Data Mining Techniques for Land Use Land Cover Classification Using Multi-temporal Awifs Data

The present study addresses the attempt made to explore the temporal (5-day revisit) and spatial resolution (56m) potential of AWiFS sensor aboard IRS-P6 to generate the land use land cover information using decision tree classification technique using See 5 data mining algorithm. The results obtained after two annual cycles and issues related to digital classification of temporal satellite data were presented and discussed. The temporal datasets were co-registered to sub-pixel accuracy and were atmospherically corrected using modified dark pixel method. Scaled reflectance values were extracted for various classes and rule sets were generated using See-5 data mining algorithm. These rule sets were ported into ERDAS Imagine Knowledge Engineer and the temporal data sets were classified. The results indicate that temporal satellite data at monthly interval found to be suitable to address the seasonal variability in agricultural cropland. The problem with temporal dynamics of cloud cover could be overcome with a little extra care during training site selection. Additional training sites should be defined in cloudy regions keeping its temporal dynamics of the target class in view. Mis-registration among temporal data sets too can influence classification accuracies. Among various land cover classes, classification accuracy is poorer in classes those devoid of vegetal cover. Overall kappa statistic was 0.866 for 200405 which was further improved to 0.908 during 2005-06.