Generating Long-term GRACE-like Total Water Storage Change Products using Conditional Generative Adversarial Networks

Abstract. The availability of data is a major challenge for hydrological modelling in large parts of the world. Remote sensing data can be exploited to improve models of ungauged or poorly gauged catchments. In this study we combine three datasets for calibration of a rainfall-runoff model of the poorly gauged Okavango catchment in Southern Africa: (i) surface soil moisture (SSM) estimates derived from radar measurements onboard the Envisat satellite; (ii) radar altimetry measurements by Envisat providing river stages in the tributaries of the Okavango catchment, down to a minimum river width of about one hundred meters; and (iii) temporal changes of the Earth's gravity field recorded by the Gravity Recovery and Climate Experiment (GRACE) caused by total water storage changes in the catchment. The SSM data are shown to be helpful in identifying periods with over-respectively underestimation of the precipitation input. The accuracy of the radar altimetry data is validated on gauged subbasins of the catchment and altimetry data of an ungauged subbasin is used for model calibration. The radar altimetry data are important to condition model parameters related to channel morphology such as Manning's roughness. GRACE data are used to validate the model and to condition model parameters related to various storage compartments in the hydrological model (e.g. soil, groundwater, bank storage etc.). As precipitation input the FEWS-Net RFE, TRMM 3B42 and ECMWF ERA-Interim datasets are considered and compared.

Improving drought simulations within the Murray-Darling Basin by combined calibration/assimilation of GRACE data into the WaterGAP Global Hydrology Model

NASA’s Gravity Recovery and Climate Experiment (GRACE) has already proven to be a powerful data source for regional groundwater assessments in many areas around the world. However, the applicability of GRACE data products to more localized studies and their utility to water management authorities have been constrained by their limited spatial resolution (~200,000 km2). Researchers have begun to address these shortcomings with data assimilation approaches that integrate GRACE-derived total water storage estimates into complex regional models, producing higher-resolution estimates of hydrologic variables (~2500 km2). Here we take those approaches one step further by developing an empirically based model capable of downscaling GRACE data to a high-resolution (~16 km2) dataset of groundwater storage changes over a portion of California’s Central Valley. The model utilizes an artificial neural network to generate a series of high-resolution maps of groundwater storage change from 2002 to 2010 using GRACE estimates of variations in total water storage and a series of widely available hydrologic variables (PRISM precipitation and temperature data, digital elevation model (DEM)-derived slope, and Natural Resources Conservation Service (NRCS) soil type). The neural network downscaling model is able to accurately reproduce local groundwater behavior, with acceptable Nash-Sutcliffe efficiency (NSE) values for calibration and validation (ranging from 0.2445 to 0.9577 and 0.0391 to 0.7511, respectively). Ultimately, the model generates maps of local groundwater storage change at a 100-fold higher resolution than GRACE gridded data products without the use of computationally intensive physical models. The model’s simulated maps have the potential for application to local groundwater management initiatives in the region.

Regions with high productivity of agriculture, such as the Beqaa Plain, Lebanon, often rely on groundwater supplies for irrigation demand. Recent reports have indicated that groundwater consumption in this region has been unsustainable, and quantifying rates of groundwater depletion has remained a challenge. Here, we utilize 15 years of data (June 2002–April 2017) from the Gravity Recovery and Climate Experiment (GRACE) satellite mission to show Total Water Storage (TWS) changes in Lebanon’s Beqaa Plain. We then obtain complimentary information on various hydrologic cycle variables, such as soil moisture storage, snow water equivalent, and canopy water storage from the Global Land Data Assimilation System (GLDAS) model, and surface water data from the largest body of water in this region, the Qaraaoun Reservoir, to disentangle the TWS signal and calculate groundwater storage changes. After combining the information from the remaining hydrologic cycle variables, we determine that the majority of the losses in TWS are due to groundwater depletion in the Beqaa Plain. Results show that the rate of groundwater storage change in the West Beqaa is nearly +0.08 cm/year, in the Rashaya District is −0.01 cm/year, and in the Zahle District the level of depletion is roughly −1.10 cm/year. Results are confirmed using Sentinel-1 interferometric synthetic aperture radar (InSAR) data, which provide high-precision measurements of land subsidence changes caused by intense groundwater usage. Furthermore, data from local monitoring wells are utilized to further showcase the significant drop in groundwater level that is occurring through much of the region. For monitoring groundwater storage changes, our recommendation is to combine various data sources, and in areas where groundwater measurements are lacking, we especially recommend the use of data from remote sensing.

Sea level rise is perceived as a major threat to the densely populated coast of the Bay of Bengal. Addressing future rise requires understanding the present‐day sea level budget. Using a novel method and data from the Gravity Recovery and Climate Experiment (GRACE) satellite, we partition altimetric sea level rise (6.1 mm/a over 2002–2014) into mass and steric components. We find that current mass trends in the Bay of Bengal are slightly above global mean, while steric trends appear much larger: 2.2–3.1 mm/a if we disregard a residual required to close the budget, and 4.3–4.6 mm/a if, as an upper bound, we attribute this residual entirely to steric expansion. Our method differs from published approaches in that it explains altimetry and GRACE data in a least squares inversion, while mass anomalies are parameterized through gravitationally self‐consistent fingerprints, and steric expansion through EOFs. We validate our estimates by comparing to Argo and modeling for the Indian Ocean, and by comparing total water storage change (TWSC) for the Ganges and Brahmaputra basins to the conventional GRACE approach. We find good agreement for TWSC, and reasonable agreement for steric heights, depending on the ocean region and Argo product. We ascribe differences to weaknesses of the Argo data, but we also find the inversion to be to some extent sensitive with respect to the EOFs. Finally, combining our estimates with CMIP5‐simulations, we estimate that Bay of Bengal absolute sea level may rise for additional 37 cm under the RCP4.5 scenario and 40 cm under RCP8.5 until 2050, with respect to 2005.

Separation of large scale water storage patterns over Iran using GRACE, altimetry and hydrological data

Assessing the interannual and subseasonal variabilities in water storage using multi-source soil moisture products and GRACE/GRACE-FO satellites and its applications

Recent studies show that data from the Gravity Recovery and Climate Experiment (GRACE) is promising for basin‐ to global‐scale water cycle research. This study provides varied assessments of errors associated with GRACE water storage estimates. Thirteen monthly GRACE gravity solutions from August 2002 to December 2004 are examined, along with synthesized GRACE gravity fields for the same period that incorporate simulated errors. The synthetic GRACE fields are calculated using numerical climate models and GRACE internal error estimates. We consider the influence of measurement noise, spatial leakage error, and atmospheric and ocean dealiasing (AOD) model error as the major contributors to the error budget. Leakage error arises from the limited range of GRACE spherical harmonics not corrupted by noise. AOD model error is due to imperfect correction for atmosphere and ocean mass redistribution applied during GRACE processing. Four methods of forming water storage estimates from GRACE spherical harmonics (four different basin filters) are applied to both GRACE and synthetic data. Two basin filters use Gaussian smoothing, and the other two are dynamic basin filters which use knowledge of geographical locations where water storage variations are expected. Global maps of measurement noise, leakage error, and AOD model errors are estimated for each basin filter. Dynamic basin filters yield the smallest errors and highest signal‐to‐noise ratio. Within 12 selected basins, GRACE and synthetic data show similar amplitudes of water storage change. Using 53 river basins, covering most of Earth's land surface excluding Antarctica and Greenland, we document how error changes with basin size, latitude, and shape. Leakage error is most affected by basin size and latitude, and AOD model error is most dependent on basin latitude.

The Gravity Recovery And Climate Experiment (GRACE) satellite mission recorded temporal variations in the Earth’s gravity field, which are then converted to Total Water Storage Change (TWSC) fields representing an anomaly in the water mass stored in all three physical states, on and below the surface of the Earth. GRACE provided a first global observational record of water mass redistribution at spatial scales greater than 63000 km2. This limits their usability in regional hydrological applications. In this study, we implement a statistical downscaling approach that assimilates 0.5° × 0.5° water storage fields from the WaterGAP hydrology model (WGHM), precipitation fields from 3 models, evapotranspiration and runoff from 2 models, with GRACE data to obtain TWSC at a 0.5° × 0.5° grid. The downscaled product exploits dominant common statistical modes between all the hydrological datasets to improve the spatial resolution of GRACE. We also provide open access to scripts that researchers can use to produce downscaled TWSC fields with input observations and models of their own choice.

The gravity recovery and climate experiment (GRACE) and GRACE-Follow on (FO) data provide valuable information about dynamic total water storage (TWS). The complexity of the computational process and the influence of various parameters on TWS changes are complicated in their interpretation. Principal component analysis (PCA) has been used to identify key components to amplify signals and reduce noise in observations. For this purpose, in this research, the Self-organizing map algorithm (SOM) has been used to cluster TWS in 4 categories. The results show that the western regions of Greenland and part of Antarctica are in the critical cluster and have a TWS rate of about –0.2 m/year, which indicates the melting of ice in these regions. The advantage of PCA-SOM is the easy interpretation of TWS, which reduces the impact of seasonal parameters, observation noise and measurement error, and facilitates global policy decisions in the face of climate change.

In areas dominated by Glacial Isostatic Adjustment (GIA), the free-air gravity anomaly rate can be converted to uplift rate to good approximation by using a simple spectral relation. We provide quantitative comparisons between gravity rates derived from monthly gravity field solutions (GFZ Potsdam, CSR Texas, IGG Bonn) from the Gravity Recovery and Climate Experiment (GRACE) satellite mission with uplift rates measured by GPS in these areas. The band-limited gravity data from the GRACE satellite mission can be brought to very good agreement with the point data from GPS by using scaling factors derived from a GIA model (the root-mean-square of differences is 0.55 mm yr-1 for a maximum uplift rate signal of 10 mm yr-1). The root-mean-square of the differences between GRACE derived uplift rates and GPS derived uplift rates decreases with increasing GRACE time period to a level below the uncertainty that is expected from GRACE observations, GPS measurements and the conversion from gravity rate to uplift rate. With the current length of time-series (more than 8 yr) applying filters and a hydrology correction to the GRACE data does not reduce the root-mean-square of differences significantly. The smallest root-mean-square was obtained with the GFZ solution in Fennoscandia and with the CSR solution in North America. With radial gravity rates in excellent agreement with GPS uplift rates, more information on the GIA process can be extracted from GRACE gravity field solutions in the form of tangential gravity rates, which are equivalent to a rate of change in the deflection of the vertical scaled by the magnitude of gravity rate vector. Tangential gravity rates derived from GRACE point towards the centre of the previously glaciated area, and are largest in a location close to the centre of the former ice sheet. Forward modelling showed that present day tangential gravity rates have maximum sensitivity between the centre and edge of the former ice sheet, while radial gravity rates are most sensitive in the centre of the former ice sheet. As a result, tangential gravity rates offer constraints on a two-layer mantle viscosity profile that are different from radial gravity rates, which can be exploited in future GIA studies.

Sustainable water resource management requires the assessment of hydrological changes in response to climate fluctuations and anthropogenic activities in any given area. A quantitative estimation of water balance entities is important to understand the variations within a basin. Water resources in remote areas with little infrastructure and technological knowhow suffer from poor documentation, rendering water management difficult and unreliable. This study analyzes the changes in the hydrological behavior of the Lake Chad basin with extreme climatic and environmental conditions that hinder the collection of field observations. Total water storage (TWS) from the Gravity Recovery and Climate Experiment (GRACE), lake level variations from satellite altimetry, and water fluxes and soil moisture from Global Land Data Assimilation System (GLDAS) were used to study the spatiotemporal variability of the hydrological parameters of the Lake Chad basin. The estimated TWS varies in a similar pattern as the lake water level. TWS in the basin area is governed by the lake’s surface water. The subsurface water volume changes were derived by combining the altimetric lake volume with the TWS over the drainage basin. The results were compared with groundwater outputs from WaterGAP Global Hydrology Model (WGHM), with both showing a somewhat similar pattern. These results could provide an insight to the availability of water resources in the Lake Chad basin for current and future management purposes.

<p>Every year, 4 billion cubic meters of irreversible underground water from Iran is pumped to the earth's surface, exposing Iran's freshwater supplies to indiscriminate and poorly managed withdrawals. The future basins water supply in central and northern Iran is projected to face significant challenges due to the rising demand for groundwater on the one hand, and the high rate of irrigation and overexploitation of water resources in some places, on the other. So investigating, monitoring and prediction of total terrestrial water storage modelling always plays crucial role to obtain more information about climate changes of the Iran.</p> <p>The GRACE (Gravity Recovery and Climate Experiment) and GRACE Follow-On (FO) satellite gravity missions enable global monitoring of the mass transport within the Earth’s system, leading to our understanding of the global water cycle in the climate changes of the Earth. The GRACE satellites provide signals of total terrestrial water storage (TWS) variations at different timescales. While the parametrization using GRACE gravity data has been extensively used to assess spatial-temporal changes in TWS, little effort has been made to exhibit mass variation based on regional harmonic basis function.</p> <p>This study focuses on the parametrization (especially in the higher frequency domain) for TWS variations at sub monthly and monthly time scales over Iran using LGD (Line of sight Gravity Differences) data obtained by GRACE-FO missions based on a regional harmonic basis function. We will apply regional harmonic basis functions. i.e. spherical cap harmonic (SCH) functions to provide nearly full extracting of gravitational signal over a spherical cap region that are generally underestimated by GRACE Level2 product. In particular, we analyzed whether the estimated storage component represented the TWS variations at monthly and sub-monthly temporal scales, and whether the estimated component is consistent across small to large spatial scales.</p> <p> </p>

Snow and glacier melting and accumulation are important processes of the hydrological cycle in the cryosphere, e.g., high‐mountain areas. Glaciers and snow cover respond to climate change notably over the Tibetan Plateau (TP) as the Earth's Third Pole where complex topography and lack of ground‐based observations result in knowledge gaps in hydrological processes and large uncertainties in model output. This study develops a snow and glacier melt model for a distributed hydrological model (Coupled Routing and Excess Storage model, CREST) using the Upper Brahmaputra River (UBR) basin in the TP as a case study. Satellite and ground‐based precipitation and land surface temperature are jointly used as model forcing. A progressive two‐stage calibration strategy is developed to derive model parameters, i.e., (1) snow melting processes (stage I) and (2) glacier melting and runoff generation and routing using multisource data (stage II). Stage‐I calibration is performed using the MODIS snow cover area (SCA) product and a blending snow water equivalent (SWE) product combined with partial in situ measurements. Stage‐II calibration is based on Gravity Recovery and Climate Experiment (GRACE) satellite‐derived total water storage (TWS) changes and streamflow observed at a gauging station of the lower reach of the UBR. Results indicate that the developed two‐stage calibration method provides more reliable streamflow, snow (both SCA and SWE), and TWS change simulations against corresponding observations than commonly used methods based on streamflow and/or SCA performance. The simulated TWS time series shows high consistency with GRACE counterparts for the study period 2003–2014, and overestimated melting rates and contributions of glacier meltwater to runoff in previous studies are improved to some degree by the developed model and calibration strategy. Snow and glacier runoff contributed 10.6% and 9.9% to the total runoff, and the depletion rate of glacier mass was ∼ −10 mm/a (∼ −2.4 Gt/a, Gt/a is gigaton (km3 of water) per year) over the UBR basin during the study period. This study is valuable in examining the impacts of climate change on hydrological processes of cryospheric regions and providing an improved approach for simulating more reliable hydrological variables over the UBR basin and potentially similar regions globally.

Global hydrological and land surface models are increasingly used for tracking terrestrial total water storage (TWS) dynamics, but the utility of existing models is hampered by conceptual and/or data uncertainties related to various underrepresented and unrepresented processes, such as groundwater storage. The gravity recovery and climate experiment (GRACE) satellite mission provided a valuable independent data source for tracking TWS at regional and continental scales. Strong interests exist in fusing GRACE data into global hydrological models to improve their predictive performance. Here we develop and apply deep convolutional neural network (CNN) models to learn the spatiotemporal patterns of mismatch between TWS anomalies (TWSA) derived from GRACE and those simulated by NOAH, a widely used land surface model. Once trained, our CNN models can be used to correct the NOAH‐simulated TWSA without requiring GRACE data, potentially filling the data gap between GRACE and its follow‐on mission, GRACE‐FO. Our methodology is demonstrated over India, which has experienced significant groundwater depletion in recent decades that is nevertheless not being captured by the NOAH model. Results show that the CNN models significantly improve the match with GRACE TWSA, achieving a country‐average correlation coefficient of 0.94 and Nash‐Sutcliff efficient of 0.87, or 14% and 52% improvement, respectively, over the original NOAH TWSA. At the local scale, the learned mismatch pattern correlates well with the observed in situ groundwater storage anomaly data for most parts of India, suggesting that deep learning models effectively compensate for the missing groundwater component in NOAH for this study region.