Where and why do Mei-yu season Heavy-rainfall quantitative precipitation forecasts in Taiwan improve the most using a higher model resolution

In this study, quantitative precipitation forecasts (QPFs) for 24-h Mei-yu rainfall at the short range (days 1–3) during May-June of 2012–2014 by a cloud model at two different grid sizes of 2.5 and 5 km are compared using point-to-point categorical measures. With strong topographic control and enhancement, abundant Mei-yu rainfall in Taiwan allows for the use of very high thresholds up to 500 mm (per 24 h), and classification based on observations is also performed to isolate the larger 16% (group A) and the largest 4% of events (group A+) from all samples. Our results show clear improvements in threat scores in heavy rainfall, with the greatest gain (by 0.16) on day 1 at the highest threshold adopted (500 mm) in the largest events of group A+, when a finer grid is used. Improvements are seen at thresholds ≥ 200 mm on day 1, ≥ 100 mm on day 2, and over 50–350 mm on day 3, mainly due to a better capability of the finer model to simulate heavy rainfall in larger events over and near the terrain. The present work provides new insights into the importance and usefulness of increasing model resolution, when and if QPFs of heavy rainfall at precise locations are crucial for hazard mitigation. Similar benefits are not as evident in the literature, likely because the thresholds used were not high enough, the larger events were not isolated, or the impact of topography on rainfall is not as strong and apparent as in Taiwan.