Identification of the correlation between land subsidence and groundwater level in Cangzhou, North China Plain, based on time-series PS-InSAR and machine-learning approaches

• Published in: Hydrogeology journal Vol. 32; no. 4; pp. 951 - 966
• Main Authors: Nafouanti, Mouigni Baraka, Li, Junxia, Li, Hexue, Ngata, Mbega Ramadhani, Sun, Danyang, Huang, Yihong, Zhou, Chuanfu, Wang, Lu, Nyakilla, Edwin E.
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.06.2024; Springer Nature B.V
• Subjects: Aquatic Pollution; Aquifers; artificial intelligence; China; Deformation; Earth and Environmental Science; Earth Sciences; evolution; Geology; Geophysics/Geodesy; Groundwater; Groundwater levels; Hydrogeology; Hydrology/Water Resources; Identification methods; Interferometric synthetic aperture radar; Interferometry; Land subsidence; Learning algorithms; Machine learning; Overexploitation; Radar; SAR (radar); subsidence; Synthetic aperture radar; Synthetic aperture radar interferometry; Time series; time series analysis; Waste Water Technology; Water diversion; Water Management; Water Pollution Control; Water Quality/Water Pollution; water table; China; PS-InSAR; Random forest; k-nearest neighbor; China; Groundwater level
• DOI: 10.1007/s10040-024-02771-5
• ISSN: 1431-2174

Land deformation is a severe environmental problem that is often caused by groundwater overexploitation. Traditional approaches, such as those based on ground leveling, are used as standard for monitoring land deformation, but they cannot collect enough information for land-deformation mapping. In this study, the time-series Persistent Scatterer Interferometry Synthetic Aperture Radar (PS-InSAR) was used as an improved method to identify land deformation in Cangzhou after the initiation of China’s South-to-North Water Diversion Project (SNWDP). Machine learning (ML) models, including random forest and k-nearest neighbor, were used to determine the relationship between groundwater pressure and land deformation. The results showed that from 2018 to 2022, the deformation rate was up to –115 mm/year in Nanpi and Dongguang and varied between –57 and –26 mm/year in Qingxian and Cangxian. Land deformation after the SNWDP implementation was less than before. The ML models’ results show that the accuracy of the random forest and k-nearest neighbor methods were 85 and 77%, respectively. Evaluation of the groundwater-level trend measured in six wells showed that after the SNWDP implementation, the groundwater pressure started to recover in Cangzhou, but a decline has been observed recently, particularly in 2022. The mean decrease in impurity (MDI) values demonstrates that aquifers IV and III contribute the most to land deformation in Cangzhou, with the highest MDI values of 33 and 26%, respectively. The study provides new insights into the evolution of regional land deformation, and the methods employed in this research can be adopted in other regions with similar conditions.