Skilful two‐month‐leading hybrid climate prediction for winter temperature over China

• Published in: International journal of climatology 2020-09, Vol.40 (11), p.4922-4943
• Main Authors: Dai, Haixia, Fan, Ke
• Format: Article
• Language: English
• Subjects: Air temperature; Arctic sea ice; CFSv2; China and Northeast China; Climate; Climate prediction; Correlation coefficient; Correlation coefficients; hybrid climate prediction; North Pacific sea surface temperature; Performance prediction; Physics; Polar environments; Sea ice; Sea ice concentrations; Sea level; Sea level pressure; Sea surface; Sea surface temperature; Surface temperature; Surface-air temperature relationships; two‐month‐leading; Winter; Winter climates; Winter temperatures; year‐to‐year increment
• ISSN: 0899-8418; 1097-0088
• DOI: 10.1002/joc.6497

Two skilful 2‐month‐leading hybrid downscaling prediction schemes in October for winter surface air temperature (SAT) over China are proposed in this paper. The schemes are based on the year‐to‐year increment approach and the coupled climate patterns between the winter SAT over China and its predictors. Observed North Pacific sea surface temperature (SST) from the preceding July to September, Arctic sea ice concentration (SIC) in the preceding August and winter sea level pressure (SLP) over pan Eurasia from version 2 of the Climate Forecast System (CFSv2) are selected as the predictors based on the fundamental physics. Individual‐predictor schemes (IP‐schemes), that is, SLP‐scheme, SST‐scheme, SIC‐scheme, indicate that these predictors exhibit prediction skills in different regions. Multi‐predictor scheme I (MP‐schemeI) is developed by combining three predictors. However, MP‐schemeI shows limited skill in predicting SAT over Northeast China (NECTA), due to the limited skill of CFSv2 over the extratropics. Thus, MP‐schemeII is established, in which a hybrid downscaling model for NECTA is constructed. These two MP‐schemes have comparable prediction skill over China, but MP‐schemeII outperforms MP‐schemeI over NEC. The temporal (spatial) anomaly correlation coefficient (ACC) increases from 0.23 (0.15) in MP‐schemeI to 0.36 (0.21) in MP‐schemeII, and the ratio of the same sign of anomalous years (Anomalous‐RSS) improves from 39.1% to 56.5% over NEC. For the winter SAT over China, the MP‐schemes greatly enhance the prediction compared with CFSv2 outputs and the IP‐schemes. The temporal (spatial) ACC increases from 0.29 (0.05) in CFSv2 to 0.52 (0.29) in the MP‐schemes, and the station‐average root‐mean‐square error decreases by about 48.0% compared with CFSv2. Moreover, the RSS (Anomalous‐RSS) of the winter SAT over China is 55.9% (47.4%) in CFSv2 and 64.7% (63.2%) in the MP‐schemes. This indicates that the MP‐schemes perform better in predicting anomalous winters. This study proposes a 2‐month‐leading hybrid downscaling prediction scheme for winter temperature over China in October based on the coupled National Centers for Environmental Prediction (NCEP)‐CFSv2 model. Both simultaneous and preceding predictors are adopted to develop the prediction model based on the understanding of specific physical processes. A different prediction model for Northeast China (NEC) is proposed due to the specificity of factors related to winter temperature over NEC. Thus, the sc