Enhancing streamflow ensemble forecasting: evaluating pre- and post-processing strategies for a spatially distributed model

Despite recent advances, streamflow ensemble forecasting systems often suffer from unreliability and under-dispersion. Improving flood forecasting quality benefits decision-makers and the public. Statistical processing methods can enhance raw forecasts. Therefore, identifying the most effective scenario among pre-processing, post-processing, or a combination of both is crucial. While most studies have focused on lumped models, more comprehensive analysis is needed for systems employing spatially distributed models. This study evaluates scenarios involving the pre-processing of both temperature and precipitation forecasts and the post-processing of streamflow forecasts in a streamflow ensemble forecasting system with a spatially distributed hydrological model. It also explores the impact of incorporating flood events and the length of the training data on the performnace of pre- and post-processing approaches through various training and validation strategies. Weighted Ensemble Dressing (WED) is coupled with two bias correction methods: Cumulative Distribution Function Matching (CDFM) and Event Bias Correction (EBC). The forecast assessment covers lead times of 1–5 days for the au Saumon watershed in southern Quebec, Canada. Although pre-processing alone marginally improves raw forecasts, it fails to fully compensate for bias in under-dispersed ensemble forecasts. The combined pre- and post-processing approach proves superior in forecast skill and reliability to other scenarios. These findings align with existing literature, reinforcing the validity of the applied methodologies within current hydrological research. Both bias correction methodsperform well with WED; CDFM is more reliable for short lead times, while EBC excels at longer lead times. Notably, incorporating flood events and optimizing data length further improve processing performance, highlighting the importance of effective data strategies in streamflow forecasting.