APPLYING MACHINE LEARNING TO TRACK LAND USE CHANGES IN TAN AN CITY, LONG AN PROVINCE IN THE PERIOD OF 2010 - 2024

Abstract

In this study, land use cover changes in Tan An city, Long An province, between 2010 and 2024 are classified and tracked using the Random Forest machine learning algorithm. The input data consists of Sentinel-2 (2024), Landsat satellite images (2010, 2015, 2020), DEM terrain data and indices like NDVI, NDWI, NDBI, BSI, LSWI, MNDWI and AWEI. The training models are built including 6 main land use classes (vacant land, water surface, perennial trees, annual trees, residential areas and traffic). The classification results are highly accurate with a total accuracy (OA) of over 91% and a Kappa coefficient of over 0.89 for each year. The land use dynamics indicate a pronounced urbanization trend - residential and transportation areas expanded markedly, while agricultural land declined. In the period of 2010 - 2024, residential land will increase by more than 1.600 hectares, mainly converted from land for perennial trees and annual trees. Meanwhile, the water surface fluctuates slightly, while the vacant land decreases sharply over the years. Phased overlay maps, spatial variation maps and land-use conversion matrices built and edited on GIS software provided a visual view of the land-use transition process. The results show a clear trend of conversion from agricultural land to construction land, reflecting the strong urbanization rate of Tan An city over the past decade. The study contributes to providing a scientific basis for the assessment and orientation of land use planning in the locality.