[A-HW24] Human- and Climate-induced variability in water cycle and (sub)surface water resources

[AHW24-02] How well do generic reservoir operation schemes in global hydrological models reproduce storage time series?

*Emmanuel Okiria¹, Naota Hanasaki¹ (1.National Institute for Environmental Studies)

Keywords:ISIMIP, H08, WaterGAP, MIROC-INTEG-LAND, DAHITI, GRSAD

Global Hydrological Models (GHMs) employ generic reservoir operation algorithms to simulate reservoir storage. However, model validation is hindered by limited spatial coverage of published in situ global reservoir data. Fortunately, advancements in satellite remote sensing of inland waterbodies have yielded several databases of observed storage time series, e.g., DAHITI, Hydroweb, GREALM, GRS and GRSAD, offering new opportunities to validate GHM predictions. A few studies, limited to the USA, have validated and intercompared GHM simulations of reservoirs. Here, we conducted a validation and an intercomparison of storage simulations of 18 globally distributed reservoirs: These are reservoirs for which there is an intersection among available in situ, satellite data and GHM simulations. The GHMs contributed to phase 3 of the Intersectoral Impact Model Intercomparison Project (ISIMIP). Firstly, we analysed the performance of storage simulations by 3 GHMs (H08, WaterGAPv2.2e and MIROC-INTEG-LAND), each forced by 3 distinct climate forcings (GSWP3-W5E5, CR20v3-W5E5, and 20CRv3-ERA5), against in situ data. Secondly, using GHM predictions as a benchmark, we confirmed if indeed satellite-derived storage time series exceeded its (benchmark’s) performance using the NSE skill score. Best overall model performance was in the order; H08 > WaterGAPv2.2e > MIROC-INTEG-LAND, with average correlation co-efficient ranks of 1.82, 2.05 and 2.12 respectively. Much as GHMs generally reproduced the phase of the normalised storage time series quite well, they were unable to reproduce the absolute storage. It was also apparent that for reservoirs whose storage capacity was < 1 km3, predictions by the MIROC-INTEG-LAND showed minimal to no seasonality at all. Overall, implementation of reservoir operation rules in GHMs has room for improvement of prediction skill. Satellite-derived storage outperformed the GHM simulations in all but a few cases, although they also showed biases when compared with in situ data. Given their broader spatial coverage (over in situ data), satellite-derived storage can be used to validate GHM storage predictions. However, currently, they can only be used to detect change in the reservoir storage volume.