No abstract

No abstract

The development and verification of a forecast method for localized, meso‐γ‐scale (2–20 km), extreme heavy rainfall (MγExHR) is important, because it can cause urban flash flooding and inundation with accompanying damage and potential loss of life. Although previous studies have examined the predictability of precipitation at a very short range (≤1 h) using extrapolation‐based nowcasts, they did not specifically focus on MγExHR. In this study, we examine the predictability of 23 selected events of MγExHR (1 h rainfall accumulation ≥50 mm) that occurred during the warm season of 2014 in Japan using High‐Resolution Precipitation Nowcasts (HRPNs) provided by the Japan Meteorological Agency, which are extrapolation‐based nowcasts. Traditional grid‐scale verification using the equitable threat score shows that the HRPNs usefully predict the heavy rainfall areas of ≥20 mm h−1 for at least 12 min at the grid scale of 1 km. Neighbourhood verification using fractions skill scores shows that HRPNs usefully predict the areas of ≥20 mm h−1 up to 29 min by tolerating 11 km displacement errors. After 30 min, a useful forecast cannot be obtained, even if the 11 km displacement errors are tolerated for the ≥20 mm h−1 areas. This result suggests that a numerical weather prediction (NWP) model, whose accuracy is useful after ∼30 min, is necessary to seamlessly provide useful forecasts for heavy rainfall areas of ≥20 mm h−1 for MγExHR with ∼10 km displacement errors, by blending the extrapolation‐based nowcast with NWP.

Torrential rainfall associated with linear precipitation systems occurred in Northern Kyushu, Japan, during July 5–6, 2017, causing severe damage in Fukuoka and Oita Prefectures. According to our statistical survey using ground rain gauges, the torrential rainfall was among the heaviest in recorded history for 6- and 12-h accumulated rainfall, and was unusual because heavy rain continued locally for nine hours. The predictability of precipitation associated with linear precipitation systems for this event was investigated using a cloud-resolving numerical weather prediction model with a horizontal grid interval of 1 km. The development of multiple linear precipitation systems was predicted in experiments whose initial calculation time was from several hours to immediately before the torrential rain (9:00, 10:00, 11:00, and 12:00 Japan Standard Time on July 5), although there were some displacement errors in the predicted linear precipitation systems. However, the stationary linear precipitation systems were not properly predicted. The predictions showed that the linear precipitation systems formed one after another and moved eastwards. In the relatively accurate prediction whose initial time was 12:00 on July 5, immediately before the torrential rainfall began, the forecast accuracy was evaluated using the 6-h accumulated precipitation (P6h) from 12:00 to 18:00 on July 5, the period of the heaviest rainfall. The average of the P6h in an area 100 km×40 km around the torrential rainfall area was nearly the same for the analysis and the prediction, indicating that the total precipitation amount around the torrential rainfall area was predictable. The result of evaluating the quantitative prediction accuracy using the Fractions Skill Score (FSS) indicated that a difference in location of 25 km (50 km) or greater should be allowed for in the models to produce useful predictions (those defined as having an FSS ≥0.5) for the accumulated rainfall of P6h ≥50 mm (150 mm). The quantitative prediction accuracy examined in this study can be basic information to investigate the usage of predicted precipitation data.

No abstract