Integration of SCATSAT Data and Unsupervised Learning for Accurate Land Use/Land Cover Characterization in Rajasthan

Abstract

Land Use/Land Cover (LULC) plays one of the crucial roles for sustainable environmental management, agricultural planning, ecosystem status monitoring, and regional development. These techniques have limitations, such as cloud cover, atmospheric disturbance and seasonal variability, which may impact the consistency of LULC mapping. All-weather day-night capabilities of microwave remote sensing systems, especially scatterometers, are useful for monitoring terrestrial surfaces continuously. Current Research work focussed on the use of the data received from the satellite SCATSAT-1 with the unsupervised machine learning approach for LULC characterization in the state of Rajasthan, India. The grouping of SCATSAT derived backscatter signatures in various land cover clusters is done by the research using K-means clustering algorithm. Afterwards, the rules are used to classify the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI), which then are compared to the original data. Multi-temporal observations from SCATSAT are used in the data cleaning and raster transformation, feature extraction and clustering workflows. Consistency of the classification and spatial correspondence of the clusters generated is evaluated by comparing all the clusters with the vegetation-based classification and the ground reference information. The diverse topography of Rajasthan (desert, agriculture, forest, scrubland and urban areas) makes it an ideal testbed. The study shows that scatterometer backscatter data can be used to differentiate major land surface features and to aid the characterization of regional scale LULC. The integration of microwave remote sensing and unsupervised learning could lead to a computationally efficient processing scheme for large area monitoring applications for land and also aid future geospatial intelligence applications based on SCATSAT observations.