Comparative study of multiple algorithms classification for land use and land cover change detection and its impact on local climate of Mardan District, Pakistan
Farnaz, Narissara Nuthammachot, Muhammad Ali
Abstract
Land use and land cover (LULC) changes significantly impact global climate change, resource management, and sustainability. This study aims to evaluate the performance of various machine learning classifiers, including Support Vector Machine (SVM), Random Forest Algorithm (RFA), K-Nearest Neighbor (KNN), and Maximum Likelihood (MLH), in detecting Land Use and Land Cover [LULC] changes in Mardan District, Pakistan, from 2015 to 2023. Sentinel-2 satellite imagery was utilized to generate LULC maps, which were subsequently analyzed to quantify changes across five land cover classes: water land, built-up areas, barren land, green land, and farmland. The study also investigates the impact of LULC changes on climate regulation and sustainability within the study area. The findings reveal that SVM and RFA classifiers exhibited the highest overall accuracy and kappa coefficients, outperforming KNN and MLH. Significant transitions were observed, including urban expansion, reforestation efforts, and agricultural stability. Furthermore, an analysis of climate data from 2015 to 2023 revealed a notable increase in minimum, maximum, and mean temperatures within the areas impacted by LULC changes. The study highlights the importance of selecting appropriate classifiers for accurate LULC change detection and underscores the need for informed decision-making in environmental management and urban planning to mitigate the impacts of climate change and promote sustainable development.