Downscaling Of GRACE Datasets Based On Relevance Vector Machine Using InSAR Time Series

Under the influence of human activities,the global climate changes greatly.Issues,such as flood,drought,sea level rise and so on,have threaten the development of human beings seriously.The occurrence of most meteorological disasters is closely related to water.As an important part of terrestrial water resources,the change in quantity and quality of groundwater will have a significant impact on human’s manufacture and life.The investigation of groundwater resources can improve people’s understanding of regional groundwater status and related influencing factors,and take timely measures for sustainable development in groundwater use.Generally,conventional methods for acquisition of groundwater reserves are high cost,limited and not intuitive.Gravity recovery and Climate Experiment(GRACE)provides a large-scale and non-contact measurement technique,and its appearance provides a new method for regional and global water resources investigation and related studies.Many scholars have studied terrestrial water resources,glaciers,oceans and ice caps using GRACE dataset,and have a comprehensive understanding of regional and global water resources.However,due to the limitation of resolution for GRACE,it is difficult to conduct studies at fine-scale directly.Many scholars have carried out studies on downscaling models for GRACE data.In this study,to generate changes in groundwater storage(? am)at a finer scale,we developed a statistical downscaling model and applied it to GRACE.Differently,the model is based on the relevance vector machine(RVM).Additionally,land subsidence rate was introduced into the model.RVM is a statistical machine learning method which is based on Bayesian theory.It has the few parameters and the necessary accuracy in regression of small datasets.Considering the anthropogenic influence on the GWS,land subsidence rate,inversed by small baseline set interference technology,was introduced.The 0.1° × 0.1° ? ammaps were obtained by constructed downscaling model and raw GRACE datasets with 1° ×1°.After validation,the model was evaluated and compared with that constructed using artificial neural network(ANN)and support vector machine(SVM)according to current studies.The simulation results of the downscaling model show that the study area suffered from a sustained groundwater reduction with an expansion of space during 2007-2010.Meanwhile,through our study,it is shown that RVM is suggested to construct the GRACE downscaling model instantly for better performance than ANN and SVM.The rate of land subsidence can be considered an important indicator for groundwater storage inversion.Finally,the limitations of the proposed downscaling model are analyzed,and the corresponding solutions and possible improvements are proposed.