Live display program

The application of GRACE and GRACE-FO observed gridded terrestrial water storage data (TWS) often requires realistic assumptions of the data variances and covariances. Such covariances are, e.g., needed for data assimilation in various models or combinations with other data sets. The formal variance-covariance matrices now provided with the Stokes coefficients can yield such spatial variances and covariances after variance propagating them through the various post-processing steps, including the filtering, and spherical harmonic synthesis. However, a rigorous variance propagation to the TWS grids is beyond the capabilities of most non-geodetic users.

That is why we developed a new spatial covariance model for global TWS grids. This covariance model is non-stationary (time-depending), non-homogeneous (location-depending), and anisotropic (direction-depending). Additionally, it allows latitudinal wave-like correlations caused by residual striping errors. The model is tested for both GFZ RL06 Level-3 TWS data as provided via the GravIS portal (gravis.gfz-potsdam.de) and ITSG-Grace2018 GravIS-like processed Level-3 TWS data. The model parameters are fitted to empirical correlations derived from both TWS fields. Both data sets yield the same model parameters within the uncertainty of the parameter estimation.

Now, the covariance model derived thereof can be used to estimate uncertainties of mean TWS time series of arbitrary regions such as river basins. Here, we use a global basin segmentation covering all continents. At the same time, such regional uncertainties can be derived from formal variance-covariance matrices as well. To this end, the formal ITSG-Grace2018 variance-covariance matrices of the spherical harmonic coefficients are used. Thus, the modelled and formal basin uncertainties can be compared against each other globally, both spatially and temporally. Further, external validation investigates the usefulness of the basin uncertainties for applications such as data assimilation into hydrological models. Our results show a high agreement between the modelled and the formal basin uncertainties proving our approach of modelled covariance to be a suitable surrogate for the formal variance-covariance matrices.

How to cite: Boergens, E., Kvas, A., Dobslaw, H., Eicker, A., Dahle, C., and Flechtner, F.: Uncertainties of Terrestrial Water Storage Anomalies for Global Basins – A Comparison Between Modelled and Formal Covariances, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-24, https://doi.org/10.5194/gstm2020-24, 2020.

The Nubian Sandstone Aquifer System (NSAS) in northeast Africa is formed of three subbasins, the Dakhla, Kufra, and the Northern Sudan Platform subbasins. The Dakhla subbasin (DSB) receives negligible precipitation (<10 mm/yr), yet displays significant seasonal variations in GRACETWS (average: 50 mm/yr, up to 77 mm/yr) across the entire subbasin. The origin of these variations could be related to one or more of the following factors: (1) leakage out from Lake Nasser, (2) leakage in from surroundings (Kufra basin [west NSAS], Northern Sudan Platform [south NSAS], Mediterranean sea [north NSAS], and Red Sea [east NSAS], and (3) recharge and rapid groundwater flow from Lake Nasser and the northern Sudan Platform. Three approaches were used to investigate the contribution of leakage (factors 1 and 2) to the observed GRACETWS signal over the DSB subbasin: (1) forward modeling (in spherical harmonic domain) of the maximum variations in Lake Nasser levels was applied to test whether the observed seasonal variation in GRACETWS across the DSB can be accounted for by leakage from Lake Nasser alone; (2) estimate (in spherical harmonic domain) the leakage in signal using the simulated TWS from the widely applied Land Surface Model (LSM), GLDAS (Global Land Data Simulation System); and (3) apply iterative forward modeling (iterations: n=30) to reconstruct the true mass variations of GRACETWS over the DSB. Findings suggest: (1) the leakage in signal over the DSB cannot account for the observed seasonal GRACETWS patterns and neither can the leakage out from Lake Nasser; (2) the leakage out signal is centered over Lake Nasser and extends to its immediate surroundings with a maximum radius of 250 km (upper boundary of leakage error); (3) the iterative modeling indicates that the maximum leakage within the 250 km buffer zone around the lake amounted to 22.6 % of the observed GRACETWS signal; (4) minimal leakage (up to 10 mm) from northerly precipitation is observed along the northern sections (~200 km deep) of the NSAS and negligible (< 4 mm) leakage is detected over the remaining sections of the DSB; and (5) the observed seasonal variations in GRACETWS over the DSB is related to an increase in groundwater storage related to seasonal recharge from Lake Nasser and rapid groundwater flow along a network of faults, fractures, and karst topography across the entire DSB.

How to cite: Abdelmohsen, K., Sultan, M., and Save, H.: GRACE-derived seasonal variations, a key to understanding aquifer sources, recharge and groundwater flow patterns, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-50, https://doi.org/10.5194/gstm2020-50, 2020.

Space-borne observations of terrestrial water storage (TWS) are an essential ingredient for understanding the Earth's global water cycle, its susceptibility to climate change, and for risk assessments of ecosystems, agriculture, and water management. However, the complex distribution of water masses in rivers, lakes, or groundwater basins remains elusive in coarse-resolution gravimetry observations. We combine machine learning, numerical modeling, and satellite altimetry to build and train a downscaling neural network that recovers simulated TWS from synthetic space-borne gravity observations. The neural network is designed to adapt and validate its training progress by considering independent satellite altimetry records. We show that the neural network can accurately derive TWS anomalies in 2019 after being trained over the years 2003 to 2018. Specifically for validated regions in the Amazonas, we highlight that the neural network can outperform the numerical hydrology model used in the network training.

How to cite: Irrgang, C., Saynisch-Wagner, J., Dill, R., Boergens, E., and Thomas, M.: Deep learning-based recovery of high-resolution terrestrial water storage from space-borne gravimetry and altimetry, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-22, https://doi.org/10.5194/gstm2020-22, 2020.

Global warming is producing climatic changes across the world that affect in major ways the livelihood of major sectors of the world’s population. Over the past decade or two, an increase in the frequency and intensity of specific climatic phenomena (e.g., hurricanes, wet or dry periods, etc.) has been reported from many parts of the globe and is believed to be climate change-related. Over the past few years, the largest and most intense precipitation events were recorded over the Tigris and Euphrates watershed (TEW), a heavily engineered watershed (> 60 main dams) that is shared by Turkey, Iran, Syria, Saudi Arabia, and Iraq. Analysis of the Global Precipitation Climatology Project (GPCP) precipitation record over the past 40 year (1979-present) across the TEW revealed a prolonged dry period (2002- to 2017; Average Annual Precipitation [AAP]: 240 km³), followed by wet years (2018 to 2020; AAP: 425 km³). The recent extensive precipitation events during the wet period are reflected in GRACE and GRACE-FO data. Throughout the dry period there was a total decline in GRACE_TWS of 212 km³ (13.3 km³/yr) followed by an increase of 246 km³ (82 km³/yr) during the wet period.  In other words, in the past 2.5 years, the TEW more than recovered its losses during the previous 15 years. This recovery was enabled in part by the impoundment of surface water behind the many dams in the riparian countries and by infiltration of precipitation that recharged the TEW aquifers. Using radar altimetry we observe an increase in surface water levels by 8 m in Lake Ataturk, 13 m in Lake Karakaya, 1.5 m in Lake Van in Turkey, 5 m in Lake Assad in Syria, and 16 m in Lake Tharthar, and 24 m in Lake Mosul in Iraq.  These translate to a volume increase of 21.7 km³ in Turkey, 3.5 km³ in Syria, and 34 km³ in Iraq during the wet period. Using GRACE data and outputs of land surface models, we estimate that groundwater storage GRACE_TWS declined at a rate of -7 km³/yr during the dry period and increased at a rate of 60 km³/yr during the wet years.

How to cite: Sultan, M., Abdelmohsen, K., and Save, H.: GRACE a witness to the Recovery of the Tigris-Euphrates Hydrologic System, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-43, https://doi.org/10.5194/gstm2020-43, 2020.

Temporal water storage is a fundamental component of the terrestrial water cycle. Almost all precipitation falling on land is transferred via a series of short- to long-term storage locations, e.g. groundwater, to rivers, and eventually ends in the oceans or, through evapotranspiration, back in the atmosphere. The intermediate storage compartments are recharged during precipitation events and subsequently purge during phases of very little precipitation input. Methods to estimate water storage variations are often limited to specific, well-monitored locations and the findings from there are often difficult to generalize or to upscale. At the same time large scale monitoring represents an average of the entire system with very little prediction power for small areas. Thus, measures of storage from small systems can be difficult to compare to large systems and vice versa. In this recently published study (Schmidt et al., 2020) we compare three independent methods of estimating water storage variations for systems spanning over three orders of magnitude in basin area: 1) GRACE, 2) hydrograph recession curve analysis, and 3) quantifying precipitation-discharge hysteresis loops. We measured storage using all three methods for 242 watersheds in Asia spanning a size range from 10³ to 10⁶ km² and find that GRACE- derived storage correlates well with the quantification of hysteresis terms but recession curve derived dynamic storage does not correlate with hysteresis terms or GRACE-derived storage. Thus, we argue that precipitation-discharge hysteresis may be able to be scaled to GRACE-derived storage as an  independent estimate of storage to systems much smaller than the typical resolution of GRACE. Hysteresis-derived storage correlates well with mean monsoon rainfall in the upstream watershed while recession-derived dynamic storage does not. This suggests that hysteresis- and GRACE-derived storage may be input limited. In contrast, recession-derived dynamic storage does not correlate with topographic, climatic, or land cover metrics, suggesting that it may be limited by the rate at which water infiltrates into deep groundwater and then enters the river system. In addition, we find that recession-derived dynamic storage is a factor of seven lower than hysteresis-derived or GRACE derived storage. Recession-derived dynamic storage represents the annual variability in deep and saturated groundwater storage, a “leaky bucket” that is recharged from the top and “leaks” into rivers from deeper storage. The GRACE and hysteresis derived storage in turn integrates groundwater variations in addition to other storage units at or close to the Earth surface, such as snowpack, lakes, and soil moisture. These data may be able to be used to better quantify storage terms in hydrologic modeling and might help to improve GRACE data products.

Schmidt, A. H., Lüdtke, S., & Andermann, C. (2020). Multiple measures of monsoon-controlled water storage in Asia. Earth and Planetary Science Letters, https://doi.org/10.1016/j.epsl.2020.116415

How to cite: Schmidt, A., Lüdtke, S., and Andermann, C.: Multiple measures of water storage in monsoon Asia , GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-57, https://doi.org/10.5194/gstm2020-57, 2020.

Land water supply for plant growth directly links the water and carbon cycles. The abundance or shortage of water storage influences plant water consumption strategies and have important implications for ecosystem drought resistance and resilience, especially for the grassland ecosystem where water is the primary factor limiting plant production. However, plant-accessible water is rarely quantified due to the lack of regional to global scale observations of deeper water storage, and the influence of deeper water supply on plant-water relation remains unknown. In this study, we evaluate the capacity of GRACE/GRACE-FO total terrestrial water storage (TWS) estimates to capture plant-accessible water supply at depth. We use ESA CCI surface soil moisture (SM) estimates to represent shallow water storage and MODIS EVI as a proxy for grassland productivity. We calculate the inter-annual correspondence of EVI against both TWS and SM over 24 GRACE mascons covering the majority of the global grassland areas. Our results show that complementary to SM measurements, TWS provides unique information about deeper water storage limiting grassland growth. We find that the seasonal change of TWS constrains plant-accessible water storage and leads to different plant-water relations in the grassland regions across the globe.

How to cite: Velicogna, I., A Geruo, G., and Zhao, M.: GRACE/GRACE-FO observed terrestrial water storage influencing global grassland growth, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-69, https://doi.org/10.5194/gstm2020-69, 2020.

Large-scale ecological restoration (ER) has been successful in curbing land degradation and improving ecosystem services. Previous studies show that ER changes individual water flux or storage, but its net impact on total water resources remains unknown. Here we quantify ER impact on total terrestrial water storage (TWS) in the Mu Us Sandyland of northern China, a hotspot of ER practices. By integrating GRACE with other satellite observations and government reports, we construct a TWS record that covers both the pre-ER (1982-1998) and the post-ER periods (2003-2016). We observe a significant TWS depletion after ER, a substantial deviation from the pre-ER condition. This contrasts with a TWS increase simulated by an ecosystem model that excludes human interventions, indicating that ER is the primary cause for the observed water depletion. We estimate that ER has consumed TWS at an average rate of 16.6 ± 5.0 mm yr^-1 during the post-ER period, an alarming rate comparable to those caused by groundwater irrigation in California’s Central Valley and North China Plain. We further discuss whether the trend will continue under different ER strategies given a projected warming and wetting climate. Our study explores a new interdisciplinary application of GRACE/GRACE-FO in quantifying land use change impact on freshwater availability. Our findings show that ER can exert excessive pressure on regional water resources. Sustainable ER strategies require optimizing ecosystem water consumption to balance land restoration and water resource conservation.

How to cite: Velicogna, I., Zhao, M., and A Geruo, G.: Impact of ecological restoration in mainland China on the terrestrial water budget using GRACE/GRACE-FO and other data., GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-70, https://doi.org/10.5194/gstm2020-70, 2020.

A unique aspect of satellite gravimetry is its ability to quantify changes in all water stored at all depths on and beneath the land surface.  Hence, GRACE and GRACE-FO are well suited for quantifying both hydrological droughts, when terrestrial water storage (TWS) is low, and pluvial events, when TWS is high.  In this study we use GRACE and GRACE-FO data assimilation within a land surface model to fill the 1-year gap between the two missions and to replace other missing data.  We apply a cluster analysis approach to identify the locations and extents of TWS extreme events in resulting data record.  We then rank these events based on their intensity, i.e., the integral of the non-seasonal water mass anomaly over the period of the event.  In this presentation we report on the largest wet and dry events over each continent.  During the period of study, Africa, North America, and Australia each had a wet event with an intensity that exceeded 10,000 km³ * month, although the 2010-2012 event in Australia can largely be attributed to a depressed baseline TWS during the period caused by the millennial drought.  With 30 more years of data it is probable that the intensity of that drought would have been greater than the recovery and wet event during 2010-2012.  As it stands, the biggest drought event was determined to be one occurred in South America during 2015-2016, with an intensity of over 10,000 km³ * month.

How to cite: Rodell, M. and Li, B.: Water Cycle Extremes in the GRACE and GRACE-FO Data Record, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-35, https://doi.org/10.5194/gstm2020-35, 2020.

Increased water storage in the Missouri River basin as inferred from GRACE gravity [Reager et al. 2014] preceded the May–June 2011 catastrophic floods. In this study, we analyze the evolution and lateral distribution of water components in the four years before and two years after the 2011 floods (believed to occur once every 500 years). We integrate GPS measurements of solid Earth's elastic response to water change and GRACE gravity data to infer change in total water storage as a function of time in the Missouri River basin. We furthermore evaluate relative water storage and discharge in 6 major sub-basins as measured by gauging stations along the Kansas, Platte, Yellowstone, and Missouri Rivers. We aim to understand how water changes between different components (snow, soil moisture, groundwater, surface water) and how the spatial distribution of water changes in the Missouri River basin.

How to cite: Anderson, M. and Argus, D.: Accumulation and Dissipation of Water Associated with Flooding in the Missouri River Basin, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-64, https://doi.org/10.5194/gstm2020-64, 2020.

Accurate estimation of terrestrial water storage (TWS) is crucial in the characterization of the terrestrial hydrologic cycle. The launch of GRACE and GRACE Follow-On (GRACE-FO) missions provide an unprecedented opportunity to monitor the change in TWS across the globe. However, the spatial and temporal resolutions provided by GRACE/GRACE-FO are often too coarse for many hydrologic applications. Land surface models (LSMs) provide estimates of TWS at a finer spatio-temporal resolution, but most LSMs lack complete, all-encompassing physical representations of the hydrological system such as deep groundwater storage or anthropogenic influences (e.g., groundwater pumping and surface water regulation). In recent years, geodetic measurements from the ground-based Global Positioning System (GPS) network have been increasingly used in hydrologic studies based on the elastic response of the Earth’s surface to mass redistribution. This study explores the potential of improving our knowledge in TWS change via merging the information provided by ground-based GPS, GRACE, and the NASA Catchment Land Surface Model (Catchment), especially for the TWS change during an extended drought period.

Ground-based GPS observations of vertical displacement and GRACE TWS retrievals were assimilated into the Catchment LSM, respectively, using an ensemble Kalman filter (EnKF) in order to improve the estimation accuracy of TWS change. The data assimilation (DA) framework effectively downscaled TWS into its constituent components (e.g., snow and soil moisture) as well as improved estimates of hydrologic fluxes (e.g., runoff). Estimated TWS change from the open loop (OL; without assimilation) and GPS DA (i.e., using GPS-based vertical displacement during assimilation) simulations were evaluated against GRACE TWS retrievals. Results show that GPS DA improved estimation accuracy of TWS change relative to the OL, especially during an extended drought period post-2011 in the western United States (e.g., the correlation coefficient R_OL = 0.46 and R_GPSDA = 0.82 in the Great Basin). The performance of GPS DA and GRACE DA in estimating TWS constituent components and hydrologic fluxes were evaluated against in situ measurements. Results show that GPS DA improves snow water equivalent (SWE) estimates with improved R values found over 76% of all pixels that are collocated with in situ stations in the Great Basin. The findings in this study indicate the potential use of GPS DA and GRACE DA for TWS characterization. Both GRACE and ground-based GPS provide complementary TWS change information, which helps correct for missing physics in the LSM. Additionally, this study provides motivation for a multi-variate assimilation approach to simultaneously merge both GRACE and ground-based GPS into an LSM to further improve modeled TWS and its constituent components.

How to cite: Yin, G., Forman, B., and Wang, J.: Exploration of Terrestrial Water Storage Characterization via Assimilation of Ground-based GPS Observations of Vertical Displacement and GRACE TWS Retrievals , GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-2, https://doi.org/10.5194/gstm2020-2, 2020.

This study explores multi-sensor, multi-variate data assimilation (DA) using synthetic GRACE terrestrial water storage (TWS) retrievals and synthetic AMSR-E passive microwave brightness temperature spectral differences (dTb) in order to improve estimates of snow water equivalent (SWE), subsurface water storage, and TWS over snow-covered terrain. In order to better assess the performance of joint assimilation, a series of synthetic twin experiments, including the Open Loop (model-only run), single-sensor DA (GRACE TWS DA or AMSR-E dTb DA), and simultaneous assimilation of GRACE TWS and AMSR-E dTb (a.k.a., dual DA), are conducted. The baseline assimilation of GRACE TWS retrievals is further modified using a physically-informed approach during the application of the analysis increments. A well-trained support vector machine (SVM) is used as the observation operator during the assimilation of AMSR-E dTb observations.

Results suggests that the single-sensor GRACE TWS DA experiment using the physically-informed update approach leads to statistically significant improvements in SWE, subsurface water storage, and TWS estimation. The application of increments based on the presence (or absence) of snowmelt further discretizes TWS into SWE and subsurface water storage more accurately, and hence, effectively enhances TWS vertical resolution. Similarly, the single-sensor AMSR-E dTb DA approach yields improvements in SWE, subsurface water storage, runoff, and TWS estimation. However, the efficacy of SVM-based PMW dTb DA is limited by the fundamentally ill-posed nature of SWE estimation using PMW radiometry coupled with limited controllability of the SVM-based observation operator during deep, wet snow conditions. Furthermore, the PMW dTb assimilation approach (i.e., multiple observations assimilated daily) can lead to SWE ensemble collapse, which can ultimately degrade the SWE estimates.

Dual assimilation, in general, maintains the benefits introduced by the single sensor assimilation of GRACE TWS retrievals and AMSR-E dTb observations. Dual DA yields the best TWS estimates (in terms of smallest RMSE) and the most reasonable ensemble spread of subsurface water storage compared to the OL and single sensor DA experiments. The assimilation of dTb observations significantly reduces the SWE ensemble spread while the assimilation of TWS retrievals reduces the ensemble spread of subsurface water storage. The assimilation of TWS helps mitigate the SWE ensemble collapse often caused by daily assimilation of dTb's, and hence, improves the SWE ensemble reliability. The assimilation of dTb observations, in general, removes snow mass whereas the assimilation of TWS retrievals, in general, adds snow mass to the system, which can, at times, lead to SWE degradation given this juxtaposed, contradictory behavior. These synthetic experiments provide valuable insights into the assimilation of “real-world” GRACE / GRACE-FO TWS retrievals and AMRS-E / AMSR-2 dTb observations in order to better characterize terrestrial freshwater storage across regional scales.

How to cite: Wang, J. and Forman, B.: Improved Terrestrial Snow Mass via Multi-sensor Assimilation of Synthetic GRACE Terrestrial Water Storage Retrievals and Synthetic AMSR-E Brightness Temperature Spectral Differences, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-11, https://doi.org/10.5194/gstm2020-11, 2020.

Reliable predictions of terrestrial water storage (TWS) changes for the next couple of years would be extremely valuable for, e.g., agriculture and water management. In contrast to long-term projections of future climate conditions, so-called decadal predictions do not depend on prescribed CO₂ scenarios but provide unconditional forecasts similar to numerical weather models. Therefore, opposed to climate projections, decadal predictions (or hindcasts, if run for the past) can directly be compared to observations. Here, we evaluate decadal hindcasts of TWS related variables from an ensemble of 5 coupled CMIP5 climate models against a TWS data set based on GRACE satellite observations.

Since data from the CMIP5 models and GRACE is jointly available in only 9 years, we access a GRACE-like reconstruction of TWS derived from precipitation and temperature data sets (Humphrey and Gudmundsson, 2019), which expands the analysis time-frame to 41 years. The skill of the decadal hindcasts is assessed by means of anomaly correlations and root-mean-square deviations (RMSD) for the yearly global average and aggregated over different climate zones. Furthermore, we compute global maps of correlation and RMSD.

We find that at least for the first two prediction years the decadal model experiments clearly outperform the classical climate projections, regionally even for the third year. We can thereby demonstrate that the observation type “terrestrial water storage” as available from the GRACE and GRACE-FO missions is suitable as additional data set in the validation and/or calibration of climate model experiments.

How to cite: Jensen, L., Eicker, A., Stacke, T., and Dobslaw, H.: Evaluation of land water storage prediction skill in CMIP5 decadal hindcasts by means of a GRACE-based data set, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-5, https://doi.org/10.5194/gstm2020-5, 2020.

The GRACE satellite mission and its follow-on, GRACE-FO, have provided unprecedented opportunities to quantify the impact of climate extremes and human activities on total water storage at large scales. The approximately one-year data gap between the two GRACE missions needs to be filled to maintain data continuity and maximize mission benefits. There is strong interest in using machine learning (ML) algorithms to reconstruct GRACE-like data to fill this gap. So far, most studies attempted to train and select a single ML algorithm to work for global basins. However, hydrometeorological predictors may exhibit strong spatial variability which, in turn, may affect the performance of ML models. Existing studies have already shown that no single algorithm consistently outperformed others over all global basins. In this study, we applied an automated machine learning (AutoML) workflow to perform GRACE data reconstruction. AutoML represents a new paradigm for optimal model structure selection, hyperparameter tuning, and model ensemble stacking, addressing some of the most challenging issues related to ML applications. We demonstrated the AutoML workflow over the conterminous U.S. (CONUS) using six types of ML algorithms and multiple groups of meteorological and climatic variables as predictors. Results indicate that the AutoML-assisted gap filling achieved satisfactory performance over the CONUS. For the testing period (2014/06–2017/06), the mean gridwise Nash-Sutcliffe efficiency is around 0.85, the mean correlation coefficient is around 0.95, and the mean normalized root-mean square error is about 0.09. Trained models maintain good performance when extrapolating to the mission gap and to GRACE-FO periods (after 2017/06). Results further suggest that no single algorithm provides the best predictive performance over the entire CONUS, stressing the importance of using an end-to-end workflow to train, optimize, and combine multiple machine learning models to deliver robust performance, especially when building large-scale hydrological prediction systems and when predictor importance exhibits strong spatial variability.

How to cite: Sun, A., Scanlon, B., Save, H., and Rateb, A.: Reconstruction of GRACE Total Water Storage Through Automated Machine Learning, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-53, https://doi.org/10.5194/gstm2020-53, 2020.

Terrestrial water storage (TWS) data derived from past Gravity Recovery and Climate Experiment (GRACE; April 2002–June 2017) and current GRACE-Follow On (GRACE-FO; June 2018–present) missions provide insights into mass transport within, and between, different Earth’s systems (e.g., atmosphere, oceans, groundwater, and ice sheets). However, there are currently temporal gaps within GRACE-derived TWS record (20 months) and between GRACE and GRACE-FO missions (11 months), within GRACE-FO-derived TWS record (2 months), and similar gaps could be experienced between GRACE-FO and GRACE-II missions. In this study, we compare the performance of different data-driven techniques in filling TWS gaps for 62 global watersheds. Additionally, these techniques are being applied to reconstruct TWS globally on a grid scale (1° × 1°). We used artificial neural networks (ANNs), support vector machines (SVMs), and multiple linear regression (MLR) models to predict TWS data (04/2002 – 03/2020) based on the knowledge of relevant climatic datasets such as rainfall, temperature, evapotranspiration, vegetation indices, climate indices. The performance of the developed models was evaluated using several standard measures such as the root mean square error (RMSE), correlation coefficient (R), and Nash-Sutcliff efficiency coefficient (NSE). Our preliminary results indicate: (1) ANN models show higher performance over the examined watersheds compared to the other models (RMSE: 5.20; R: 0.93; NSE: 0.88), (2) the performances of ANN, MLR, and SVM models depend mainly on the nature of factors that control TWS in each of the examined hydrologic systems, and (3) higher model performance is achieved when the model input data were further spectrally decomposed. Results of our research could be used to validate GRACE-FO datasets. Our research will promote additional and improved use of GRACE products by the scientific community, end-users, and decision makers by providing a continuous uninterrupted TWS record from GRACE and GRACE-FO missions.

How to cite: Ahmed, M., Gyawali, B., and Wiese, D.: Applications of Data‐Driven Techniques in Filling Temporal Gaps Within and Between GRACE and GRACE-FO Records, GRACE/GRACE-FO Science Team Meeting 2020, online, 27–29 Oct 2020, GSTM2020-56, https://doi.org/10.5194/gstm2020-56, 2020.