天气雷达定量降水估测订正优化算法

Research on Optimization of Quantitative Precipitation Estimation Calibration Method

摘要:

本文利用2020年5—9月山东省青岛市S波段天气雷达和地面雨量站数据，开展了基于最优插值的天气雷达定量降水估测订正方法的优化研究。根据青岛地区地面雨量站的空间分布，确定了用于订正格点估测降水的雨量站的搜索范围。通过分析用于订正的雨量站个数与分析误差的关系，确定了参与订正的最佳雨量站数量。采用随机抽样和交叉检验的方法，分析了不同相关函数模型对降水订正结果的影响，最终得到适合青岛地区的相关函数优化参数。进一步检验优化参数的订正结果发现，订正后估测降水的统计评分有明显的提升，估测降水的强度和空间分布与实况更加一致。

Abstract:

The optimal interpolation method is widely used in meteorological applications worldwide. However, due to the different precipitation climate characteristics and the spatial distribution of gauges, the implementation of the optimal interpolation method involves many empirical models and key parameters. There is still some uncertainty on how to derive a set of optimised parameters for the application of the optimal interpolation method in a local region. This study analyses and optimises the key parameters of the optimal interpolation method for the calibration of weather radar quantitative precipitation estimation using radar and gauge observations from May to September 2020 in Qingdao, Shandong. The search radius of gauges from a grid point is determined by analysing the spatial distribution of all gauges relative to the analysis point. The optimal number of gauges used to calibrate the precipitation estimation of a grid point is determined by analysing the change of relative analysis error with respect to the number of gauges. Eight groups of sensitivity experiments with different correlation functions are compared by random sampling and cross-validation to find the best set of parameters for the Qingdao radar. The verification of calibration results produced by the best parameters shows that the calibration significantly improves the accuracy of the quantitative precipitation estimation. The median values of M_AE, R_MAE, and B_IAS are 1.5 mm, 1.0, and 0.03 mm respectively, and the C_ORR is higher than 0.9. Comparing the quantitative precipitation estimation at different levels of precipitation before and after calibration produced by the best parameters shows that the M_AE and R_MAE of light rain are reduced by 90%, and the C_ORR is about 0.87. The M_AE and R_MAE in moderate to heavy rain are decreased by 89%, and the C_ORR is higher than 0.9. The M_AE and R_MAE of rainstorm are decreased by more than 83.9%, and the C_ORR is about 0.77. Based on the verification of a widespread rainfall case on 26 August 2020, the intensity and spatial distribution of the quantitative precipitation estimation after calibration are closer to the gauge observations. The original quantitative precipitation estimation is relatively smooth and lacks small-scale variations. The calibrated results can reflect the characteristics of the local change that is consistent with the observation of gauges. The results of this paper suggest that the optimal interpolation method with optimised local parameters can significantly improve the accuracy of the quantitative precipitation estimation, which has important application value for rainstorm warning and flood disaster prevention.