Urban Land Use and Land Cover Change Prediction via Self-Adaptive Cellular Based Deep Learning With Multisourced Data

Abstract

The urban sustainable development becomes an essential goal presented in “2030 Agenda for Sustainable development” by the United Nations. Urban land use and land cover (LULC) change prediction, which is a significant indicator to evaluate urban construction strategy, urges to be solved. Remote sensing technique and neural network, are two practical tools widely utilized in urban LULC change prediction. Because of the rapid improvement of remote sensing technique, urban data can be periodically captured in a short time interval. A large amount of data cause the conventional neural networks method having bad efficiency in dealing with it. Moreover, as human activities are in high intensity in the urban area, society related factors require to be taken into consideration in urban land change trend prediction. In this article, a self-adaptive cellular-based deep learning analysis method by utilizing the multisourced data is proposed for urban LULC change prediction. Multisourced data, including weather related data, economy related data, construction related data, and remote sensing data, are normalized and formalized by the proposed self-adaptive cellular-based method. Deep learning long short term memory neural network, which is an advanced model of recurrent neural network and has a strong ability in dealing with sequence data, is utilized to urban LULC change prediction. Remote sensing data captured from 1984 to 2016 are utilized to conduct experiments. Experiment results illustrate that the proposed method can effectively and efficiently make LULC change prediction, and the accuracy is up to $\text{93.1}\%$ .