Global scale predictability of hydrological drought: evaluating the skill of the GloFAS - Copernicus EMS sub-seasonal to seasonal forecast of river discharge

Major impacts associated to hydrological droughts are often neglected in early warning systems. Extensive research in hydrological drought forecasting is crucial to develop an effective early warning strategy. This work aims at quantifying the sub-seasonal to seasonal predictability of these extreme hydroclimatic events globally, by evaluating the skill of the Global Flood Awareness System (GloFAS), as part of the Copernicus Emergency Management System (CEMS). Two river discharge datasets for the period 1991-2020 from the LISFLOOD hydrological model were used, based on: 1) reanalysis (ERA5) forcings, and 2) seasonal forecast (SEAS5). River discharge values were converted into anomalies, namely the Standardized Streamflow Index (SSI), at three-time horizons (1, 3, and 6 months ahead). The skill metrics computed between the SSI reanalysis (reference) and the forecasts were the Pearson correlation coefficient (r), the Gilbert Skill Score (GSS), and the Heidke Skill Score (HSS). Moreover, the signal-to-noise ratio (SNR) of the ensemble forecast was used as a complementary metric to quantify the skill. The study evaluated the overall forecast predictability for the full year, as well as seasonal and spatiotemporal differences in the predictability and the effects of initial conditions. On average, forecast skill is higher for 1 and 3 months ahead (r= 0.81 and r= 0.70, respectively) compared to 6 months ahead (r= 0.61), with similar results in terms of spatial patterns. Seasonal differences in predictability can be well explained by average river discharge seasonality, with highest skill when river discharge is low. The forecast skill spatial patterns indicate a strong dependency on the inter-annual variability of initial conditions and precipitation, especially in summer and spring-summer seasons for the former and in winter and autumn-winter for the latter. Overall, high skill is associated with high SNR, suggesting that SNR could be used as a proxy variable for forecasting skill in operational applications. The results underline the potential of the evaluated sub-seasonal to seasonal forecast for hydrological drought predictions, suggesting a potentially successful implementation as a product as part of the CEMS Global Drought Observatory (GDO) system.

Acknowledgements:
This work is funded by the European Union, under the HORIZON-CL4-2023-SPACE-01 project “Strengthening Extreme Events Detection for Floods and Droughts” (SEED-FD), grant no. 101135110.