Comparative Evaluation of DYNA-CLUE and a Python-Based Machine Learning Framework for Long-Term Land-Use and Land-Cover Simulation

Accurate simulation of land-use and land-cover change (LUCC) is essential for understanding landscape dynamics and supporting environmental decision-making. This study compares DYNA-CLUE with a Python-based LUCC framework (PBLCF) under identical demand, driver, and validation conditions. Unlike regression-based CLUE approaches that derive suitability mainly from a single baseline map, PBLCF learns transition behavior from two temporal land-cover maps using Random Forest and embeds these transition probabilities within a demand-constrained allocation process. This provides a simpler and more flexible hyperparameter structure for controlling transition intensity, scenario-based change, and unrealistic conversions. LUCC Results show that PBLCF achieves higher predictive accuracy (OA = 0.864; Kappa = 0.815) than DYNA-CLUE and improves representation of ecotonal and disturbance-driven classes by capturing non-linear interactions among drivers. In addition, class-specific constraints enhance the persistence of stable land-use types, reducing unrealistic transitions. The framework demonstrates improved long-term stability and provides a flexible, reproducible approach for scenario-based LUCC simulation. • This study compares DYNA-CLUE with a Python-based LUCC simulation framework. • Machine learning models simulate nonlinear land-use transitions. • The framework integrates suitability modelling with demand-driven allocation. • The framework improves long-term simulation of dynamic LUCC. • The open-source workflow supports reproducible LUCC simulations.