邻域法在天气预报中的应用研究进展

Application and research progress of the neighborhood method in weather forecasting

摘要:

邻域法在天气预报中有着广泛的应用,其关键应用领域包括2个方面,一是基于邻域法的高分辨率数值模式检验,二是邻域概率或者集合预报的邻域概率。首先,回顾了邻域法“一对多”和“多对多”的两种邻域法检验框架,归纳了邻域法数据处理方法、常用评分指数的物理意义。其次,总结了网格尺度上的邻域概率和大于网格尺度邻域概率的基本思想和统计意义,重点阐述了与集合预报相结合产生的邻域集合概率(neighborhood ensemble probability,NEP)预报、邻域最大集合概率(neighborhood maximum ensemble probability,NMEP)预报的算法流程和内在含义。第三,进一步结合典型应用个例,分析了邻域法检验和邻域集合概率的优缺点和适用性。总体来说,邻域法检验可以在不同的时空尺度上比较预报产品的性能,具有独特的优势。虽然NEP和NMEP两种邻域概率都可以提高降水的预报评分,但NEP更适合于大尺度、系统性降水预报,NMEP对对流性、极端性降水有更好的应用效果。最后,给出了使用邻域法应注意的问题以及未来研究应用的发展方向。

Abstract:

The traditional dichotomous contingency table test, which evaluates the objective performance of numerical weather prediction (NWP) based on the point-to-point matching between forecasted and observed events, has notable limitation when applied to high-resolution NWP or convection-allowing models (CAM). The neighborhood method addresses these limitations by relaxing the grid scale matching constraints between forecasted and observed events, making it particularly valuable for evaluating high-resolution numerical weather forecasts and the post-processing of objective probability forecasts. This paper systematically reviews the key applications of the neighborhood method in weather forecasting, focusing on two key aspects: one is the verification of high-resolution numerical models using neighborhood method; and other is the neighborhood probability or neighborhood probability of ensemble forecasts. First, the study outlines the verification frameworks of two neighborhood methods,“one-to-many” and “many-to-many”, and discusses the data processing techniques associated with the neighborhood method, alongside the physical interpretation of common scoring matrices such as FBS (fractions brier score) and FSS (fractions skill score). It is concluded that, in addition to traditional dichotomous contingency table-based verification metrics, the neighborhood method facilitates comparisons of forecast performance across multiple spatial and temporal scales. This enables the derivation of diagnostic metrics for NWP forecast performance based on scale changes, providing unique advantages. Second, it summarizes the fundamental concepts and statistical meaning of the grid scale neighborhood probability and the neighborhood probability at scales larger than the grid. Discussion focuses on expounding the algorithm workflow and internal meaning of neighborhood ensemble probability (NEP) forecast and neighborhood maximum ensemble probability (NMEP) forecast derived from ensemble forecasts. Third, by examining typical application cases, it analyzes the advantages, disadvantages and applicability of the neighborhood method and neighborhood ensemble probability. Results show that both NEP and NMEP enhance precipitation forecast scores. NEP performs better for large-scale and systematic precipitation forecasts, whereas NMEP is more effective for convective and extreme precipitation events. However, the selection of an appropriate neighborhood radius remains a critical technical challenge, as it is influenced by variations in underlying surface conditions and the optimal neighborhood scales of different NWP products. Finally, the paper discusses future directions for the application of the neighborhood method in weather forecasting. Promising areas of research and application include integrating neighborhood ensemble probability with the temporal dimension, developing metrics for the rare-event ensemble neighborhood probability, and exploring synergies between the neighborhood method and artificial intelligence. These directions hold significant potential for advancing the utility and impact of the neighborhood method in weather forecasting.