Analysis of cultivated land changes and driving factors in the Alar Reclamation Area (1990–2019) based on multi-temporal Landsat data and machine learning algorithms

Clarifying the dynamic changes in cultivated land and their driving factors is crucial for ensuring national food security and optimizing land use structure. This study focuses on the Alar Reclamation Area in southern Xinjiang, China, based on Landsat satellite images from seven key years (1990, 1994, 2000, 2006, 2010, 2015, and 2019) and corresponding socio-economic and meteorological data. Six machine learning algorithms-Spectral Angle Mapper (SAM), Artificial Neural Network (ANN), Minimum Distance Classification (MDC), Maximum Likelihood Classification (MLC), Support Vector Machine (SVM), and Support Vector Machine-Conditional Random Field (SVM-CRF)-were compared to identify the optimal method for land use/cover classification. The results showed that the SVM-CRF algorithm achieved the highest accuracy (Overall Accuracy = 0.95; Kappa = 0.94). Cultivated land area increased by 729.97 km² from 1990 to 2019, showing an outward expansion trend. Path analysis, based on annual regional data, revealed that total population, GDP, total fixed asset investment, total agricultural output value, and cotton price were the major drivers. GDP had a negative direct effect on cultivated land area, reflecting industrialization and urban expansion. This study demonstrates that integrating machine learning classification with socio-economic and natural indicators provides a robust approach for understanding land use change mechanisms in arid oasis regions.