High-Resolution, Multi-Model Hydrological Seasonal Forecasting for Water Resources Management over Europe

To support the decision-making process at the seasonal time scale, ensembles of hydrological forecasts with a high temporal and spatial resolution are required to provide the level of information needed by water managers. So far high-resolution, multi-model, ensemble seasonal forecasts have been unavailable due to 1) lack of availability in meteorological seasonal forecasts, 2) the coarse temporal resolution of meteorological seasonal forecasts, requiring temporal downscaling, 3) lack of consistency between observations and seasonal forecasts, requiring bias-correction.

As part of the EDgE (End-to-end Demonstrator for improved decision making in the water sector in Europe) project commissioned by ECMWF under the EU Copernicus Climate Change Service (C3S) initiative, we have created a unique dataset of hydrological seasonal forecasts derived from four global climate models with daily reforecasts from 1993-2010 going out 6 months. The seasonal models are: ECMWF-S4 (with 15 ensembles), LFPW (15 ensembles), CanCM3 (10 ensembles), and GFDL-FLOR (12 ensembles). Their forecast are used to force four global hydrological models (PCR-GLOBWB, VIC, mHM, Noah-MP), with an ESP forecast is used as a benchmark. The forecasts provide a daily temporal resolution at 5-km spatial resolution and are bias corrected against E-OBS meteorological observations. Consistency in the LSM parameterization (e.g. soil properties and vegetation) ensures consistency in the static fields of the hydrological forecasts. The forecast results are communicated to stakeholders via Sectoral Climate Impact Indicators (SCIIs) that have been created in collaboration with the end-user community of the EDgE project. The MME set results in 208 forecasts at any given day over Europe, for each SCII of which there are 7 related to seasonal hydrological forecasts. This MME set is used to assess forecast uncertainty and skill using metrics such as the Brier Score, Reliability Diagrams, Probability of Detection, Probability of False Detection and False Alarm Rate. The results are made available to users via a user-friendly dashboard and web interface.

Results show that skillful discharge forecasts can be made throughout Europe 3 months in advance, with predictability up to 6 months for Northern Europe due to the impact of snow. Drought forecasts have high skill over Europe due to the large spatial and temporal extent of drought events that is picked up by the seasonal forecasts. The predictability of soil moisture is limited to the first three months, due to the significant impact of precipitation and the short memory in the initial conditions (only for the first month). The groundwater predictability surpasses 6 months throughout Europe, with the lowest forecast skill for western Europe. We observe that the multi-LSM system provides added value when we want to quantify the forecast reliability and increase the consistency of the forecast system. The SCIIs, that we have developed in this project prove to be a good way to communicate the uncertainty in the seasonal forecasts and summarize the forecast results, including uncertainty.

Overall the new system provides an unprecedented ensemble for seasonal forecasts with significant skill over Europe to support water management. The consistency in both the GCM forecasts and the LSM parameterization ensures a stable and reliable forecast framework and methodology, even if additional GCMs or LSMs are added in the future.