Machine Learning-Based Detection of Land Use and Land Cover Change in Phu Quoc Special Zone, An Giang Province, Vietnam, Using Google Earth Engine

Abstract

Phu Quoc Special Zone is one of the most developed island destinations in Vietnam. The island has undergone significant land use and land cover (LULC) changes due to urbanization, infrastructure development, and tourism. The objective of this study is to analyze LULC changes from 1995 to 2025 in Phu Quoc usiing Landsat satellite imagery and machine learning (ML) algorithms on the Google Earth Engine (GEE) platform. Four ML algorithms, namely Random Forest, Classification and Regression Tree, Gradient Tree Boosting, and Support Vector Machine, were employed to identify distinct LULC classes, including forest, agricultural land, built-up area, bare land and water body. The classification results yielded an overall accuracy ranging from 85.24 ± 2.81% to 94.41 ± 1.59% and Kappa coefficients ranging from 0.80 ± 0.03 to 0.93 ± 0.02 across the analyzed years, indicating that RF outperformed the other algorithms. Although forest remained the most dominant LULC class, its area decreased by 11.56% over the study period. Conversely, built-up area expanded rapidly by 627.72%, primarily through conversion from forest, agricultural land, and bare land, highlighting the profound impacts of urbanization, infrastructure development, and tourism on LULC changes. The findings demonstrate the effectiveness of integrating ML algorithms within the GEE platform for long-term LULC monitoring. The identified LULC changes provide valuable spatial information to support land-use planning, management, and sustainable land use.