Application of an Ensemble Canonical Correlation Method in Summer Temperature Prediction over China

Using geopotential height at 500 hPa, sea level pressure, surface temperature, and temperature at 850 hPa in winter over East Asia as predictors, predictability in summer temperature over China is analyzed. Based on the detrended datasets during the period 1951-2009, individual forecasting models produced separately by Barnett-Preisendorfer canonical correlation analysis (BP-CCA) are established, and the ensemble canonical correlation (ECC) prediction based on one-year-out cross validation is used to predict the summer temperature over China during the same period. Independent sample tests are then performed based on these datasets over the period 2010-2014. Analyzing the BP-CCA mode shows that the spatial patterns of BP-CCA can in general reflect the remote correlation characteristics between predictor and predictand. By the prediction test based on one-year-out cross validation, it is confirmed that circulation and thermal fields can provide effective information for temperature prediction. Since ECC prediction collected the skill of each predictor in different areas, its skill is higher and more stable than any individual BP-CCA prediction. Compared with these individual BP-CCA models, the ECC model in the independent sample test shows a better and more stable performance in predicting summer temperature, which is effective for seasonal temperature prediction.