Minimum Error Land Cover Separability Analysis And Classification Of Modis Time Series Data

Three minimum-error land cover classifiers are compared on coarse resolution MODerate-resolution Imaging Spectroradiometer (MODIS) surface reflectance data for discerning between vegetation and settlement classes. It is shown that good class separability can be achieved using only the seasonal component of Normalized Difference Vegetation Index (NDVI) data, or the mean component of several other MODIS land bands. It is also shown why particular classifiers fail in certain spectral bands. Finally it is shown that after NDVI, band 2 has the highest separability of all the MODIS land bands, and band 5 has the lowest separability.