Utilising land use scenario modeling and machine learning for mitigating drought risks in degraded landscapes

Land-use change is a key driver of environmental degradation and increasing drought risk. This study assesses drought dynamics in the South Malang Plateau, East Java, by integrating remote sensing data with the Random Forest (RF) algorithm. Three land use scenarios were developed: Business-as-Usual (BAU) for 2030 (predicted using the CA-ANN method in QGIS), participatory mapping (PM), and land capability classification (LCC). Using 175 stratified random field points (70% for training, 30% for validation), the analysis integrated 25 predictor variables across climatic, anthropogenic, topographic, and vegetation index factors. The RF model used for drought classification achieved an overall accuracy of 92.57%. Based on unsupervised classification of historical satellite imagery, between 2017 and 2023 multistrata agroforestry declined by nearly 50%, natural forest cover decreased by 27.6%, and settlements more than doubled. Under the 2030 BAU scenario, forest cover is projected to decline further to 9,195.16 ha. Drought analysis shows a peak in 'Severe Drought' at 18.1% in 2019, dropping to 3.1% by 2030, while 'Extreme Drought' steadily rises from 6.2% to 7.0%, particularly in deforested areas. Among the scenarios, the integrated LCCPM approach demonstrated higher potential to reduce drought vulnerability and land degradation. The integrated land capability classification- participatory mapping (LCCPM scenario) is recommended to strengthen landscape resilience and promote sustainable land management.