Short-term Climate Prediction for Summer Rainfall Based on Time-scale Decomposition

By using one set of hindcasted integration of NCEP Climate Forecast System Version 2.0 (CFSv2) beginning from April during 1982-2008, for summer seasonal forecast, along with observations and reanalysis datasets, a downscaling scheme with time-scale decomposition is developed for summer rainfall prediction of the Yangtze-Huai Basins and North China. First, both the predictand and selected predictors are decomposed into inter-annual and decadal scales through Fast Flourier Transformation filtering. And then two downscaling models are separately built, predicted results for two timescales are combined to represent the total prediction. By using the scheme, the summer rainfall of 1982-2008 is hindcasted and compared with CFSv2 raw prediction first. A cross validation shows that skills in the present scheme are significantly improved with increased spatial and temporal correlation coefficients and decreased root mean square error, in comparison with the raw prediction. The spatial correlations with observations for both the Yangtze-Huai Basins and North China have the maximum exceeding 0.8 and a long-term average of 0.53, 0.51, greater than the original-0.06, -0.01 for two regions. The predicted rainfall temporal correlation at each station is also improved, with the regional mean increased from-0.2 to 0.2 in raw model prediction to about 0.5 after downscaling, significant at 0.01 level. The root mean square error exhibits a decrease with the rate of exceeding 10% at most of stations. Then a five-year hindcast from 2009 to 2013 is performed and used for validation as independent samples. Results suggest that spatial correlations of the predicted rainfall with the observed in five samples are significantly higher than the raw prediction, with the mean increased from 0.24, 0.08 to 0.37, 0.44 for two regions. Spatial patterns of rainfall anomaly percentage in two of these independent samples are reasonably closer to observations. Also, the predicted rainfall strength is much closer to the observation, comparing to the raw prediction. Finally, the scheme is applied for the real-time prediction of summer rainfall in 2014. The prediction result displays more rainfall over the mid-lower Reaches of the Yangtze and over the north region of the Yellow River valley, with an anomaly percentage of 20%, along with rainfall anomaly percentage of-10%. Compared with observations, the rainfall anomaly pattern can be predicted to some extent through the downscaling method, especially over the southern region of Yellow River.