Evaluating the performance and reliability of CA-Markov and CA-ANN models for land use/land cover change simulation in Southwestern Nigeria

Accurate Land Use/Land Cover Change (LULCC) modelling and forecasting are crucial for understanding environmental dynamics and guiding sustainable land management in rapidly urbanising regions. This study compares the predictive accuracy and reliability of two hybrid models, Cellular Automata–Markov (CA–Markov) and Cellular Automata–Artificial Neural Network (CA–ANN) for Southwestern Nigeria. Reliability was assessed as a composite measure combining statistical agreement (accuracy and kappa), spatial consistency (pattern agreement between simulated and observed maps), and class-level stability (persistence of landcover categories over time). Multi-temporal Landsat imagery (2002, 2012, and 2022) was classified and processed in TerrSet’s Land Change Modeller using four GIS-derived drivers: distance to roads, distance to rivers, slope, and elevation. Transitional potentials were calibrated on 2002–2012 and validated against the observed 2022 map. Built-up land increased by over 340%, while wetlands declined by 47%, indicating rapid urban expansion. Accuracy assessment showed CA-Markov (Kappa = 0.846, overall accuracy = 89.4%) outperformed CA-ANN (Kappa = 0.707, overall accuracy = 78.9%), with discrepancies exceeding 15% in simulated areas and 10% in spatial allocation most pronounced in built-up and wetland classes. CA–Markov slightly overestimated urban growth, whereas CA–ANN underestimated built-up expansion and showed greater misclassification, particularly higher emission errors for built-up and water classes. CA-Markov provided the more reliable simulation under these conditions, while CA-ANN requires additional driving variables and parameter optimisation. The study underscores the value of integrated statistical, spatial, and class-level evaluation for LULCC model assessment to support land management and sustainable development planning in Southwestern Nigeria.