Prediction of urban land subsidence by SBAS-InSAR and improved BP neural network

In response to the issues of excessive reliance on subsidence data and a lack of model diversity in existing urban ground subsidence prediction methods, this study focuses on the main urban area of Kunming City, Yunnan Province. A novel approach for predicting urban ground settlement is proposed, incorporating a multi-temporal sequence and multifactor perspective into the improved BP neural network. Firstly, SBAS-InSAR technology is utilized to acquire monitoring values of ground settlement in the main urban area. Subsequently, gray correlation analysis and factor analysis in SPSSAU software are employed to identify the influencing factors of ground settlement in this specific area. Based on the obtained settlement monitoring values and the identified influencing factors, GA-BP and PSO-BP prediction models are constructed from a multifactor multi-temporal sequence viewpoint. The optimal prediction model is determined and its performance is evaluated through comprehensive validation. Experimental results demonstrate that SBAS-InSAR effectively monitors urban ground settlement, while the GA-BP algorithm outperforms the PSO-BP algorithm in terms of prediction accuracy and overall performance. This method allows for long-term and large-scale predictions of urban ground settlement, as well as forecasting the settlement trends of specific points over multiple periods. Consequently, it serves as an effective tool for urban ground settlement prediction, providing governmental departments with an efficient and fast decision-making approach.