Modelling and prediction of land use and land cover change using machine learning in the Michiru Forest Reserve, Southern Malawi

Abstract

Monitoring land use and land cover (LULC) change is crucial for ecological sustainability and urban planning. The Michiru Mountain Forest Reserve in Malawi faces mounting pressure from rapid urbanization and population growth, making predictive modeling essential for conservation planning. This retrospective focus limits the ability of policymakers to plan proactively, highlighting a critical need for predictive tools that can model the complex, non-linear dynamics of land use change in diverse landscapes. This study employs a Multi-Layer Perceptron (MLP) neural network to model LULC dynamics and Random Forest classifier for land cover classification in Malawi's Michiru Mountain Forest Reserve from 2004 to 2024. The optimized MLP model achieved 87.2% accuracy, a 0.83 skill measure, and a Kappa coefficient of 0.74, with minimal overfitting (training RMS: 0.0368, testing RMS: 0.0396). During this period, vegetation cover declined from 85.77 to 60.07 km2 (30% loss), while bare land increased by 123% and built-up areas nearly doubled from 4.58 to 8.77 km2. Spatial analysis shows 70% of urban expansion occurred within 500 m of roads, emphasizing the role of infrastructure. Forecasts for 2034 project built-up areas to extend to 23.20 km2 and further vegetation decline, highlighting ongoing ecosystem degradation and underscoring the urgent need for integrated land management strategies.