资源环境科技发展态势分析平台

A hybrid downscaling model is established to forecast the winter temperature at 182 stations over northeast China based on the year-to-year increment approach (differences in variables between the current and previous year). As winter temperature over China is related to general circulation and oceanic circulation signals, the three most suitable predictors at specific regions are ultimately considered in our model after analysing the physical processes involved. Accordingly, winter sea level pressure (SLP) from version 2 of the NCEP's Climate Forecast System (CFSv2) is used as a current-year predictor, along with two other previous-year predictors-sea surface temperature (SST) in August and sea-ice concentration (SIC) in November, from observation. Four separate downscaling schemes are then proposed, three of which involve just one predictor (i.e. SLP-scheme, SST-scheme, and SIC-scheme), and the fourth involves all of the predictors [i.e. hybrid downscaling (HD)-scheme]. We evaluate the schemes, through cross-validation, by examining the spatial and temporal anomaly correlation coefficients (ACCs) and reduction in root-mean-square error percentage (RMSEP). Considerable evidence is found that all of the schemes improve the skill in predicting winter temperature over northeast China, especially the HD-scheme, as compared with real-time CFSv2 forecasting. The 33-year spatial ACC average for the HD-scheme rises approximately from -0.017 to 0.30, which is much larger than the threshold value of 0.19, showing statistical significance at the 99% confidence level. Additionally, the temporal RMSEP (i.e. the difference between the RMSE of the CFSv2 outputs and the downscaling results) decreases by more than 30% for all four schemes. Moreover, the cold winter temperature pattern over northeast China in 2013 is reproduced well by the HD-scheme.