Mapping Diseased Oak Trees Using ADAR Imagery

Abstract

We investigated the ability of high spatial‐resolution 4‐band imagery (Airborne Digital Acquisition and Registration ‐ ADAR) to discern moisture stress in trees affected by Sudden Oak Death (SOD). We wanted to test if the imagery could be used to distinguish between green oak trees with advanced SOD trunk symptoms, and green oaks with no SOD trunk symptoms. ADAR imagery of China Camp State Park in Marin County, California was flown in spring 2000 and 2001. Training samples from the field consisting of the locations green healthy oaks and green symptomatic oaks were used to derive spectral signatures for the two classes. Both hierarchical unsupervised classification (HUC) and maximum likelihood classification (MLC) were used to classify the imagery. Accuracy assessment and other spectral measurements were performed to analyze the separability of the two signatures. Poor overall accuracy 55.17% was obtained by the HUC method. A better overall accuracy 74.19% was obtained by MLC method, but the low transformed divergence (1448) indicated poor separability of the training samples. The poor accuracy results can be explained by the fact that ADAR image has relatively broad spectral bands that combine narrow moisture‐ stress‐sensitive regions with broader stress‐insensitive regions; such combination could decrease the capability of ADAR to detect moisture stress. In addition, healthy oaks in the area display a marked variability in canopy condition, making it difficult to separate healthy trees from those experiencing some stress. In conclusion, this research indicated the inability to automate mapping of moisture stress in oaks using ADAR imagery, and limited success in using methods that require extensive field data.