An evaluation of AVHRR NDVI data for monitoring western spruce budworm defoliation [electronic resource] /

by Majeed, Zainal A.

An Evaluation of AVHRR NDVI Data for Monitoring Western Spruce Budworm Defoliation By Zainal A Majeed This thesis is an evaluation of the possibility of using NOAA Advanced Very High Resolution Radiometer (AVHRR) satellite imagery to map defoliation by the western spruce budworm in Washington and Oregon. The AVHRR data was obtained as Normalized Difference Vegetation Index (NDVI) maximum value composites from the USGS. Additional data sets included ten years of defoliation data (1989-1998), an EPA ecoregion map, a USFS forest type group map, and a digital elevation model. A biogeographical analysis of the distribution of the defoliation and its potential habitat showed that defoliation was most extensive over the altitude ranges over which the habitat was concentrated. However, when defoliation was expressed as a proportion of the potential habitat, it was found that defoliation was concentrated towards the higher elevations within the potential habitat for the Ponderosa Pine forest type group in Ecoregion 4 [northcentral Washington], whereas in Ecoregion 9 [eastern Cascade Mountains] , the defoliation was more extensive at lower elevations. An automatic classification of defoliation from the NDVI data using maximum likelihood classification was investigated using two different approaches. The first approach utilized inter-annual NDVI differenced data and the second used NDVI data from single years. Classification using both strategies resulted in excessive errors, particularly the classification of non-defoliated forest as defoliated forest. However, the single year approach provided a generally better result than the NDVI differenced data, indicating that the differencing technique did not improve discrimination. Overall classification accuracy of the defoliation and non-defoliation patterns ranged between 54.9 and 78.0 %. Unsupervised classification did not produce any class with a greater than 30% association with defoliation, thus suggesting that defoliation does not appear to have a distinct temporal pattern of NDVI values in the data. Possible causes of the low classification accuracy include (1) the coarse spatial resolution of the AVHRR data, (2) issues relating the generation of the multitemporal compositing, including the lack of radiometric and atmospheric calibration, and (3) the use in this study of a single defoliation class, instead of multiple levels of defoliation. iii