Evaluation of 8-Day Typhoon Quantitative Precipitation Forecasts in Taiwan by the 2.5-km CReSS Model: Overall Performance during 2012–16 and Impact of Domain Size

Abstract The skill of 24-h daily quantitative precipitation forecasts (QPFs) in Taiwan by a cloud-resolving model out to 8 days was evaluated for 27 typhoons (84 days) during 2012–16. Overall, the skill increased as the lead time shortened, and most of the QPFs were of high quality on both day 1 (0–24 h) and day 2 (24–48 h) with a minimum tendency to overforecast rainfall. At longer lead times from day 4 (72–96 h) to day 8 (168–192 h), where underforecasting became increasingly apparent, decent QPFs that resembled the observation could still be realized if the track errors were small enough due to better quality in the initial and boundary conditions from the global model. For the 14 typhoons that had an observed peak daily amount of ≥350 mm, such QPFs occurred in 10 of them (71%) and the earliest around day 5 on average. Thus, good rainfall scenarios could be provided quite early for many typhoons by this high-resolution system. By comparing the QPF results above to those by the 3-day forecasts that used a smaller domain, potential benefits of an enlarged fine domain were assessed. The only major difference occurred on day 1 at the 750-mm threshold, where the threat score (TS) was 0.17 (0.05) using a larger (smaller) domain, contributed only by two typhoons. While this result appeared random, the larger domain simulated the responsible rainband slightly better and had a higher accumulation to produce hits at a key station on both occasions despite a very similar tropical cyclone (TC) track and thus could be beneficial for typhoon QPFs.