| Land subsidence,as a worldwide geological hazard,has become a complex system problem that requires a comprehensive and interdisciplinary approach.It severely constrains the sustainable development of society and economy,with groundwater-related subsidence being the most common form.Therefore,it is crucial to study and track the complete evolution process of surface deformation in the past and present,predict future development trends,effectively monitor the aquifer,and estimate changes in large-scale groundwater storage to mitigate disaster risks and protect human property,as well as reduce the negative impact of groundwater depletion.Interferometric Synthetic Aperture Radar(InSAR)technology is a relatively mature space-based observation technique developed over the past thirty years,which has the advantages of wide coverage,high spatial and temporal resolution,and being unaffected by clouds and rain.It can obtain surface deformation monitoring results with millimeter-level accuracy,providing new opportunities and possibilities for multi-scale and multi-dimensional surface deformation monitoring and related physical parameter inversion research.However,InSAR technology still faces many challenges in the high-precision monitoring and interpretation of large-scale deformation,such as multi-source deformation signal decomposition,error correction,multi-orbit data fusion,and quantitative interpretation.Additionally,in recent years,with the deepening development of machine learning,it has been widely applied in remote sensing.By introducing machine learning models,it is possible to jointly predict the spatiotemporal land motions of land surface,and reveal the processes of different remote sensing dataset’s effects on surface deformation and groundwater changes.Based on this,this study aims to use a combination of time-series InSAR technology and machine learning methods to carry out high-precision,multi-dimensional,multi-scale surface parameter inversion research related to groundwater in the context of ground subsidence as a geological hazard.The research focuses on ground subsidence in four types of geological conditions in domestic and international regions,including urban areas,farmland,coastal plains,and river deltas.The study deepens the application potential of InSAR technology in this field,while also expanding the breadth and depth of machine learning in InSAR time-series deformation analysis.The main research contents and achievements of the paper are as follows:1.Employ a combination of multi-band and multi-orbit(ascending and descending)geometric SAR data and interferometric point target time series analysis techniques in response to the challenges of multi-scale and multi-dimensional monitoring and interpretation of surface deformation.The study addresses the issues of atmospheric errors,phase unwrapping errors,and multi-orbit data fusion in large-scale deformation monitoring,and reveals the multi-scale temporal and spatial evolution processes of surface deformation in Xi’an and the Shandong Peninsula.The study successfully constructs a depth inversion model for ground subsidence caused by groundwater exploitation in Xi’an using the Okada homogeneous elastic half-space rectangular dislocation model and determines the optimal model parameters.Additionally,multi-dimensional monitoring of surface deformation in the coastal area of the Shandong Peninsula is conducted,and the various inducing and driving factors of deformation in this region are quantitatively analyzed.2.Propose a decomposition strategy based on independent component analysis(ICA)model for InSAR time-series deformation to address the problem of large-scale surface InSAR interferometric time-series signal mixing.Taking ground subsidence deformation in the Willcox Basin of the United States and the comprehensive deformation field of the Iran-Iraq earthquakes as research objects,the contribution of deformation at different scales in the comprehensive time-series deformation field was successfully separated based on the ICA analysis method.Spatial-temporal distribution maps of different signal sources were plotted,revealing the spatiotemporal evolution process of time-series deformation under different geological processes and seismic events.Simultaneously,by decomposing the phase signal related to atmospheric delay and thermal noise,the signal could be removed to optimize InSAR deformation monitoring results.3.Conduct study on the estimation of parameters and water storage of confined aquifers based on Willcox deformation and groundwater level time series observations.A joint method for estimating the parameters of the confined aquifer and the irrecoverable water storage was proposed,which is based on the Independent Component Analysis(ICA)decomposition.The study focused on the Willcox Basin in the United States and obtained high-precision Willcox deformation results of the area.The skeletonized specific storage coefficient and spatiotemporal distribution of water storage change of the confined aquifer were inferred through the inversion of the groundwater head data.The results shows that the average specific storage coefficient in the Willcox Basin decreased annually from 0.008 to 0.005,indicating that the long-term overpumping of groundwater has changed the structure of the confined aquifer system and resulted in the degradation of its storage capacity.This irreversible and irrecoverable deformation of the confined aquifer system indicates that groundwater exploitation in the area is unsustainable.4.Proposed an innovative spatiotemporal framework for surface deformation prediction,aiming to address the low prediction accuracy issue of large-scale surface InSAR temporal deformations,namely the ICA-assisted LSTM deep learning model.Considering the spatial heterogeneity of large-scale InSAR spatiotemporal deformations,machine learning-based spatiotemporal clustering was conducted for homogeneous InSAR spatiotemporal deformation points to achieve higher precision in broad-scale InSAR deformation prediction.The experimental study was conducted using the surface deformation data of the Willcox Basin in the United States,and the results showed that the proposed method achieved a 34%improvement in surface deformation temporal prediction accuracy compared to traditional single models.5.Preliminary investigation on the integration of Random Forest and multi-Source remote sensing datasets for analyzing and predicting annual variation of groundwater level and land subsidence in subsidence areas.Using Salt Lake Basin in Iran as a case study,we combine 11 types of multi-source remote sensing products,including meteorological and hydrological data,InSAR observations,and low-resolution water storage estimates from GRACE satellite.Based on the random forest model,the relationships between various datasets and predicted the changes in land surface deformation and groundwater level is analyzed.Our results demonstrate that this machine learning model can predict 92% of groundwater level changes and 82% of land subsidence deformation in the Salt Lake Basin area.By integrating multiple remote sensing datasets and the random forest model,we can more accurately describe the nonlinear dynamic changes in groundwater and land surface deformation. |
PDF Full Text Request