Preliminary Study on the Scoring Methods of Cloud-free Rainfall/Snowfall and Air Temperature Forecasts

Scientific and reasonable forecast scoring method is fundamental for evaluating the efficiency of the weather forecast objectively.By comparing forecast results with the corresponding observed data of 373 stations in China, the problems in existing forecast scoring method of the cloud-free rainfall/snow fall and temperature are investigated. Experimental amendments are made to the method, too. As a widely used method, forecast accuracy of the cloud-free rainfall/snowfall proves to be simple and practical in judging the effects of forecast to some extent.But without considering influence of rainfall probability, the forecast accuracy cannot distinguish blind prediction or persistent prediction accuracy effectively, and may bring abnormal high value even some mistakes.Skill-score of the cloud-free rainfall/snowfall forecast accuracy is positively correlated to the no-rainfall frequency at a single station during the months when regional rainfall probability difference is significant. In terms of temperature forecast, daily variation of the air temperature is a significant factor that affects the skill-score.When adopting 1℃ or 2℃ as the absolute standard value threshold, the skill-score of temperature forecasting is negatively correlated to the daily temperature variation.P_F method and Index Threshold method are proposed in order to reduce the influences of rainfall probability and daily temperature variation.The results indicate the forecast score of precipitation by P_F method is not closely correlated with rainfall frequency at single station during the months when regional rainfall probability difference is dominant in China. When adopting 2/3 or 1/2 as the index threshold, the linear regression coefficient between the daily temperature variation and the temperature forecast score can be significantly reduced.The correlation coefficient also decreases obviously with the value down to below 0.15, which is clearly lower than adopting the absolute standard value threshold. In other words, skill-score of temperature forecast using Index Threshold method is less sensitive to the daily temperature variation than using forecast accuracy method. Above all, the new methods proposed can effectively reduce the influences of rainfall probability and daily temperature variation on the skill-score of the cloud-free rainfall/snowfall and air temperature. It also improves the comparability of the weather forecast scores in the regions with different climate background. Therefore, it can be applied to the quality test and assessment on the weather forecast.