Benchmarking global hydrological and land surface models against GRACE in a medium-sized tropical basin

Abstract

Abstract. The increasing reliance on global models to address climate and human stresses on hydrology and water resources underlines the necessity for assessing the reliability of these models. In river basins where availability of gauging information from terrestrial networks is poor, models are increasingly proving to be a powerful tool to support hydrological studies and water resources assessments (WRA). However, the lack of in situ data hampers rigorous performance assessment, particularly in tropical basins where discordance between global models is considerable. Remotely sensed data of the terrestrial water storage obtained from the Gravity Recovery and Climate Experiment (GRACE) satellite mission can provide independent data against which the performance of such global models can be evaluated. However, how well GRACE data represents the dynamics of terrestrial water storage depends on basin scale and hydrological characteristics. Here we assess the reliability of six global hydrological models (GHMs) and four land surface models (LSMs) available at two resolutions. We compare the dynamics of modelled Total Water Storage (TWS) with TWS derived from GRACE data over the Magdalena–Cauca basin in Colombia. This medium-sized tropical basin has a well-developed gauging network when compared to other basins at similar latitudes, providing unique opportunity to contrast modelled TWS and GRACE data across a range of scales. We benchmark monthly TWS changes from each model against GRACE data for 2002–2014, evaluating monthly variability, seasonality, and long-term variability trends. The TWS changes are evaluated at basin level, as well as for selected sub-basins with decreasing basin size. We find that the models poorly represent TWS for the monthly time series, but they improve in representing seasonality and long-term variability trends. The high-resolution GHM World-Wide Resources Assessment (W3RA) model forced by the Multi-Source Weighted Ensemble Precipitation (MSWEP) is most consistent in providing the best performance at almost all basin scales, with higher-resolution models generally outperforming lower-resolution counterparts. This is, however, not the case for all models. Results highlight the importance of basin scale in the representation of TWS by the models, as with decreasing basin area, we note a commensurate decrease in the model performance. A marked reduction in performance is found for basins smaller than 60 000 km2. Although uncertainties in the GRACE measurement increase for smaller catchments, the models are clearly challenged in representing the complex hydrological processes of this tropical basin, as well as human influences. We conclude that GRACE provides a valuable dataset to benchmark global simulations of TWS change, in particular for those models with explicit representation of the internal dynamics of hydrological stocks, offering useful information for continued model improvement in the representation of the hydrological dynamics in tropical basins.

Benefit of Multivariate Model Calibration for Different Climatic Regions

“…The larger uncertainty of GRACE data for smaller catchments is often used as an argument for working with large catchments (Bai et al., 2018; Liu et al., 2022), and recent work by Bolaños Chavarría et al. (2022) showed that for many models it becomes challenging to reproduce total water storage if basins are smaller than 60,000 km 2 . Independent of catchment area, work with Australian catchments showed that the AWRA model (Australian Water Resources Assessment system) had most difficulties in reproducing total water storage in western and central Australia (Van Dijk et al., 2011), which is in line with our findings.…”

Parameter transferability of a distributed hydrological model to droughts

“…Here, the model properly represented ET and TWSA seasonality and long-term variability, while model performances slightly decreased for their deviations from seasonality, coherently with previous literature. Bolaños Chavarría et al (2022) for instance showed that a set of global hydrological and land surface models well represented TWSA seasonality and long-term variability in a tropical river basin in Colombia, but not the TWSA monthly time series that account for the deviations from seasonality. However, model capabilities in simulating TWSA anomalies were comparable during moderate droughts and a severe drought.…”

“…This highlights that evaluating hydrological models against multiple hydrological fluxes and states may represent a way to analyse causes of poor model transferability, verify model internal consistency and so move towards more robust modelling (Guo et al, 2017). Today ET and TWSA remote sensing-based products can be particularly useful for model evaluation, especially for distributed models (Rakovec et al, 2016b;Hulsman et al, 2021;Bolaños Chavarría et al, 2022) as they allow to check also their spatial representativeness. Nonetheless, assessment of model transferability to severe droughts using independent and spatially distributed ET and TWSA remote sensing-based products is still rare.…”

Hydrological model skills change with drought severity; insights from multi-variable evaluation

“…This indicates that evaluating hydrological models against multiple hydrological fluxes and states represents a way to analyze causes of poor model performances and hence move towards more robust modelling [31]. ET and TWS remote sensing-based products can be particularly useful for distributed model evaluation [32,33] as they allow to check also the spatial representativeness of models.…”