SSIM在高分辨率降水数值预报产品空间检验中的应用

Application of SSIM in Spatial Testing of High-Resolution Precipitation Numerical Forecasting Products

关键词:  降水数值预报  0-1化格点场  空间检验  SSIM应用

Key words:  precipitation numeric prediction  0-1 formatted grid fields  spatial verification  applied SSIM

摘要:

引入图像结构相似度指数 SSIM 应用于高分辨率降水数值预报产品的空间检验，通过对 SSIM 应用场景和算法的适用性改进，提出了适用于 0-1 化降水格点场空间检验的 SSIM 应用方法，并针对 SSIM 指标不能正确反映降水场中雨带位置变化的缺陷进行了改进，提出了修正的 SSIM*指标。应用该新方法选取 2020 年夏季贵州地区的暴雨个例，对中国气象局/广东省区域数值天气预报重点实验室华南区域中尺度模式输出的高分辨率降水预报产品进行了空间检验，并与采用另外两种检验方法即 MODE 和邻域法的检验结果进行了比较。结果表明改进的 SSIM 方法在高分辨率降水数值预报场空间检验中有较好的应用效果，三种方法得到的检验结果在有无预报能力的定性评价上差异不大，而新的 SSIM*指标与 MODE 的 MMI 指标表现较为接近，且在对个例预报能力的评价上更为合理。该方法以 0-1化降水格点场作为空间检验对象，能够得到其它方法所不能得到的降水场整体检验特征和分项检验特征，提供从雨区面积、雨带结构及其平均位置多个方面评估预报性能的参考指标，且计算简单无主观参数设定，得到的检验结果具有唯一确定性，克服了当前 MODE 方法在应用中的缺点，具有较大的推广应用价值，值得进一步深入研究。

Abstract:

This study proposes a new scheme for comparing and analyzing 0-1 formatted grid fields using the image structure similarity index method (SSIM). With the original application scenario and algorithms optimized, the SSIM index is improved to reflect the position change of rainbands in precipitation fields. The new method is applied to the spatial verification of high-resolution precipitation numerical forecast products generated by the regional modeling system of South China (CMA-GD), specifically for rainstorm cases in Guizhou during the summer of 2020. Comparative analysis with results obtained using the MODE and neighborhood methods shows that the qualitative evaluation of prediction cases is consistent, and the performance of the SSIM* index closely aligns with the MMI index of MODE. It shows the effective application of the improved SSIM method and the new SSIM index in spatial verification for high- resolution precipitation numerical prediction products, and results of the new method are even more reasonable in the assessment of prediction cases. With the whole 0-1 formatted grid fields of precipitation as the spatial objects considered, the new method can capture overall spatial structure characteristics that are not attainable through other methods, offering valuable references for assessing prediction cases based on rain area, rainband structure, and average position. Furthermore, the new method has the advantages of simple and efficient calculation, no subjective parameters setting, and well-determined results that overcome the limitations of the MODE which is the current mainstream method in spatial verification for the precipitation numerical prediction products. Therefore, this research holds good application prospects and is worthy of further study.