Structures of Error Covariance in Global Navigation Satellite System Reflectometry

Abstract

Observations represent the largest cost driver in the environmental prediction enterprise. This dissertation aims to enhance the return on investment for satellite observatories by maximizing both the quantity and quality of data collected. Taking NASA’s Cyclone Global Navigation Satellite System (CYGNSS) as a case study, this work proposes several novel contributions to extract the most value in observation data.

CYGNSS uses a technique known as Global Navigation Satellite System – Reflectometry (GNSS-R), which is an opportunistic bistatic radar measurement of the Earth’s surface using GNSS as a signal-of-opportunity. This technique enables several surface sensing products for both land and ocean parameters. This dissertation primarily focuses on CYGNSS’s ocean surface windspeed measurement, which drove the design of the mission architecture and performance objectives.

This thesis is structured into four distinct, but related lines of effort. The first is an evaluation of the CYGNSS onboard calibration system that substantially increased the amount of usable science data. Prelaunch design decisions and conservative estimates of the thermal loading led to a suboptimal calibration sequence that significantly impacted science duty cycle. This work proposed a longer calibration cadence with minimal data quality degradation, increasing the science duty cycle from 90% to 98%.

The second line of effort produces a mechanism to price the cost of representativity error in satellite observations that are not exactly simultaneous and collocated. If two measurements are nearby in space and time, but not exactly simultaneous or collocated, it begs the question whether the two measurements are representative of the same target. Applying a technique from optimal interpolation, this work produces a simple metric that can be employed by satellite observation planners for future missions, which may feature proliferated constellations of disaggregated sensors.

The third line of effort builds and validates a physical model of correlated error structure in GNSS-R. In addition to building up a “bottom-up” error inventory for GNSS-R, this construction is also generalizable to other observation types. This is especially useful for numerical weather prediction, as data assimilation schemes frequently discard large amounts of observations because the correlated error structure is not well-defined. This is particularly problematic for GNSS-R due to its unique sampling characteristics.

The fourth and final line of effort builds upon the instrument correlated error model and evaluates the CYGNSS windspeed observation error covariance in observation space. This analysis also produced a method to partition certain sources of representation-retrieval error and suggests new corrections to the CYGNSS windspeed product that improves retrieval RMSE by 11%.

This work focuses on not just enhancements to CYGNSS, but also analysis to explore the structure of certain sources of error or impairment. Taken together, this dissertation both increases the quantity of useful CYGNSS data and makes CYGNSS data more useful.