Abstract/Summary

Globally, flash floods are one of the costliest natural hazards for property damage and loss of life. The low accuracy of flash flood forecasts beyond a few hours limits their use in early warning systems. This study aims to compare the performance of ECMWF raw ensemble (ENS) and ecPoint rainfall forecasts. These two systems are profoundly different because the first provides forecasts at a grid-scale while the second provides forecasts at a point-scale that mirror point observations such as rain gauges. ecPoint rainfall forecasts have been shown to provide better guidance than ENS in predicting extreme (localized) rainfall up to 10 days ahead. Hence, this study assesses whether ecPoint also enhances ENS performance in identifying areas at risk of flash floods up to medium-range leads. A one-year objective verification was conducted for flash flood events in Ecuador. The country’s varied climate and the existence of a comprehensive database describing the occurrence of flash flood events made Ecuador an attractive site for the verification analysis. Knowing the magnitudes of flash-flood-triggering rainfall events is imperative to carry out the objective verification analysis. The authors did not have this information at their disposal, and suitable rainfall observations were not available to estimate the magnitudes of flash-flood-triggering rainfall events. This study proposes a methodology that uses short-term ecPoint rainfall forecasts as a proxy for point rainfall observations to define the magnitude of flash-flood triggering rainfall events when suitable rainfall observations are unavailable. Due to the probabilistic nature of the ecPoint output, this approach simulates a very high-density observational network in data-scarce regions. From a general comparison of the rainfall thresholds defined using short-range ecPoint rainfall forecast and those computed from a low-density observational network, overall, the flash-flood-triggering rainfall thresholds seem reasonable, but they might be slightly underestimated in the coastal region and overestimated in the Andean region. The verification results suggest that ecPoint generally provides better guidance than ENS in identifying areas at risk of flash floods when rainfall originates from small-scale convective systems. In these cases, ENS might completely miss the flash flood event, while ecPoint was able to anticipate it. When rainfall originates from large-scale convective systems, ecPoint and ENS performances are comparable, showing that ENS can predict areas at risk of flash floods around the globe up to medium-range leads under certain weather conditions. However, ecPoint provides better guidance on the magnitude of the flash-flood-triggering rainfall events. The findings of this study have the potential to significantly improve early warning systems, aid decision-makers, and enhance emergency preparedness worldwide against flash floods. This would ultimately contribute to better mitigation of the devastating impacts of flash floods on communities and infrastructure.