Recognition of arable lands using multi-annual satellite data from spectroradiometer MODIS and locally adaptive supervised classification

The arable lands recognition method is developed based on multiannual time-series of remote sensing data acquired by spectroradiometer MODIS on board of Terra and Aqua satellites. The method involves producing of the recognition features set, which exploits differences of seasonal and interannual changes of spectral reflectance for arable lands on one hand and other types of agricultural lands and natural vegetation on another hand. The arable lands recognition utilizes the locally-adaptive supervised classification algorithm, which accounts the spatial variability of the considered features for classes to be discriminated. The developed method has been applied to produce the arable lands map for entire Russia. The arable lands map validation based on Pareto optimum approach and reference data has been performed for the test region in order to estimate the method's accuracy and potential for its further enhancement.