Accuracy assessment of biomass and forested area classification from modis, landstat-tm satellite imagery and forest inventory plot data
The objective of this study was to determine how well forestfnon-forest and biomass classifications obtained from Landsat-TM and MODIS satellite data modeled with FIA plots, compare to each other and with forested area and biomass estimates from the national inventory data, as well as whether there is an increase in overall accuracy when pixel size (spatial resolution) decreases. A subset of 1049 inventory plots (100% forested, 100% non-forested) was used to classify the land cover and model the biomass in 20 counties of East Kentucky. Forest inventory data have been further subdivided into two datasets containing 100% forested/non-forested, and only 100% forested plots. Separately, each of these two datasets was used in a decision tree modeling process applied to Landsat-TM, MODIS satellite data, and ancillary data to classify the land cover and model the forest biomass. The satellite, ancillary, and plot data have been processed in See5 and Cubist software. Classification results from trials with Landsat-TM and MODIS show that overall classification accuracy for the percent of pixels correctly classified (%PCC) increased from 85.9% to 89.9%. Classifications from Landsat-TM and MODIS modules show an increase in biomass and forest area when compared to forest inventory estimates, but Landsat-TM module performed better. Comparison between classified forest area with MODIS and Landsat-TM, forest area shows a 2.9% increase. The forest/nonforest single layer classification from each trial was used to mask out non-forested areas for the forest biomass classification. Accuracy of modeled forest biomass was compared with plot data estimates of forest biomass. Biomass obtained from Cubist models with 100% forested forest inventory plots and Landsat-TM images, when compared to the biomass from the published plot data estimates, show a difference less than 2.5%.