Geo-spatial dynamics and machine learning insights: SVM and RF-driven land use and cover change detection in São Paulo, Brazil

A quantitative evaluation of land-use/land-cover (LULC) changes is required due to the rapid acceleration of LULC change in recent decades, which has been influenced by population growth, economic expansion, and industrial development, particularly in emergent nations. The Sao Paulo region in Brazil faces significant LULC changes due to industrial development, urbanization, and agricultural growth, impacting ecosystems, biodiversity, and water supplies. Addressing these changes involves using Landsat satellite data, sustainable land management, conservation programs, and community involvement. This study compares random forest (RF) and support vector machine (SVM) techniques for classifying LULC features. RF achieved higher accuracy (0.89) compared to SVM (0.76). LULC distribution in 1993 was 3% water, 20% agriculture, 49% forests, 27% built-up, and 1% barren. Projections for 2023 show changes to 2% water, 35% agriculture, 42% forests, 35% built-up, 3% barren, and 5% mining. RF is identified as the superior classifier, though further testing in diverse conditions is recommended.