Rapid Identification of Beech Bark-Diseased Trees Using High Resolution NAIP Imagery

Abstract

Non-native diseases and insects can have a significant impact on forest health. Locating outbreaks and patterns of spread is important in order to mitigate spread (where possible) or plan for changes in forest-species composition. Beech Bark Disease (BBD) is a two-step disease involving a beech scale insect, Cryptococcus fagisuga, and a fungi of the genus Nectria. BBD is actively affecting northeastern US forests, including those of northern Lower Michigan, the location of this study. Remote sensing technologies have potential advantages of being able to monitor for forest health events over broad landscapes and to track change over time. The goal of my study was to use publicly available imagery and open-source software to develop a remote sensing-based BBD mapping approach that can be efficiently replicated by other land managers at the landscape scale. My study landscape was the upland area of the ~4200-ha University of Michigan Biological Station (UMBS) in northern Lower Michigan. I used the National Agriculture Imagery Program (NAIP) imagery plus field data and performed my analyses in the R software environment. My specific objectives were to: 1) develop field data characterizing BBD infestation over the study landscape; 2) collect training and testing data of BBD-affected tree crowns plus those of senescing aspen trees for use in the remote sensing classification, 3) evaluate remotely sensed characteristics of BBD-affected image pixels and assess their spectral separability from those of healthy beech trees and senescing aspen, 4) use information derived from the above objectives, along with multi-year NAIP imagery, to map BBD-affected tree crowns and track BBD outbreaks over several years. Results from the field data showed that BBD is widespread on the study landscape in all regions where beech has been a significant component (at least 57% of study landscape). Visual identification of specific BBD infected canopies had very high accuracy (94%) and the automated classifier had an accuracy of 82%. Spectral analyses showed that diseased beech canopies are mostly unique in their spectral signatures when compared to both healthy beech canopies and senescing aspen canopies with minor overlap. Use of NAIP imagery facilitated replicating the classification process on recent historical imagery (every other year for 8 years) to observe the pattern and progression of the disease. Overall this study demonstrates that showed that BBD-infected American beech canopies can effectively be identified using openly available imagery and software.