Predicting flash floods in Ecuador and beyond

"ecPoint" is a statistical post-processing technique that anticipates sub-grid variability and biases in numerical weather prediction (NWP) model outputs, de facto downscaling them from grid-box to point-scale. ecPoint-Rainfall is the branch of the ecPoint family of products that post-processes ECMWF ensemble (ENS) rainfall forecasts. Global verification over a 1-year period has shown that, versus rain gauge observations, ecPoint-Rainfall provides more reliable and skilful rainfall forecasts than raw ENS (up to 10-day lead times and especially in case of extremes, e.g. rainfall >= 50 mm/12h). 

Flash flood forecasting could be a natural downstream application for ecPoint-Rainfall. One of this field's challenges is indeed finding reliable and skilful forecasts for localized rainfall extremes, which tend to be the primary cause of flash flood events. Radar-derived rainfall estimates can feed into nowcasting systems (which provide forecasts with lead times up to a few hours), whilst km-scale NWP models can provide reasonable guidance for localized extreme rainfall for slightly longer leads (e.g. up to 24 hours). However, such small lead times limit the mitigating actions that end-users can take. ecPoint-Rainfall targets longer leads.

ecPoint-Rainfall forecasts have been verified against flash flood observations in Ecuador to understand whether they can better predict flash floods compared to other medium-range rainfall forecasts (e.g. raw ECMWF ENS). It will be shown that, in the context of flash flood reports, in regions dominated by small scale convective systems (e.g. Ecuador's Andean region), ecPoint-Rainfall outperforms ENS. In contrast, in areas dominated by large-scale convective systems (e.g. coastal areas of Ecuador), the performance of the two forecasting systems is comparable. This talk will discuss these findings and how they are helping with the creation of global medium-range, ecPoint-Rainfall-based flash flood warnings.