Research On The Temporal And Spatial Variation Characteristics Of Regional Groundwater Storage And Drought

The Indus Basin plays an important role in Pakistan’s economic growth due to unique natural resource conditions.The Indus Basin Irrigation System(IBIS)located in the Punjab and Sindh provinces of Pakistan,is an important agricultural production area.In recent years,the significant reduction of groundwater in basin poses a serious threat to sustainable water resources management and agricultural development.Therefore,assessment of groundwater storage is important for estimating groundwater depletion and natural disasters(i.e.,groundwater drought).Meanwhile,there is also an urgent need to study drought characteristics for comprehensive planning of water resources in the basin.This study used terrestrial water storage(TWS)and groundwater storage(GWS)data retrieved from the Gravity Recovery and Climate Experiment(GRACE)satellite,precipitation from the Tropical Precipitation Measurement Program(TRMM),assisted by the Global Land Data Assimilation System(GLDAS)and groundwater level data set from the Punjab Irrigation and Drainage Authority(PIDA)to assess groundwater storage trends,depletion and groundwater drought characteristics in the basin.Different methods were used in this study including uncertainty analysis,non-parametric trend analysis,empirical orthogonal function analysis,downscaling of GRACE data using machine learning models,and wavelet coherence analysis.First,the generalized three-cornered hat method(GTCH)was conducted to evaluate the uncertainty of terrestrial water storage anomaly(TWSA)through different products of GRACE data in the Lower Transboundary Indus Basin(LTIB).The spatial and temporal variability of groundwater storage retrieved by GRACE was obtained by empirical orthogonal function(EOF)analysis.In addition,monthly GRACE-derived GWSAs were assessed using the GLDAS model data,and Mann Kendal trend tests and robust regressions were used to assess time-series trends and groundwater storage depletion in the LTIB.The results show,TWSA and precipitation data delineate the seasonal characteristics of summer peaks and winter troughs,reflecting changes in GWSA,respectively.GRACE-derived GWSA is being depleted at a rate of 4.16±0.26 mm(2.97±0.19 km~3)per year.The measured correlations(R~2)of the long-term monthly mean of the GRACE-derived GWSA with the PCRaster Global Balance Model(PCR-GLOBWB)and the Water Gap Global Hydrological Model(WGHM)were 75%and 81%,respectively.On the seasonal and annual scales,the in situ data for the GRACE-derived GWSA showed good correlations of 0.69 and 0.82,respectively.Secondly,a machine learning-based random forest model(RFM)and artificial neural network(ANN)model were used to downscale GRACE data(TWS and GWS)from 1°to a higher resolution(0.25°)due to coarse resolution of GRACE data.For this purpose,GRACE’s TWS changes in combination with geospatial variables including digital elevation models(DEM),slope,aspect,and hydrological variables such as soil moisture,evapotranspiration,rainfall,surface runoff,canopy water,and temperature were used in downscaling.The model va lidation results show that RFM outperformed the ANN model with a Pearson correlation coefficient(R)of 0.97,a root mean square error(RMSE)of 11.83 mm,and a mean absolute error(MAE)of 7.71 mm,and the Nash efficiency(NSE)was 0.94,using training data set from 2003 to 2016.This indicated that RFM is suitable for downscaling the GRACE data on a regional scale.After downscaling,the correlation coefficients between the downscaled GWS and the observation well GWS were 0.67and 0.77,respectively.Therefore,it can be concluded that RFM has the potential to downscale GRACE data at spatial scales suitable for predicting GWS.Third,this study applied four machine learning models to a training dataset of GRACE TWS and GWS data to increase the resolution from 1°to 0.25°to assess the groundwater drought.The results show that the XGBoost model outperformed with a correlation coefficient(0.99),NSE(0.99),root mean square error(RMSE)(5.22 mm),and mean absolute error(MAE)(2.75 mm).The groundwater storage downscaled by the XGBoost model show ed a good correlation with the in situ groundwater storage.This study also analyzed the spatiotemporal evolution characteristics of GGDI in the basin from 2003 to 2016,and used the wavelet coherence method to evaluate the relationship between teleconnection factors and GGDI.GGDI was validated using 1,3,and 6-month cumulative standardized precipitation evapotranspiration index(SPEI)and self-calibrated Palmer Drought Severity Index(sc-PDSI)drought pattern.The teleconnection factors had a significant effect on the GGDI shown by the wavelet coherence technique.In general,over-exploitation has led to a downward trend in groundwater storage on the temporal and spatial scales,resulting in groundwater drought in the basin.Secondly,the framework proposed in this study can serve as a tool for drought monitoring and provides valuable information for understanding extreme hydroclimatic conditions in the Indus Basin Irrigation System and similar climatic regions.