Performance evaluation of MLE, RF and SVM classification algorithms for watershed scale land use/land cover mapping using sentinel 2 bands

Abstract

The land use and land cover map plays a significant role in agricultural, water resources planning, management, and monitoring programs at regional and national levels and is an input to various hydrological models. Land use and land cover maps prepared using satellite remote sensing techniques in conjunction with landform-soil-vegetation relationships and ground truth are popular for locating suitable sites for the construction of water harvesting structures, soil and water conservation measures, runoff computations, irrigation planning and agricultural management, analyzing socio-ecological concerns, flood controlling, and overall watershed management. Here we use a novel approach to analyze Sentinel–2 multispectral satellite data using traditional and principal component analysis based approaches to evaluate the effectiveness of maximum likelihood estimation, random forest tree, and support vector machine classifiers to improve land use and land cover categorization for Soil Conservation Service Curve Number model. Additionally, we use stratified random sampling to evaluate the accuracies of resulted land use and land cover maps in terms of kappa coefficient, overall accuracy, producer's accuracy, and user's accuracy. The classifiers were used for classifying the data into seven major land use and land cover classes namely water, built-up, mixed forest, cultivated land, barren land, fallow land with vertisols dominance, and fallow land with inceptisols dominance for the Vishwamitri watershed. We find that principal component analysis with support vector machine is able to produce highly accurate land use and land cover classified maps. Principal component analysis extracts the useful spectral information by compressing redundant data embedded in each spectral channel. The study highlights the use of principal component analysis with support vector machine classifier to improve land use and land cover classification from which policymakers can make better decisions and extract basic information for policy amendments.