A Comparison of Standard and Hybrid Classifier Methods for Mapping Hardwood Mortality in Areas Affected by “Sudden Oak Death”

The sudden oak death (SOD) epidemic in California has resulted in hundreds of thousands of dead trees in the complex of oak (Quercus) and tanoak (Lithocarpus) woodland that exist in patches along the California coast. Monitoring SOD occurrence and spread is an on-going necessity in the state. Remote sensing methods have proved to be successful in mapping and monitoring forest health and distribution when a sufficiently small ground resolution is used. Supervised, unsupervised, and “hybrid” classification methods were evaluated for their accuracy in discriminating dead and dying tree crowns from bare areas and the surrounding forest mosaic utilizing 1-m ADAR imagery covering both tanoak/redwood forest and mixed hardwood stands. In both study areas the hybrid classifier significantly outperformed the other methods, producing low omission and commission errors among information classes. The hybrid method was then further refined by varying three parameters of the algorithm (iteration number, homogeneity threshold, and number of classes) and accuracy was assessed. The results demonstrate that while the hybrid method outperformed the other classifiers, the parameters that yielded highest accuracy for the algorithm differed between the two study areas. The use of a randomly selected subsample of training pixels was compared to the use of polygonal training areas, and we found that polygonal training data provided better classification accuracies in both cases.