基于逐步回归模型的华南区域模式气温预报产品释用技术研究

Research on Interpretation and Application of Temperature Forecast Products of Regional Model in South China Based on Stepwise Regression Model

基于2017-2019年华南高分辨率区域数值模式2 m气温预报产品,设计了3套模型的构建方案,利用25个显著因子建立97个国家基本站的逐步回归订正模型,对模式和3套订正模型的预报与实况进行偏差分析,结果表明:(1)华南区域模式分辨率较高,但无法细致反映湖南复杂地形影响导致的区域气温变化特征,模式预报的气温,山区的偏差大于平原的;在1-72 h预报时效内的准确率为54.08%,预报偏差随预报时效延长显著增加,且具有明显的周期振荡,偏差最大值出现在日最高/低气温发生时段。(2)基于逐步回归模型,依据模式的起报时次、预报时效建立了3套气温订正模型。其中,方案1为不分起报时次和预报时效建模,方案2为分起报时次不分预报时效建模,方案3为分起报时次和预报时效建模。3套针对不同起报时次和预报时效的建模方案皆对湖南2 m气温的时空分布特征有较好刻画,对复杂地形影响区域的气温预报有一定的订正能力,但随着预报时效的延长订正效果下降,订正效果总体为00时起报的优于12时起报的。(3)3套建模方案2 m气温预报同样具有周期振荡特征,但没有明显的季节差异,最大偏差出现在最高气温发生时段,对夜间温度,尤其是对05时前后的温度预报订正效果明显,其中方案1的气温订正效果总体优于方案2和方案3的,1-72 h预报时效内绝对误差减小了0.68℃,准确率提高了17%以上。

Based on the 2 m temperature prediction products of South China mesoscale weather prediction model from 2017 to 2019,three sets of model construction schemes were designed using 25 significant factors to establish the stepwise regression correction model of 97 national basic stations.The deviation between the revised prediction of the three sets,the models and the actual situation were analyzed.The results show that:(1)although the regional model of South China has high resolution,it could not reflect the characteristics of regional temperature change caused by the complex terrain of Hunan.The deviation in mountainous areas is greater than that in plain areas.The accuracy is 54.08%within 172 hours.The prediction deviation increases significantly with the extension of prediction time,which has obvious periodic oscillation.The maximum deviation occurs with daily maximum/minimum temperature.(2)Based on the stepwise regression model,three sets of temperature correction models are established according to the starting time and prediction time of the model.Scheme 1 is the modeling not distinguishing starting time and prediction time,scheme 2 is the modeling distinguishing starting time and not distinguishing forecasting time,and scheme 3 is the modeling distinguishing both starting time and prediction time.The three sets of modeling schemes all have a good description of the temporal and spatial distribution characteristics of 2 m air temperature in Hunan.It has certain correction ability for the temperature forecast in the area affected by complex terrain.However,with the extension of prediction time,the correction effect decreases.The overall correction effect at 00:00 start time is better than that at 12:00.(3)The 2 m temperature forecast of the three modeling schemes has the characteristics of periodic oscillation,and no obvious seasonal difference.The maximum deviation occurs with maximum temperature.The correction effect is obvious for the night temperature,especially before and after 05:00.The temperature correction effect of scheme 1 is better than that of scheme 2 and scheme 3.The absolute error within 172 h prediction time is reduced by 0.68℃,and the accuracy is improved by more than 17%.