Open-Data GeoAI for Short-Term Urban Land-Use/Land-Cover Forecasting in Rapidly Growing African Cities: A Reproducible Workflow and the Kinshasa (DRC) Case Study

Rapid urban growth in many African cities frequently outpaces official mapping and monitoring, limiting evidence-based urban planning in data-scarce contexts. This study develops and tests a fully open-data, reproducible geospatial artificial intelligence (GeoAI) workflow to forecast short-term urban land-use/land-cover (LULC) change, using Kinshasa (Democratic Republic of the Congo) as a demonstrative case study. Multi-temporal LULC maps for 2016, 2019, 2022 and 2025 were produced at 10 m resolution from Sentinel-2–derived Dynamic World Version 1 in Google Earth Engine and combined with open explanatory variables derived from NASADEM topography and OpenStreetMap proximity layers. Transition potentials were modelled in TerrSet Land Change Modeler using a multilayer perceptron (MLP) and a support vector machine (SVM) and then integrated with Markov-chain allocation to simulate near-term scenarios for 2028 and 2031 (three-year steps). From 2016 to 2025, built-up areas increased by approximately 114.48 km², while vegetation and croplands declined by about 149.64 km² and 50.80 km², respectively, with the most dynamic changes occurring during 2019–2022; vegetation was the principal donor class to urban conversion (≈110.65 km²). Predictive performance was high, with area-under-the-curve values exceeding 0.89 and overall Kappa statistics ranging between 0.78 and 0.83, indicating reliable short-term forecasting skill. The projected maps for 2028 and 2031 suggest continuing peri-urban sprawl and progressive fragmentation of vegetated zones, demonstrating that open-data GeoAI can deliver actionable, transferable spatial foresight for rapidly growing African cities.