Modeling urban expansion by integrating a convolutional neural network and a recurrent neural network

Simulating urban expansion (UE) accurately is fundamental for projecting ecological and environmental impacts of future UE, for optimizing the urban landscape patterns, and for improving urban sustainability. We proposed a new UE model by integrating a convolutional neural network (i.e., U-Net) and a recurrent neural network (i.e., long short-term memory, LSTM), and applied it in the Beijing-Tianjin-Hebei urban agglomeration (BTHUA). The results yielded a high overall accuracy (99.18 %), a Kappa coefficient of 0.88 and a figure of merit of 0.13, which are greater than those of existing models. Such improvements are attributed to the multiscale neighborhood information powered by U-Net and the time series information of historical urban expansion uncovered by LSTM. The urban land in the BTHUA is projected to peak at 8736-9155 km(2) during the period 2039-2043, which is an increase in the range of 10.99-16.31 % compared with that in 2020. The results are useful for supporting urban planning in the BTHUA, while the proposed UE model has the potential to be employed worldwide.

Automated summary

Accurately simulating urban expansion is crucial for predicting ecological and environmental impacts, optimizing urban landscape patterns, and improving urban sustainability. In this study, a new model integrating U-Net and LSTM was proposed and successfully applied in the Beijing-Tianjin-Hebei urban agglomeration, yielding better results than existing models. The proposed model has the potential to be used worldwide.