Machine learning-based analysis of wetland loss and agricultural expansion dynamics in the Long Xuyen Quadrangle, Vietnamese Mekong Delta (1990–2023)

Abstract

Land use and land cover (LULC) dynamics are critical for understanding environmental changes and their implications for sustainable land management. This study presents the first comprehensive comparative analysis of three widely-used machine learning classifiers (Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Trees (CART)) to assess LULC changes in the Long Xuyen Quadrangle (LXQ), a significant agricultural zone in the Vietnamese Mekong Delta (VMD), over the period 1990–2023. Employing Landsat imagery and machine learning classifiers – RF, SVM, and CART – within the Google Earth Engine (GEE) platform, we evaluated classification accuracy and analyzed LULC transformations. The RF classifier achieved the highest accuracy (overall accuracy: 95.90%; Kappa coefficient: 0.951), significantly outperforming SVM and CART. Results highlighted substantial LULC shifts, notably the extensive expansion of arable cropland, urban growth, and substantial reductions in wetlands and upland forests. Specifically, wetlands declined dramatically by approximately 89% from 2593.96 km² in 1990 to 293.74 km² in 2023, while arable cropland increased by approximately 100% from 1996.81 km² to 3994.39 km² during the same period. These changes were primarily driven by socio-economic factors such as agricultural intensification and urbanization, leading to significant ecological impacts, including biodiversity loss and hydrological alterations. This study underscores the effectiveness of advanced geospatial technologies and machine learning for monitoring and understanding LULC dynamics. The study highlights critical land management needs, including wetland preservation, careful urban planning, and sustainable agricultural practices to mitigate adverse ecological impacts in rapidly developing regions.