Use of Savitzky - Golay Filters to Minimize Multi-temporal Data Anomaly in Land use Land cover mapping
Land use land cover characterization and mapping have become a prerequisite in all environmental Planaing. The array of satellites deployed in the space provides multi-temporal images that can be used for the land use land cover classification. But, much often these multi-temporal images have data noise and anomaly owing to the cloud and atmospheric effects. This brings pseudo hikes and lows in data adding classification with possible errors. We present a method for the removal of data anomaly where monthly data of MODIS (Moderate Resolution Imaging Spectroradiometer) Normalized Difference Vegetation Index (MODIS 13Q1) was used for the classification of images over a large area encompassing the SAARC nations. MODIS multi-temporal data were filtered usinga Savitzky-Golay (S-G) algorithm which provided smoothened data and the seasonality (start, end of the season) were identified. Phenology profile curves were created for the characterization of the agriculture and forestry feature classes. The S-G filtered images and raw MODIS data phenology profile curves were compared for the eleven classes of land cover, viz., ever green needle forest, ever green broad leave, deciduous broad leave, shrub, savannas, grass, agriculture, built-up, water, snow (ice), and barren. Spectral signature separability was also compared using Euclidean spectral distance method. In conclusion, it was observed that multi-spectral S-G filtered data were more useful for the classification of agriculture and forestry classes for a larger coverage.