大气科学学报2017，Vol.40Issue(6)：749-758,10.DOI:10.13878/j.cnki.dqkxxb.20160419002
基于多种集合的中国冬季气温逐月滚动预测
A monthly rolling prediction for winter surface air temperature over China based on multi-ensemble
摘要
Abstract
With the development of social economy and the improvement of people's living standard,the demand of country and society for the short-term climate prediction is increasing.Though current methods including statistic,dynamical-statistic and numerical methods for the prediction of surface air temperature in wintertime are more,the prediction lead time is usually short and the forecast skill is not stable.For example,the seasonal prediction of climate model for winter temperature is still low outside the tropics and the most models cannot give reliable results in many areas of China.So it is very important to carry prediction experiment of winter air temperature and expand valid prediction lead time,in order to meet the needs of the society.Based on NCC(National Climate Center of China) monthly surface air temperature data of 160 stations in China and NCEP/NCAR monthly mean reanalysis data during 1979-2015,the predictive factors are selected from early winter geopotential height at 500 hPa and velocity potential at 850 and 200 hPa during 1979/1980-2008/2009.Considering the combination of different predictive factors and their independence,the monthly rolling forecasting models are separately established by the multi-variable regression ensemble,the cross validation test ensemble and the monthly rolling prediction ensemble,in order to perform independent predictive tests for the winter temperature in China during 2010/2011-2014/2015.The velocity potential can reflect the external forcing source of atmospheric system,and 500 hPa height can denote the basic state of atmospheric circulation.Although the memory of internal evolution within atmosphere circulation is about a week or so,the initial time potential function at 850 and 200 hPa can reflect variations of the upper and lower level boundary forcing anomalies and their influences on the future atmosphere.Besides,it is simple and practical to select factors from the predictands on the above three levels.Results show that the multi-variable regression ensemble(ENC1) may increase predictable station number.Combined using of the multi-variable regression ensemble and the cross validation test ensemble(ENC2) can improve stability and prediction skill,which is negatively affected by unstable statistic relationship between predictor and predictand.The comprehensive ensemble of multi-variable regression,cross validation test and monthly rolling prediction (ENC3) can not only increase the predictable station number,but also make the prediction more stable,which improves the feasibility and stability of objective quantitative prediction of short-term climate.Although the data used in establishment of prediction model are less and not complex,the final prediction model,through the comprehensive application of the three ensemble methods,has a certain predictive ability for the winter surface air temperature in China,and the prediction lead time is relatively long.Therefore,the statistic model established here will make the long-lead prediction reliable and effective with valuable skill,which is very important in practical use in short-range climate prediction.In addition,the comprehensive application of multi-ensemble methods can also be employed to correct the numerical model products by the establishment of dynamic statistical forecasting model.关键词
逐月滚动预测/统计预测模型/多集合预测/冬季气温Key words
monthly rolling prediction/statistic prediction model/multi-ensemble prediction/winter surface air temperature引用本文复制引用
谭桂容,尹丝雨,王永光..基于多种集合的中国冬季气温逐月滚动预测[J].大气科学学报,2017,40(6):749-758,10.基金项目
公益性行业(气象)科研专项(GYHY201206016 （气象）
GYHY201306028) （）
国家自然科学基金资助项目(41475088) （41475088）
