Global Evaluation of Runoff Simulation From Climate, Hydrological and Land Surface Models

Recent advances in global hydrological modeling yield many global runoff data sets that are extensively used in global hydrological analyses. Here, we provide a comprehensive evaluation of simulated runoff from 21 global models, including 12 climate models from CMIP6, six global hydrological models from the Inter-Sectoral Impact Model Inter-Comparison Project (ISMIP2a) and three land surface models from the Global Land Data Assimilation System (GLDAS), against observed streamflow in 840 unimpaired catchments globally. Our results show that (a) no model performs consistently better in estimating runoff from all aspects, and all models tend to perform better in more humid regions and non-cold areas; (b) the interannual runoff variability is well represented in ISIMIP2a and GLDAS models, and no model performs satisfactorily in capturing the annual runoff trend; (c) the runoff intra-annual cycle is reasonably captured by all models yet an overestimation of intra-annual variability and an early bias in peak flow timing are commonly found; and (d) model uncertainty leads to a larger uncertainty in runoff estimates than that induced by forcing uncertainty in ISIMIP2a, and model uncertainty in GLDAS is larger than that in ISIMIP2a. Finally, we confirm that the multi-model ensemble is an effective way to reduce uncertainty in individual models except for CMIP6 regarding mean annual magnitude and annual runoff trend. Overall, our findings suggest that assessments/projections of runoff changes based on these global outputs contain great uncertainties and should be interpreted with caution, and call for more advanced, observation-guided ensemble techniques for better large-scale hydrological applications.

Automated summary

Recent advances in global hydrological modeling have generated multiple global runoff datasets used extensively in global hydrological analyses. This study evaluates the simulated runoff from 21 global models against observed streamflow in 840 catchments globally and finds that model performance varies across regions and aspects of runoff estimation. The study highlights the need for caution in interpreting assessments/projections of runoff changes based on these models due to uncertainties and calls for advanced observation-guided ensemble techniques for better large-scale hydrological applications.