Modeling land use change prediction using multi-model fusion techniques: A case study in the Pearl River Delta, China

Abstract

Cellular automata (CA) model has powerful spatial analysis capabilities and easy coupling with other models. However, the coupling based on single model and CA model is difficult to take into account both the accuracy of the model and the interpretation of the driving factors. Moreover, the "bottom-up" feature of the model makes it difficult to take into account the macro regulation. Based on this, we proposed a multi-model coupled multi-agent system (MAS), convolutional neural network (CNN) and cellular automata (CA) model for land use change simulation. Using the Pearl River Delta (PRD) as the study area, the simulation study was conducted from 2000 to 2010 as the test period and from 2010 to 2020 as the validation period, respectively. The results of the study are as follows: (1) Compared with the CNNCA model, the MAS-CNNCA model has improved accuracy validation in both the test period and the validation period, which indicates that the strong classifier composed of multiple agents as weak classifiers is more accurate in identifying the suitability of land use change. (2) The dual perspective consisting of macro-agent and micro-agent together is more profound for land use change mechanism exploration. (3) The coupling of simulation results and the geographic detector model can reveal the interactive relationship between drivers and land use change.