Design of a land use change prediction model incorporating neural networks

Abstract

Listen

Translate article

Abstract Land use change has gradually developed into a core area of global environmental change research. In this paper, we use object-orientated modeling to construct a model that combines Markov models, neural networks, and cellular automata. We extend the Markov model to the traditional CA, fully utilizing the advantage of ANN in simplifying the definition of land use transformation rules and obtaining a large number of spatial variable parameters of the model. This successfully simplifies the structure of the model and the definition of transformation rules. We apply the constructed Ann-CA-Markov land use change analysis model to the evolution and prediction of land use in County A. It has been found that the proportion of arable land area in County A decreased from 23.3% to 12.1%, and the proportion of construction land increased from 28.07% to 50.87%. From 2000 to 2020, other land continued to converge into construction land in large quantities, so the land area of County A increased from 1224.73km² to 1295.15km² in 2020. The area of arable land converted out is the largest among the five types of land, with an arable land area of only 308.11km² by 2020. The probability of conversion of four land types, namely, arable land, forest land, grassland, and watershed, to construction land is 21.7%, 10.5%, 10.9%, and 9.2%, respectively, by 2030, while the probability of conversion of construction land to arable land is 21.7%, 10.5%, 10.9%, and 9.2%, respectively. The probability of converting land to cropland is 13.9%. The model constructed in this paper shows strong performance in the analysis of land use change evolution and prediction of County A, which is in line with the design expectation and makes an innovative exploration for realizing the effective simulation of spatial and temporal land use changes.