Reconstructing GRACE-like Time Series Of Water/ice Mass Anomalies

Water resources sustain global ecosystems and provide important support for food and economic activities.Accurate estimates of water storage changes are critical to ensuring food supply security,maintaining human and ecosystem health and harmony,and promoting energy production and social stability.The implementation of the gravity satellite missions,GRACE(Gravity Recovery and Climate Experiment)and GRACE-FO(GRACE Fllow-On),provide an unprecedented tool for people to effectively monitor long-term,continuous,and large-scale changes in water storage.However,there are also problems,such as the limited observation time span(from 2002 to the present)and the inability to isolate specific driving sources(climate driving or human intervention).Although other satellite geodetic observation methods,such as satellite laser ranging,High-low satellite-to-satellite tracking method,and satellite altimetry,can obtain the information on water mass transports prior to the GRACE period.Subject to coarse spatial and(or)temporal resolution,the application of these data is limited to large areas(e.g.Greenland and Antarctica).Existing methods for reconstructing GRACE-like mass anomalies,the machine learning approaches,have been widely used in the extension of time series of mass anomalies.However,they are not good at reconstructing long-term trend items.In this study,based on statistical models,we investigated the following two aspects:reconstructing the time series of GRACE-like mass anomalies and separating different driving signals from mass anomalies.1.We proposed a new statistical model for reconstructing climate-driven water storage anomalies(CWSAs)with full signal(seasonal and non-seasonal variations)using precipitation and temperature data as input.The seasonal signal of CWSA was reconstructed using statistical model for the first time.The evaluations based on 55 river basins show that for the basins with few human interventions,our reconstruction results are closer to GRACE observations than these calculated using land surface models,and are similar to those based on machine learningbased methods.In contrast,reconstruction misrepresents CWSA in some basins with widespread frozen ground and glaciers,or with huge lakes,because of limitations of the proposed method(only for hydrological signal).2.We explored and validated the feasibility of quantitatively separating human-driven water storage anomalies(HWSA).Total water storage anomaly(TWSA)detected by GRACE includes the contributions from climate variability and human intervations.Therefore,we further discussed the feasibility of isolating the effect of direct human intervention on water storage from GRACE observation based on the correct reconstruction of CWSA.Discrepancies between reconstructed CWSA and GRACE TWSA occurr in high-intensity human-intervention basins,and mainly reflect different trends between the reconstruction and GRACE data.Isolated HWSAs in four representative basins(the Haihe,Huaihe,Central Valley and Tigris-Euphrates basins)accorded with the characteristics of anthropogenic water depletion in both seasonal and non-seasonal signals.Our isolated results were also verified by the groundwater well observations in the Haihe basin and in-situ observations of reservoir storage in the Yangtze basin.3.We proposed a statistical model for reconstructing the time series of GRACE-like mass anomaly on high mountain glaciers(HMGs),using precipitation and temperature data as input.Given that the mass anomalies GRACE observed on HMG are the mixed signals,i.e.,glacier and snow(GS)signal and hydrological signal,thus,our model includes two functions:(i).reconstructing the mass anomalies GRACE detected,(ii).separating GS signal and hydrology signal from the reconstruction.The results of reconstructing HMG mass anomalies in 14 regions based on five precipitation/temperature(P/T)products showed that our model that can effectively and feasibly reconstruct GRACE-like mass anomalies on HMGs and retrospect them.4.We reconstructed the time series of HMG mass anomalies in 14 regions starting from1979 and effectively separated the GS and hydrological signals from our reconstructions.The reconstructions in the 14 regions were excellently consistent with GRACE/GRACE-FO observations.Our reconstructions’ average median CC/NSE values based on five P/T products were 0.92 and 0.84,respectively.The time series of mass anomalies observed by SLR verified the capacity of this model for retrospecting mass anomalies prior to the GRACE period.Compared with the deficiency in trend estimation based on the machine learning method,our model could more effectively reconstruct the trend of GS mass anomaly.External independent data(e.g.,ICESat,GLDAS,and MODIS)verified the reliability of our separated GS signals.