Combined Optical and SAR remote sensing for LULC mapping of Imphal valley using Machine Learning Algorithm

Abstract

Imphal Valley (India) is the part of United Nations Educational, Scientific and Cultural organization (UNESCO) biodiversity hotspots in India where the Imphal River and Loktak Lake form their lifelines. This region is dynamically changing at an unprecedented pace and the effects of such Land use change need to be quantified accurately. In this work, we used both optical satellite data and microwave (Synthetic Aperture Radar (SAR)) data to prepare the Land Use Land Cover (LULC) map of the area for 2015 and 2021. We used Landsat-S and Sentinel 1A satellite data for 2015 and 2021. We compared our results with optical-derived LULC as well as earlier published literature. We have classified the area into seven classes namely waterbody, Phumdis, settlement, horticulture farm, paddy farm, forest, and bare land, and used the Random Forest Machine learning (ML) algorithm for image classification. We observed a significant decrease in waterbody, Phumdis, and forest areas with an increase in the settlement, paddy, and horticulture farm between 2015 and 2021. We observed that settlement and water bodies were better classified when we used both the data rather than optical data only. The k-fold overall accuracy (OA) and Kappa coefficient (KC) for combined data was S7% and 0. S3 for 2015 and OA-90%, KC-0.S6 for 2021, where k=10. These values were lower for only Landsat-8 derived LULC.