A hybrid landscape metric-enhanced cellular automata model (LE-CA) for land use/land cover change simulation: An application to coastal wetlands

Coastal wetlands are critical for carbon sequestration and shoreline protection, yet their rapid transformation demands accurate land-use change modeling for sustainable management. Traditional Cellular Automata (CA) models often neglect landscape structure, limiting their ability to replicate real-world spatial dynamics. To address this gap, we propose a landscape-enhanced CA (LE-CA) model that integrates landscape metrics into the simulation framework. The LE-CA model couples an artificial neural network (ANN) to assess land-use suitability, a Markov module to estimate transition probabilities, and a genetic algorithm (GA) that incorporates landscape indices into the optimization process. The framework was applied to the Luoyangjiang River wetland in southeastern China, using land-use data and driving factors from 2018 to 2022. Model calibration was conducted for 2018–2020 and validation for 2020–2022. Comparative analysis with conventional models (Markov-CA and ANN-Markov-CA) revealed that the LE-CA model achieved superior performance in both overall accuracy (OA = 0.8014) and figure of merit (FoM = 0.3548). It also demonstrated better landscape structural similarity, with a lower RMSE (0.4899) and reduced combined landscape error (1.2305) during validation. These results highlight the LE-CA model’s enhanced ability to capture complex spatial patterns and dynamic land-use processes. By embedding landscape structure into the modeling process, the LE-CA framework offers a more realistic and reliable approach for simulating land-use change in sensitive coastal wetland ecosystems.