R.D. Lufira, E. Suhartanto, U. Andawayanti, S. Marsudi, A. Azzahra, R.T. Utami
Accurate land-use/land-cover (LULC) prediction underpins sustainable urban and watershed management. This study advances LULC simulation for the Banjir Kanal Timur watershed by coupling Cellular Automata (CA) with Artificial Neural Networks (ANN) in MOLUSCE to enhance spatial realism and predictive skill. Multi-temporal Landsat imagery (2004, 2014, 2024) was classified using supervised Maximum Likelihood. Elevation, slope, distance to roads and rivers, and proximity to built-up areas informed ANN transition potentials, while CA governed spatial allocation. A 2014 to 2024 back-cast validated against the 2024 reference map achieved an overall accuracy of 92.3% and a Kappa of 0.829, reproducing the direction and magnitude of urban expansion, especially intensification along transport corridors and near existing built-up areas. The 2034 forecast indicates continued consolidation of built-up land and encroachment into peri-urban zones, with implications for flood risk, green-space preservation, and infrastructure planning. The novelty of this study introduces a comprehensive CA-ANN approach in MOLUSCE that distinctly differentiates map classification accuracy from simulation proficiency, validates transitions through a 2014-2024 back-cast, and provides a reproducible, policy-ready scenario for 2034 in a tropical urban watershed. The findings support a robust, transferable approach to policy-relevant LULC forecasting in rapidly urbanizing settings. © 2026 Published under licence by IOP Publishing Ltd.
Water Resources Engineering Department, Faculty of Engineering, Universitas Brawijaya, Malang, Indonesia
