Scenario-Based Land Use Change Prediction Using a Hybrid CA-Markov and XGBoost Model: A Case Study of Sejong City, South Korea

Abstract This study develops a hybrid modeling approach that integrates Cellular Automata (CA)-Markov with XGBoost-based logistic regression topredict future land use and land cover (LULC) changes under policy scenarios in Sejong City, South Korea. LULC data from 2005 to 2020is used to predict 2030 and 2035 under four development restriction scenarios: Strict Conservation (S1), Moderate Conservation (S2), Partial Development Allowed (S3), and Business As Usual (S4; BAU). The hybrid model outperforms logistic regression in simulating complex and nonlinear land transition processes(Area Under the Curve > 0.75, Kappa = 0.73). Stricter conservation policies result in increased forest area (55.8 km² under S1 by 2035), decreased agricultural land, and limited expansion of built-up and transportation infrastructure, according to scenario-based simulation. The BAU scenario illustrates the greatest reduction of agricultural land and the least forest expansion, highlighting the detrimental effects of unregulated development. The findings also emphasize barren land may adapt to ecological restoration and urban expansion based on the scenarios. Our findings support Sejong’s 2040 master plan, highlighting development control’s role in climate goals. This study shows methodological advantages of integrating machine learning into the CA-based spatial modelling improve decision support for sustainable urban growth management and carbon mitigation.