基于时空不确定性的对流尺度集合预报效果评估检验

针对对流尺度天气系统的高度非线性特征和高分辨率模式预报结果存在时、空不确定性现象，以及当前邻域概率法主要考虑高分辨率预报结果的空间位移误差，而不能有效解决预报结果存在时间超前与滞后问题，将时间因素引入到邻域概率法中，结合一次强飑线过程进行对流尺度集合预报试验，并基于改进后的新型邻域概率法与分数技巧评分，对降水预报进行了不同时、空尺度的效果评估检验。结果表明:（1）邻域集合概率法和概率匹配平均法在极端降水的分数技巧评分远高于传统集合平均，弥补了集合平均对极端降水预报能力偏低的缺陷。（2）对于此类飑线过程的对流尺度天气系统而言，邻域半径为15—45 km的空间尺度能够改善降水位移误差的空间不确定性，并使其预报效果达到最优，其中15—30 km的邻域半径对于尺度更小的大量级降水事件预报能力更强。（3）对流尺度降水预报考虑时间尺度与降水强度存在着对应关系，不同时间尺度可以捕获到不同量级降水的时间不确定性。同时，时间尺度与空间尺度对于降水预报效果的影响是相互关联的。（4）改进的邻域概率法能够同时体现高分辨率模式预报结果在对流尺度降水事件上存在的时、空不确定性，实现了对流尺度降水在时、空尺度上的综合评估，并能为不同量级降水提供与其时、空尺度相匹配的概率预报结果。 

The assessment and verification of convection-allowing ensemble forecast based on spatial-temporal uncertainties

Considering the highly non-linear characteristics of convective-scale weather systems and spatial-temporal uncertainties in high-resolution numerical forecasting, a convection-allowing ensemble forecast experiment has been conducted to simulate a strong squall line. The neighborhood probability (NP) method mainly considers the spatial displacement error in high resolution model forecasts, and cannot effectively address the problem of difference in event occurrence time between forecasts and observations. Therefore, a time factor is introduced into the NP method in this study. And based on the improved NP method and fractions skill score (FSS), the precipitation forecast of the strong squall line is verified on different spatial-temporal scales. The conclusions are as follows:(1) The FSS of extreme precipitation produced by the neighborhood ensemble probability (NEP) and probability matched mean (PMM) methods is higher than that produced by the traditional ensemble mean (EM) method, and the former two methods overcome the shortcomings of the EM method in predicting extreme precipitation. (2) For the squall line process investigated in the present study, the spatial scale of 15-45 km can reduce the spatial uncertainty in displacement error of precipitation forecast and optimize the forecast effect. The spatial scale of 15-30 km exhibits a better forecast capability for smaller-scale extreme precipitation events. (3) The convective-scale precipitation forecast has a corresponding relationship between temporal scale and rainfall intensity, and different temporal scales can capture temporal uncertainties in precipitation with different magnitudes. Meanwhile, the spatial and temporal scales are inter-related for the precipitation forecast effect. (4) The improved NP method can simultaneously show temporal and spatial uncertainties in high-resolution model forecast of convective-scale precipitation, achieve comprehensive evaluation of convective-scale precipitation on temporal and spatial scales, and provide skillful probabilistic forecast results for precipitation with various magnitudes and corresponding spatial-temporal scales.