c + (X-VV)d (5) where the C-VV and X-VV channels were selected because they were sensitive to changes in leaf area index and stem density, respectively, and c and d are coefficients determined by least-squares regression techniques. Upon completion of the derivation of the initial connecting model based upon the results of the sensitivity analysis, actual radar data from the test sites used to validate the backscatter model can be used

Kasischke and Christensen 16 to refine the coefficients of the connecting model. Finally, the precision of the connecting model can be evaluated using additional test sites from the study area. The generality of such models can be measured using test sites at outside of the original area. The end product will be a family of models which utilize specific remotely-sensed signatures to predict forest features relevant to ecological models (or vice versa). The modeling process described above is sufficient to define the stationary forest characteristics/properties of interest over the landscape of interest. For instance, we may be able to estimate aboveground biomass (standing crop), stand density or GLAI for bigtooth aspen forests. However, if the goal is to measure or predict forest processes such as primary production, carbon flux, evapotranspiration, or nitrogen. cycling, additional modeling steps are required, as discussed in section 3. For example, a succession model such as FORET (Shugart 1984) might be used to estimate biomass change from tree size and density data. Many of these process models were not explicitly developed to be driven by the forest data directly provided by remotely sensors. Some models require Inputs that are not likely to be obtained by remote sensing. For example, most forest hydrologic models require input of Information on soil characteristics. Thus, such models may require extensive modification to make effective use of the data remote sensors provide. 6.0 CONCLUSIONS We have outlined a procedure to develop a set of algorithms to relate microwave remote sensing signatures to forest ecosystem models. This process is dependent on a validated remote sensing model which describes microwave scattering from a forest canopy. The validation of the backscatter model itself can be greatly facilitated through the use of forest tree/stand growth models to provide inputs for the backscatter model. In summary, the following steps are followed in the overall

Kasischke and Christensen 17 development of a forest ecosystem model which utilizes remote sensing-derived parameters as inputs: 1. Utilizing tree/stand growth models and actual ground-based radar and SAR measurements, develop and validate microwave backscatter models which predict the radar signature as a function of radar system parameters and tree stand characteristics. 2. Parametrically exercise the validated backscatter models to determine which desired forest stand characteristics are most influential on the radar signature. 3. Based on the above sensitivity analysis, identify which forest ecosystem input parameters can potentially be provided from the SAR data set. 4. Develop the, connecting models to provide the inputs for the forest ecosystem models. 5. Reconfigure the forest ecosystem models so that they utilize the remote sensing estimates from the connecting models as their primary inputs and identify the sources for the remaining inputs. 7._0 ACKNOWLEDGEMENTS The methodology presented in this paper was developed under funding provided by the National Aeronautics and Space Administration (NASA) under Grants No. NHEW-1339 to Duke University and Grant Nos.NASW-4360 and NAGW-1101 (under subcontract to the University of Michigan) to the Environmental Research Institute of Michigan. The authors would like to thank M. Craig Dobson of the University of Michigan for providing outputs of the MIMICS model for inclusion in this paper. Many of the ideas presented in this paper were the direct result of concepts which evolved during numerous round-table discussions with a group of scientists loosely organized under the banner of the International Forest Investigation Team or IFIT. The authors would like to acknowledge the contribution of these discussions to this paper. Special thanks go to JoBea Cimino of the Jet Propulsion Laboratory,

Kasischke and Christensen 18 8.0 REFERENCES Blanchard, A., 1989, Identification of Forest Information In Radar Signatures Identified Through Measurements, Int. J. Remote Sensing (this issue). Burkhart, H.E. and S.B. Walton, 1985, Incorporating Crown Ratio into Taper Equations for Loblolly Pine Trees, Forest Sci., 31, 478484. Byrne, J.C. and D.D. Reed, 1986, Complex Compatible Taper and Volume Estimation Systems for Red and Loblolly Pines, Forest Sci., 32, 423-443. Christensen, N.L., 1988, Analysis of Duke Forest Permanent Plots: Patterns of Production and Mortality of Seedlings and Trees, in Forest Response Program Annual Meeting - Project Status Reports, Vol. 1, NCSU Atmospheric Research Program, Raleigh, NC, 85-89. Christensen, N.L., Jr. and E.S. Kasischke, 1987, The Use of Active Microwave Remote Sensors in Forest Ecology Studies, Proc. 1987 Inter. Geosci. Remote Sens. Sy)p., Ann Arbor, MI, 457-466. Cimino, J.B., J. Paris, D. Casey, F. Ahern, N. Christensen, M.C. Dobson, D. Gates, F. Ulaby, J. Weber, R. Hoffer, M. Imhoff, E. Kasischke, A. Milne, J. Richards, A. Sieber, P. Churchill, D. Simonett, C. Slaughter and L. Viereck, 1989, The Effect of Changing Environmental Conditions on Microwave Signatures of Forest Ecosystems, Int. J. Remote Sensing (this issue). Clark, A. III, and M.A. Taras, 1976, Comparison of Aboveground Biomass of Four Major Southern Pines, Forest Products Journal, 26, 2529. Daniels, R.F. and H.E. Burkhart, 1975, Simulation of Individual Tree Growth and Stand Development in Managed Loblolly Pine Plantations, Virginia Polytechnic Institute, Blacksburg, VA. Dewers, R.S. and D.M. Moehring, 1971, Effect of Soil Water Stress on Initiation of Ovulate Primordia In Loblolly Pine, Forest Sci., 16, 219-221. Dobson, M.C., Diurnal and Seasonal Variation in the Microwave Dielectric Constant of Selected Trees, 1988, Proc. Int. Geoscience and Remote Sens. Syqp., Edinburgh, Scotland, 1754. Dobson, M.C., J.F. Paris, F.W. Ulaby and K. McDonald, Status of Radar Scatterometer Measurements Over Forested Areas, Int. J. Remote Sensing (this issue).

Kasischke and Christensen 19 Durden, S.L., J.J. van Zyl, and H.H. Zebker, 1989, Polarimetric Radar Forest Signatures, Int. J. Remote Sensing (this issue). Grano, C.X., 1957, Indices to Potential Cone Production of Loblolly Pine, Jour. Forestry, 55, 890-891. Held, D.N., W.E. Brown, and T.W. Miller, 1988, Preliminary Results from the NASA/JPL Multifrequency, Multipolarization SAR, Proc. 1988 IEEE Radar Conf., 7-8. Hepp, T.E. and G.H. Bristler, 1982, Estimating Crown Biomass in Loblolly Pine Plantations in the Carolina Flatwoods, Forest Sc1., 28, 115-127. Hodges, J.D. and P.R. Lorio, Jr., 1971, Comparison of Field Techniques for Measuring Moisture Stress in Large Loblolly Pines, Forest Sci., 17, 220-223. Houghton, R.A., J.E. Hobble, J.M. Mell1lo, B. Moore, III, B.J. Peterson, G.R. Shaver and G.M. Woodwell, 1983, Changes In the Carbon Content of Terrestrial Biota and Soils Between 1860 and 1980: A Net Release of C02 to the Atmosphere, Ecol. Monogr., 53, 235-262. Kasischke, E.S., A.L. Maffett and R.W. Larson, 1987, Statistical Modeling of Speckle Distributions on Airborne SAR Imagery, Proc. 1987 Inter. Geosci. Remote Sens. Symp., Ann Arbor, MI, 1357-1362. Kinerson, R.S., K.0O. Higginbotham and R.C. Chapman, 1974, The Dynamics of Foliage Distribution within a Forest Canopy, J. Appl. Ecol., 11, 347-353. Kinerson, R.S., C.W. Ralston, and C.G. Wells, 1977, Carbon Cycling in a Loblolly Pine Plantation. Oecologia 29, 1-10. Kozma, A., A.D. Nichols, R.F. Rawson, S.J. Shackman, C.W. Haney and J.J. Shanne, Jr., 1986, Multifrequency, Multipolarization SAR for Remote Sensing, Proc. IGARSS'86 Symposium, Zurich, Switzerland, Ref. ESA SP-254, 715-719. Labyak, L.F. and F.X. Schumacher, 1954, The Contribution of Its Branches to the Main-Stem Growth of Loblolly Pines, J. Forestry, 52, 333-337. Mitchell, K.J., 1969, Simulation of Growth of Even-Aged Stands of White Spruce, Yale School of Forestry Bull 75, 48 p.

Kasischke and Christensen 20 Moore, B. R.D. Boone, J.E. Hobble, R.A. Houghton, J.M. Melillo, B.J. Peterson, G.R. Shaver, C.J. Vorosmarty, and G.M. Woodwell, 1981, A Simple Model for Analysis of the Role of Terrestrial Ecosystems in the Global Carbon Budget, in B. Bolin (ed.), Carbon Cycle Modeling, Scope 16, John Wiley and Sons, New York. NASA, 1983, Strategy for Earth Science Issues, Land-Related Global Habitability Sciences Working Group, NASA Technical Memorandum 85841, Washington, D.C., 112 p. National Academy of Sciences, 1985, A Strategy for Earth Science from Space in the 1980'2 and 1990's - Part II: Atmosphere and interactions with Solid Earth, Oceans and Biota, National Academy Press, Washington, D.C., 194 p. Nehmeth, J.C., 1971, Dry-Matter Production in Young Loblolly (Pinus Taeda L. and Slash (Pinus Elliottii Englm.) Plantations, Ph.D. Dissertation, North Carolina State University, 75 p. Pastor, J. and M, Huston, 1986, Predicting Ecosystem Properties from Physical Data: A Case Study of Nested Moisture-Climatic Gradients Along the Appalachian Chain, in Coupling of Ecological Studies with Remote Sensing; Potentials at Four Biosphere Reserves in the United States, ed. by M.I. Myer and D.A. Crossley, Jr., 82-95, U.S. Department of State Publication 9504. Pastor, J. and W.M. Post, 1985, Development of a Linked Forest Productivity-Soil Process Model, ORNL/TM-9519, Oak Ridge National Laboratory, Oak Ridge, TN. Pastor, J. and W.M. Post, 1986, Influence of Climate, Soil Moisture, and Succession on Forest Carbon and Nitrogen Cycles, Blogeochem., 2, 3-27. Peet, R.K. and N.C. Christensen, 1987, Competition and Tree Death, Bioscience, 37, 586-595. Peet, R.K. and O.P. Council, 1982, Rates of Biomass Accumulation in North Carolina Piedmont Forests, unpublished report to the North Carolina Energy Institute. Pickett, S.T.A., 1988, Space-for-Time Substitution as an Alternative to Long-Term Studies, Proc. Cary Conference II (in press). Richards, J.A., G-Q. Sun and D.S. Simonett, 1987, L-band Radar Backscatter Modeling of Forest Stands, IEEE Trans. Geosci. Remote Sens., 25, 487-498.

Kasischke and Christensen 21 Running, S.W., 1984, Documentation and Preliminary Validation of H20TRANS and DAYTRANS, Two Models for Predicting Transpiration and Water Stress in Western Coniferous Forests, U.S.D.A. For. Serv. Res. Pap. RM-252. Running, S.W., D.L. Peterson, M.A. Spanner, and K.B. Teuber, 1986, Remote Sensing of Coniferous Forest Leaf Area, Ecol., 67, 273276. Running, S.W., R.R. Nemani, D.L. Peterson, L.E. Band, D.F. Potts, L.L. Pierce and M.A. Spanner, 1988, Mapping Regional Forest Evapotranspiration and Photosynthesis by Coupling Satellite Data with Ecosystem Simulation, Ecology (in press). Shugart. H.H., 1984, A Theory of Forest Dynamics: The Ecological Implications Forest Succession Models, Springer-Verlag, New York, 278 p. Sullivan R., A. Nichols, R. Rawson, C. Haney, F. Darreff, and J. Schanne, Jr., Polarimetric X/L/C-band SAR, 1988, Proc. 1988 IEEE Radar Conf.,*9-14. Sun, G-Q. and D.S. Simonett, 1988, Simulation of L-band HH Microwave Backscattering from Coniferous Forest Stands: A Comparison with SIR-B Data, Photogrammetric Eng. and Remote Sensing, 54, 1195-1201. Thames, J.L., 1963, Needle Variation in Loblolly Pine from Four Different Geographic Seed Sources, Ecology, 44, 168-169. Ulaby, F.T., R.K. Moore and A.K. Fung, 1986, Microwave Remote Sensing, Active and Passive-Volume III: From Theory to Application, Artech House, Inc., Dedham, MA, 1097 pp. Ulaby, F.T, K. Sarabandi, K. McDonald, M. Whitt, and M.C. Dobson, Modeling Microwave/Tree Interactions Paper, Int. J. Remote Sensing, this issue. Wenger, K.F., 1957, Annual Variation in the Seed Crops of Loblolly Pine, Jour. Forestry, 55, 567-569. Wu, S-T. and S.A. Sader, Multipolarization SAR Data for Surface Feature Delineation and Forest Vegetation Characterization, IEEE Trans. Geosci. Remote Sens., GE-25, 67-76, 1987. Zobel, B.J. and R.R. Rhodes, 1957, Specific Gravity Indices for Use in Breeding Loblolly Pine Trees, Forest Scl., 3, 281-285.

Kasischke and Christensen 22 TABLE 1 LOBLOLLY PINE CHARACTERISTICS REQUIRED FOR INPUTS INTO MICROWAVE BACKSCATTER MODEL* TREE CHARACTERISTIC SOURCE Average tree diameter (dbh) Measured directly Average tree density/acre (dt) measured directly Average tree height (h) Measured directly and allometric equations Dielectric constants of bole, branches and needles measured directly Litter depth and surface roughness measured directly Litter dielectric properties measured directly Average height to first branch whorl (hb) Christensen 1988. Measured directly on selected sites, allometric equations Crown ratio (C) Allometric equations, or hb/h Main stem taper ratio (tr) Burkhart and Walton 1985 Byrne and Reed 1986 Total crown weight (TWT) Kinerson et al. 1974 Hepp and Bristler 1982 Total branch weight (BWT) Hepp and Bristler 1982 Nehmeth 1971 Peet and Council 1982 Total needle weight (NWT) Kinerson et al. 1974 Clark and Taras 1976 Labyak and Schumacher 1954 relationships further established with direct measurements *Variables which will be measured directly are indicated. References indicate sources of allometric equations to predict a given variable.

Kasischke and Christensen 23 TABLE 1 LOBLOLLY PINE CHARACTERISTICS REQUIRED FOR INPUTS INTO MICROWAVE BACKSCATTER MODEL (Concluded)* TREE CHARACTERISTIC SOURCE Average aboveground tree biomass (Bm) Nehmeth 1971 Peet and Council 1982 Clark and Taras 1976 Total number of branch whorls (BW) relationships established with direct measurements Stems/branch whorl (Sbi) relationships established with direct measurements Average horizontal branch angle/whorl (1i) relationships established with direct measurements Average distance between pranch whorls (di) relationships established with direct measurements *Variables which will be measured directly are indicated. References indicate sources of allometric equations to predict a given variable.

Kasischke and Christensen 24 LIST OF FIGURES Figure 1. Schematic diagram of the relationship between remote sensing data, radar backscatter models, and forest ecosystem models (see text for an explanation) Figure 2. Typical trends in important forest characteristics during secondary forest succession on the southeastern Piedmont. (Values are expressed as a percent of their maximum value during the chronosequence. The range of maximum values for GLAI = 6-8 (Nehmeth 1972), pine density = 1000-30,000 stems per hectare, hardwood density = 5000-6000 stems per hectare, aboveground pine biomass = 20-40 kg per m2, and aboveground hardwood biomass = 18-30 kg per m2 [Peet and Christensen 1987]. The actual maximum values, as well as the actual rates of change, for any particular stand vary as a consequence of local site conditions, historical factors, and management interventions.) Figure 3. Diagram of the forest/tree inputs required for the MIMICS backscatter model. Figure 4. The change in canopy depth (distance from the top of the crown to the lowest leafy branch) as a function of stem density. (Each line represents data from a single stand from 1933 to 1988. All stands were 8 years of age in 1933 and were allowed to thin naturally. After 55 years, the stands converge to a canopy depth of 5-6 m, however the trajectory of change varies with initial stand density or stocking.) Figure 5. Changes in tree crown component weights as a function of crown ratio (dbh = 7 inches; after Hepp and Bristler 1982). Figure 6. Changes in the proportion of branch (branch ration) and needle (needle ratio) comprising total canopy biomass with increasing tree diameter (after Hepp and Bristler 1982). Figure 7. Predicted height distribution of canopy biomass components for a 15 year old loblolly pine tree using a model developed by Hepp and Bristler (1982). Figure 8. Measured diurnal variations in trunk water potential and needle relative water content (after Hodges and Lorio 1971). Figure 9. Seasonal variation in loblolly pine needle biomass distribution as predicted by a model developed by Kinerson et al. 1974. Figure 10. Vertical distribution of old and new foliar biomass at four times beginning in the first week of the growing season (after Kinerson et al. 1974).

Kasischke and Christensen 25 Figure 11. Total HH-polarized canopy backscatter estimated by MIMICS for a bigtooth aspen stand at X, C and L-bands as a function of tree height, diameter, stem density and leaf area index (LAI). Figure 12. Total VV-polarized canopy backscatter estimated by MIMICS for a bigtooth aspen stand at X, C and L-bands as a function of tree height, diameter, stem density and leaf area index.

88-21 826R 1 Forest Ecosystem Biosphere Ecosystem Community Population ~Remote Connecting Sensing ModelsIndividua Data Tree Tree Microwave Physiology Backscatter Models -E

8&81172 ESTABLISHMENT PHASE THINNING PHASE j-4- TRANSITION PHASE-..... —-.-.-. STEADY STATE l~X1 --''''/'.~ Bt 50 ff!_; PINE DENSITY HARDWOOD BIOMASS Z. HARDWOOD DENSITY P 2S. ~o T0 50O T1 200X i-8 ~~~~~~\ ~APPROXIMATE AGE

1 DC,, I:'_ c -'- C -H Region I H. I l Region 11 1t b Region III GEOMETRICAL PARAMETERS I. Crown Region: HI, foliage height Dc, foliage diameter Branches and Needles: fc (l,d,e,e); cylinder PDF I = cyl length, d. cyl diameter, (9e,). cyl orientation Leaves: ~(t,d,9,e); disc PDF t = disc thickness, d a disc diameter, (e,e) = disc orientaton II. Trunk Region: Ht,trunk height )t, trunk diameter Rt, trunk surface roughness Lt, spacing between trunks 111. Ground Region: S, surface rms height Is, surface correlation length DIELECTRIC PARAMETERS e of leaves e of branches e of trunks e of ground 22 _ o -

8 6~ E 0 z 1,000 10,000 DENSITY (stemshaf') f~l~A" Yf

88-11770 80 70 60 ~~~~~~60 ~~~~Total Weight E 50 40 | / Branch Weight 30 20 10 20 35 50 65 80 Crown Ratio %

88-1171 0.8 Branch 0.6 O rQ 0 0.4 0.2.. 4 6 8 10 12 14 DBH (in.)

1.4 Total Weight 1.2 Branch Weight.CD 0.8 Needle Weight 0.4 0.2 24 28 32 36 40 44 Height Above Ground (ft) F~hA^'7

88-11074 Trunk Water Potential 0 -4;ii -8 E -12 0) -16 -20 -24 0 12 24 Hours Needle Relative Water Content (RWC) 90 a. 00 o 80 70 -I 0 12 24 Hours

Needle Biomass 88-1176 1000 Total New E cm U, E.9 Old M J J A S O N D J F M A F1" f 5

16 Week 0 Week 10 Week 20 Week 30 12 O O N N N (a) (b) (c) (d) 0 I I I.. 0 1.5 0 1.5 0 1.5 0 1.5 Foliage Density (m -2 m 3) c

TOTAL CANOPY BACKSCATTER HH - Polarization 10.0 5.0 5.0 50.0 L _ _ 0 0 0.0 *0* i * * * ffi & * * | * * * 10.0 20.0 30.0 40.0 SO.O -- L BAND -o- C BAND 1 5-0 -0- X BAND -10.0-~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ( 5.0 -I~~~~~0. ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~-0 --— C BAND 15.0 X BAND 0.0 0.0.5.0 - L BAND -o- C BAND D Cl)5o -,o0AN -- X BAND co r 0 0U 0 CANOPY DENSITY (Trees per Square Meter)N j~ ~~~~~BAJ

TOTAL CANOPY BACKSCATTER VV - Polarization 10.0 I'''I... " 5.0 s10~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~0.0.0 0~c ~ I 0.000 0 ~~~~~~~~~~~~~~~~ oP o o 0 1 P U UJ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~( -- l BAND r 4 -o- C BAND -BAND -- X BAND C') (.,1 -o. BAND U) UZ 0.-a X BAND O C.) 00 0 4 -15.0 10.0 20.0 30A 40A 50.0 41I.0 TRUNK DIAMETER (cm) 4.00 6.00 8.o00 10.00 12.00 14.00 16.00 TRUNK HEIGHT (Meters). 50~~~~~~~~~~~~~~~~~~~~~~~~~~~~0 - L BAND o- C BAND A0-X'BAND oa o.0 oV 4.0 4 -- C BAND.' - X BAND U) -6. m~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~I.10A ~ I -101C -1OA -15... -' — 6.0 7~~~~~-'.C. -15.Q ON 0.~ -rS Q05 0.10 0.15 0.20 0.25 2A0 40 5.0 60 7.0 CANOPY DENSITY (Trees per Square Meter) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~LAARAIDE

6.0 George Washington University Progress Report A progress report has not been forthcoming from George Washington University. 3 9015 028269