Radiometric scene correction of temporal multi-spectral satellite data for crop discrimination

Abstract

Multi-date satellite images under different conditions of the same area are difficult to compare because of change in atmospheric propagation, sensor response and illuminations. To overcome this problem, a radiometric normalization technique, which is based on the statistical invariance of the reflectance of man-made in-scene elements (pseudo invariant features) was attempted. The LISS-III data of IRS-1D of three dates were taken for discrimination of crops and retrieval of crop statistics. To develop temporal NDVI profile of the various crop types, relative image-to-image radiometric scene normalization of each band was done using linear transformation. Water body, orchard and other less dynamic features were excluded and multidate-NDVI image having only agricultural crops was obtained for identification and classification of various crops. Nine classes were identified and discriminated as different crops by analyzing temporal NDVI profile pattern based on ground truth, crop calendar and information on crop sowing and harvesting time. Spatial distribution of different crops was analyzed and crop area statistics was computed.