Skill of seasonal flow forecasts at catchment-scale: an assessment across South Korea

Abstract

Abstract. Recent advancements in numerical weather predictions have improved their forecasting performance at longer lead times for several months. As a result, seasonal weather forecasts, providing predictions of weather variables for the next several months, have gained significant attention from researchers due to their potential benefits for water resources management. Many efforts have been made to link seasonal weather forecasts with Seasonal Flow Forecasts (SFFs) using diverse hydrological models. However, generating SFFs with good skill at finer scales such as catchment remain challenging, hindering their application in practice and adoption by water managers. Consequently, water management decisions, not only in South Korea but also in many other countries, continue to rely on worst-case scenarios and the conventional Ensemble Streamflow Prediction (ESP) method. This study examines the potential of SFFs in South Korea at a catchment-scale. The analysis was conducted across 12 operational reservoir catchments of various size (from 59 to 6648 km2) over a last decade (2011–2020). Seasonal weather forecasts data (precipitation, temperature and evapotranspiration) from the ECMWF (European Centre for Medium-Range Weather Forecasts, system5) is used to drive a Tank model (conceptual hydrological model) to generate the flow ensemble forecasts. The actual skill of the forecasts is quantitatively evaluated using the Continuous Ranked Probability Skill Score (CRPSS), and it is probabilistically compared with ESP, which is the most popular forecasting system. Our results highlight that precipitation is the most important variable in determining the skill of SFFs, while temperature also plays a key role during the dry season in snow-affected catchments. Given the coarse resolution of seasonal weather forecasts, a linear scaling method to adjust the forecasts is applied, and it is found that bias correction is highly effective in enhancing the skill of SFFs. Furthermore, bias corrected SFFs showed higher skill than ESP up to 3 months ahead, and it was particularly evident during abnormally dry years. To facilitate future applications to other regions, freely available Python packages for analysing seasonal weather and flow forecasts have been made accessible.