Improving spatial land-use and land-cover change simulations with machine learning: A Python adaptation of the CLUMondo model

: Simulating land-use and land-cover change (LULCC) is key for assessing sustainable land-management strategies. Traditional models often rely on logistic regressions (LR) to estimate land-use suitability, which are limited in modelling complex, nonlinear relationships. We present CLUMondoPy, a Python-based adaptation of CLUMondo allowing the integration of advanced machine learning (ML) methods for suitability modelling. Using Mato Grosso, Brazil, we apply ML to simulate LULCC from 2012–2022. We compared different suitability modelling approaches regarding statistical performance (AUC) and impact on land-use allocation (figure of merit – FOM). ML outperformed LR, with random forest showing higher predictive accuracy for suitability (AUC 0.95 vs. 0.82). For land allocation, XGBoost was most accurate (FOM 0.148 vs. 0.074). Projections to 2040 indicated 14.5% less forest loss with XGBoost compared to LR, alongside smaller aboveground biomass declines (10.2% vs. 12.4%). CLUMondoPy demonstrates that ML integration enhances LULCC modelling accuracy and reproducibility, improving scenario reliability to inform decision-making.