Open AccessAn Evaluation of WRF Rainfall Forecasts in Taiwan during Three Mei-Yu Seasons from 2008 to 2010
Author(s) -
Chung Chieh Wang,
Fang Ching Chien,
Sahana Paul,
Dong In Lee,
Pei-Yu Chuang
Publication year - 2017
Publication title -
weather and forecasting
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.393
H-Index - 106
eISSN - 1520-0434
pISSN - 0882-8156
DOI - 10.1175/waf-d-16-0190.1
Subject(s) - weather research and forecasting model , climatology , precipitation , environmental science , diurnal cycle , flash flood , ridge , monsoon , rain gauge , morning , meteorology , geology , flood myth , geography , medicine , paleontology , archaeology
The performance of the Weather Research and Forecasting (WRF) Model in quantitative precipitation forecasts (QPFs) across Taiwan during three mei-yu seasons between 2008 and 2010 is evaluated using observations from about 400 rain gauges. The QPFs, spanning a range of 12–36 h and run for two nested domains at grid sizes of 15 and 5 km, are verified. Both visual and statistical-based verification methods are used to provide complementary results. More emphasis is placed on intraseasonal variation and the diurnal cycle of mei-yu rainfall, as these aspects have been less well explored previously. While the categorical statistics indicate skill levels comparable to past studies, the model performs better for frontal rainfall in May than monsoon rainfall in June. The two WRF domains are found to capture the overall rainfall amount, its general spatial pattern, the increased rain from May to June, and the basic diurnal cycle to a reasonable extent. However, both domains exhibit a persistent eastward shift in rainfall areas throughout the season, from the upwind slope to near the ridge, mainly because of excessive daytime rainfall over the mountains that starts and ends too early (more so in June), combined with insufficient rainfall upstream (in plains and slopes) since the morning. Also, the 15-km domain has total rainfall amounts closer to the observations, but the 5-km domain suffers a larger underforecast with rainfall only at the resolvable scale. Despite this, the finer mesh is more capable of predicting the peak values and local variations in rainfall and, thus, has the same skill with higher hit percentages, especially toward the high thresholds.