GRAPES＿GFS在西南地区的预报稳定性及其误差与地形的关系

Prediction stability of GRAPES＿GFS in Southwest China and the relationship between its error and the terrain

根据地形特征,将西南地区划分为高原区、边坡区和盆地区,引入统计学"不稳定度"定量描述模式预报稳定性,对2016年6月—2017年9月全球中期天气预报(GRAPES_GFS)和欧洲中期天气预报中心(EC)在西南地区的高层形势场、主要的天气影响系统和地面要素预报性能进行了主客观检验,一定程度揭示了GRAPES_GFS和EC在西南地区的预报稳定性、地形的影响以及二模式预报性能的异同。结果显示:GRAPES_GFS高空高度场、温度场预报不稳定度分布呈北高南低型,相对湿度、风速预报不稳定度大值区在高原边缘;各要素预报不稳定度季节性周期最为显著,其位相和振幅因要素不同而有所不同;地形主要影响温度和风向预报误差值,但对相对湿度和风速预报的影响则体现在误差随时效的增长速率差异上;"漏报"是模式对西南地区天气系统的主要预报误差源,"低报"则是模式对西南地区2 m温度预报误差的最大来源;模式对西南地区降水落区预报有效率大约为50%,但强度预报通常偏低。EC与GRAPES_GFS的误差特征没有本质区别,但EC误差更小,稳定性更高。

By introducing the statistics concept of"degree of instability"and dividing Southwest China into three kinds of areas, namely, the plateau, the side slope and the basin area in consideration of the terrain complexity there, numerical model GRAPES_GFS and EC's prediction stabilities, their temporal and spatial variations, the influence of the terrain on the models'performance as well as the difference between the two models were analyzed subjectively and objectively based on their predictions over the period from July, 2016 to September, 2017. The results showed that GRAPES_GFS's prediction instability of geopotential height and temperature fields appeared higher in the north of the area, while higher instability of relative humidity and wind speed was around the Tibet Plateau to the north and east. The instability had obvious seasonal fluctuation with different phases and amplitude as different elements. Terrain impacted mainly on the prediction error value of temperature and wind direction. For both relative humidity and wind speed, its impact is more on the rate of error growth. The"lower prediction"led to a high failure rate of GRAPES_GFS in 2 m temperature prediction, and"missing"was its major error source of weather systems prediction. GRAPES_GFS predicted the location of rainfall in the Southwest China by about effective rate of 50%, although its prediction of rainfall intensity was usually lower than the observation. The two models had little essential difference in the characteristics of prediction error. Compared to GRAPES_GFS, EC had less errors and higher stability.