Abstract
Heavy precipitation episodes are particularly hazardous in mid-sized alpine catchments where runoff tends to increase rapidly after strong rainfall, leaving limited time for warning. A high-quality, unbiased forecast of heavy precipitation with long enough lead times (2–5 days) and an adequate spatial resolution is thus crucial for decision makers. However, at present, weather forecast models are sometimes still too coarsely resolved for such catchments or limited to short lead times. Here we present a computationally cheap post-processing approach for operational applications to bias-correct deterministic heavy precipitation forecasts at medium-range lead times. We assessed forecast performance of uncorrected ECMWF Integrated Forecasting System (IFS) high-resolution (HRES) and ensemble median run (ENS) forecasts with lead times of 2 and 5 days between 2010 and 2019 and evaluated the improvement after interpolation and bias correction. The study is based on observations from 787 meteorological stations, with a focus on areal precipitation in three medium-sized Swiss catchments (Emme, Simme, and Vispa). We found a significantly higher performance of the HRES forecast compared to ENS, especially for a lead time of 5 days. The post-processing approach that we present in this study removes large biases, and lowers false alarm rates. Nonetheless, improvements may be limited for single heavy precipitation events where the model considerably underestimates daily precipitation. Also, a lower false alarm rate can be at the cost of lower hit rates. Quantile mapping is clearly limited for short term prediction in the cases where the error does not substantially exceed natural variability. Despite these limitations, we conclude that bias-corrected forecasts can help improving forecast credibility, which is a key element for decision makers to take actions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have