Coupling Machine Learning and Cellular Automata-Markov Chain to Model Urban Expansion in a Fast Developing Area: A Case Study of Liangjiang New District of Chongqing, China. GeoComputation 2019

Timely monitoring and modeling of landuse change is vital to manage land resources effectively and to rectify defective land use policies, especially in rapidly urbanizing areas. Our research is applied to an annual land use dataset in order to simulate urban expansion in a fast developing area, through a hybrid model coupling Artificial Neural Networks (ANNs), cellular automata (CA), and Markov Chain (MC). The ANNs were optimized to create the urban suitability index (USI) map that was then integrated with CA-MC to spatially allocate urban expansion cells. Two ANNs, multiple-layer perceptron (MLP) and long short-term memory network (LSTM), were implemented comparatively. Since LSTM is able to take into account more temporal information, it outperformed MLP in modelling urban expansion process over a short temporal interval. The results, validated using kappa and fuzzy kappa simulation, indicate that the integration of ANNs with CA-MC can capture the possible nonlinear relationship between urban expansion and its drivers, hence it can accurately simulate and predict urban expansion in the study area.