Development of a method with high-reliability to forecast the occurrence of location and time of local heavy rainfall within 24 hours.

利用模式集成方法，结合中尺度气象模式PSU/NCAR MM5中多个参数化得到的32个成员集合，建立了局地暴雨预报方法。多参数集合预报与使用20公里分辨率全球分析JMA GSM-GPV和雷达数据同化系统初始化的模式相结合。通过对日本气象厅(JMA)雨量实测值与预报雨量之间的形变误差进行评估，提出了一种基于图像处理方法的降雨短时预报新方法，以减小JMA雨量短时预报的误差。对强降雨过程的验证实验结果表明，该方法考虑了雨区复杂的变形过程，能显著提高降雨短时预报的精度。此外，它清楚地表明，新方法对未来一小时的预报值比原来的JMA降雨量现在预报的准确度至少提高了10%。

Forecasting method of local heavy rainfall was developed by using a model ensemble approach combining 32 member ensembles obtained from multiple parameterizations in the mesoscale meteorological model PSU/NCAR MM5. The multi-parameterization ensemble forecasts were combined with the model initialized using the 20km resolution global analyses JMA GSM-GPV with the Radar-data assimilation system. The forecasts averaged by 32 members are found to be very skillful in indicating the area of severe local storm activity, while each member deviates largely from the ensemble average.A new rainfall nowcasting approach based on the image processing method taking in complicated changes of rainfall area was developed, by evaluating the deformational errors between observed and predicted rainfall obtained by Japan Meteorological Agency(JMA) so as to minimize errors of the JMA rainfall nowcasting in real time based on its evaluation. Results of the verification experiments for intense rainfall events verified that the new approach can remarkably improve the accuracy of rainfall nowcasting by considering complicated deformation processes of rainfall area. Further, it made clear that the predictive value for one hour ahead forecasted by the new approach is at least 10% more accurate than that by the original JMA rainfall nowcasting.