Grid graph convolutional network with neighborhood learning for spatio-temporal prediction

We mapped the subsidence of Wuhan using Sentinel-1 synthetic aperture radar (SAR) images acquired during 2015–2019. Overall subsidence coincides with the distribution of engineered geological regions with soft soils, while the subsidence centers shifted with urban construction activities. Correlation between karst subsidence and concentrated rainfall was identified in Qingling–Jiangdi. Results indicate that interferometric SAR can be employed to routinely monitor and identify geohazards.