Catchment memory explains hydrological drought forecast performance

Skillful Drought Early Warning Systems (DEWSs) to predict drought a few months in advance are of utmost importance to reduce the impacts of the drought hazard. Previous studies on meteorological drought forecasts e.g. using the standardized precipitation index (SPI) show that drought can be sufficiently predicted up to 1-3 months ahead. The skill of hydrological drought forecasts e.g. using the standardized runoff index (SRI) and standardized ground index (SGI), on the other hand, is even 2-3 months higher than the meteorological ones. The high skill in hydrological drought forecasts is anticipated coming from the catchment storage/memory (e.g. lakes, soils, groundwater) that pools, attenuates, and lengthens the effect of the driving forces (i.e. precipitation). Yet, the importance of catchment memory in explaining hydrological drought forecast performance has not been studied. Here, we have conducted a pioneering study that investigates the importance of catchment memory on the forecast performance of streamflow drought across Europe. We identified streamflow drought using the Standardized Streamflow Index (SSI). The observed and forecasted streamflow droughts at major European rivers were derived from the streamflow data obtained from the European Flood Alert System (EFAS) driven by observed and forecasted weather data. Catchment memory was derived from the Baseflow Index (BFI) and the groundwater Recession Coefficient (gRC), which through the streamflow, give information on the catchment memory. Performance of streamflow drought forecasts was evaluated using the Brier Score (BS) for rivers across Europe. Results show that the use of higher accumulation periods in the SSI (e.g. SSI-3) forecasts improves forecast performance. The performance is even higher for catchment that has large memory. We found that BS is negatively correlated with BFI, meaning that rivers with high BFI (large memory) yield better drought prediction (low BS). A significant positive correlation between gRC and BS demonstrates that catchments slowly releasing groundwater to streams (low gRC), i.e. large memory, generates higher drought forecast performance. The higher performance of hydrological drought forecasts in catchments with relatively large memory (high BFI and low gRC) implies that Drought Early Warning Systems have more potential to be implemented there and will appear to be more useful.