| 摘要: | This study compares rainfall forecasts by four cumulus parameterization schemes (CPSs), the Anthes-Kuo, Betts-Miller, Grell, and Kain-Fritsch schemes, using the fifth-generation Pennsylvania State University-National Center for Atmospheric Research Mesoscale Model (MM5) nesting down to 15-km grid spacing. Six rainfall events over the Taiwan area are selected to investigate the CPS performance. The precipitation forecast is evaluated over the model grid points using statistical scores (threat score, equitable threat score, and bias score) for different threshold values based on island-wide rain gauge observations.
The results show that except for the warm-season events (spring rainfall and summer thunderstorm cases), the 15-km MM5, using any of the four CPSs, shows good skill for predicted coverage of measurable 6-h rainfall over the Taiwan area. For rainfall-area and rainfall-amount predictions, the model performs better in cold-season events (winter cold-air outbreak and autumn cold front cases) than in warm-season events, in agreement with previous studies. None of these CPSs consistently outperforms the others in all measurements of forecast skill, and each CPS performs very differently for precipitation prediction under different synoptic forcings. For precipitation events with strong synoptic-scale forcings (like the winter cold-air outbreak and Mei-Yu front cases), the synoptic forcing and Taiwan's topography provide the primary control on the model's rainfall forecast, and the CPSs used in the model only modify precipitation prediction slightly.
In the wettest 3 of 6 cases, the ensemble prediction with an arithmetic average of rainfall forecasts by four CPSs has the best threat score at 0.25-mm threshold. The 15-km MM5 generally overpredicts the area of light rainfall, and underpredicts the area of heavy rainfall, and the model tends to have better predictive skill in the lower land than over the mountainous area in Taiwan. Except for the spring rainfall case, all CPSs underestimate precipitation amount, especially for heavy rainfall cases (Mei-Yu front and Typhoon Otto). The Grell experiment has the best skill in total accumulated rainfall prediction in four out of six cases; the Betts-Miller experiment has the best performance for the rainfall maximum forecast in three out of six cases.
The characteristic responses of four CPS experiments over the Taiwan area are: (i) Anthes-Kuo experiment tends to overpredict the rainfall area, especially for light precipitation events; (ii) Betts-Miller experiment is inclined to produce heavy rainfall, and it underpredicts the rainfall area; (iii) Kain-Fritsch experiment has the best skill in rainfall-area prediction for the winter cold-air outbreak case; and (iv) Grell experiment has the best predictive skill for the heavy-rainfall events of the Mei-Yu front, and Typhoon Otto cases. |
