Computer-generated weather forecasts divide the Earth’s surface into gridboxes, each currently spanning about 400 km2, and predict one value per gridbox. If weather varies markedly within a gridbox, forecasts for specific sites inevitably fail. Here we present a statistical post-processing method for ensemble forecasts that accounts for the degree of variation within each gridbox, bias on the gridbox scale, and the weather dependence of each. When applying this post-processing, skill improves substantially across the globe; for extreme rainfall, for example, useful forecasts extend 5 days ahead, compared to less than 1 day without post-processing. Skill improvements are attributed to creation of huge calibration datasets by aggregating, globally rather than locally, forecast-observation differences wherever and whenever the observed “weather type” was similar. A strong focus on meteorological understanding also contributes. We suggest that applications for our methodology include improved flash flood warnings, physics-related insights into model weaknesses and global pointwise re-analyses.

<p>Accurate predictions of heavy and intense rainfall are vital for impact-based forecasting that can be essential for mitigating the significant damage and loss of life across the globe. However, producing reliable forecasts capable of capturing the rainfall values is challenging in complex mountain terrain due to the forecast uncertainty and computational cost especially in data-scarce regions. Central Asia is one of these regions, where extreme rainfall leads to flash floods, landslides and debris flows in the mountains and foothills. The risk of these events increases with global warming, and the early warning systems based on reliable forecasts are particularly important to manage the risk in the region and adapt to climate change.</p><p>In this study, we have evaluated and compared the skills of two probabilistic forecasts developed by the European Centre for Medium-Range Weather Forecasts (ECMWF): standard Ensemble Forecasts (ENS) which consists of an ensemble of 51 members and ecPoint Rainfall produced by statistical post-processing of the ENS and delivers probabilistic forecasts of rainfall totals for points within a model gridbox (18 km resolution) that can be particularly useful in the mountains. Skills of both forecasts were assessed in relation to the forecast of debris flows in Central Asia.</p><p>Both forecast products were verified against SYNOP (surface synoptic observations) data for stations over Central Asia, mainly for the debris flow season (March-October) in 2022. In this case, two popular verification methods were used: Brier Score and Receiver Operating Characteristics (ROC) diagram for the exceedance of precipitation thresholds of 1 mm, 10 mm and 25 mm.</p><p>Verification trials over the 2022 debris flow season in Central Asia show that the performance of ecPoint Rainfall depending on the forecast lead-time can be a good proxy for the range of point rainfall values to define the warning areas of debris flow risk over the study area. The ecPoint Rainfall is recommended for the operational application of heavy rainfall leading to debris flow formation which can support impact-orientated forecasting and early warning systems in Central Asia.</p>

To establish climatologies that facilitate the contextualization of extreme, high-impact weather events in relation to historical occurrences or to comprehend the influence of climate change on their intensity and frequency, an extensive collection of observations extending as far back as possible is essential. Yet, these observations exhibit inaccuracies and uneven distributions in space and time. These attributes may lead to a distorted representation of past weather and climate, particularly for variables like rainfall, which can exhibit substantial variations in space and time.  Reanalyses and reforecasts fill the gaps in the observational records. Existing literature has demonstrated that both reanalysis and reforecast datasets offer a more accurate representation of past weather and climate, owing to their global completeness and temporal consistency. Nonetheless, reanalyses and reforecasts may not adequately depict localized and/or rare events as effectively as observational climatologies might do (provided sufficient observations are available) due to the coarse spatial resolutions of both modelled datasets. This misrepresentation is particularly pertinent for discontinuous variables, such as precipitation.  In this presentation, we will examine the representation of point-rainfall climatologies by four distinct global model datasets: ERA5_EDA (reanalysis, 62km), ERA5 (reanalysis, 31km), ECMWF reforecasts (reforecasts, 18 km), and ERA5_ePoint (reanalysis, point-scale on a 31km grid), and compare with observed point climatologies from raingauge sites. Furthermore, we will discuss the implications of this study on future calculations of reference climatologies for localized extreme precipitation events.

Flash floods are one of the most devastating natural hazards. They can occur in very large or small rural or urban areas, with little to no warning. Extreme (localized) rainfall plays a crucial role.   This presentation compares the rainfall forecast performance, for the raw ECMWF ensemble (ENS) and post-processed point-scale output derived from that (ecPoint), in pinpointing areas at risk of flash floods. Performance evaluation is based on location and timing accuracy for the flash floods. Long-term objective verification and case studies are used to compare. Although ENS effectively identifies areas at flash flood risk in instances of large-scale rainfall, its performance falters when confronted with localized extreme convective systems. We show how ecPoint yields superior results for both scenarios, pinpointing well areas at flash flood risk up to medium-range timescales. This enables decision-makers to extend their preparedness and action time window.  This presentation will also demonstrate forecast system strengths and weaknesses, and how forecasters can leverage these to produce better predictions of areas at flash flood risk up to medium-range lead times.