Sub-Optimization based Random Forest Algorithm for Accurate and Efficient Land use and Land Cover Classification using Landsat Time Series Data

The land use and land cover (LULC) play an essential role to investigate the impacts of environmental factors and socio-economic development in the Earth’s surface. Extracting the hidden information from the remote sensing images in the observed earth environment is the challenging process. In this research, implemented a model that uses Landsat data to investigate the LULC changes. Utilized the Landsat 5,7 and 8 as inputs for the 1985 to 2019 by Google Earth Engine (GEE) is applied for the robust classification. This paper proposed a Sub-forest optimization based Random forest (SO-RF) classifier with faster diagnosis speed for LULC classification. Moreover, to increase the multispectral Landsat band’s resolution from 30 m to 15 m, the pan-sharpening algorithm is utilized. In addition, analyzed the various image configurations grounded numerous spectral indices and other supplementary data such as land surface temperature (LST) and digital elevation model (DEM) on final classification accuracy. The proposed SO-RF produced higher accuracy (0.97 for kappa, 96.78% Overall accuracy (OA), 0.94 for f1-score) than Copernicus Global Land Cover Layers (CGLCL) map and state of art methods like K-Nearest Neighbor (KNN), Decision Tree (DT), and Multi-class Support Vector machine (MSVM).