An Improvement in a Time-Scale Decomposition Statistical Downscaling Prediction Model for Summer Rainfall over North China

This paper applies partial-correlation predictor selection and a conditional downscaling method to improve a Time-Scale Decomposition (TSD) statistical downscaling model of summer (July and August, JA) rainfall over North China. A new preceding predictor, the North Atlantic-Eurasia Teleconnection (AEAT) in June is found by using the partial-correlation predictor selection method. This predictor stores its signal in the tripole sea surface temperature pattern in the North Atlantic and impacts on the development of depressions over Baikal in the following July and August, which further influences the rainfall over North China. A conditional TSD statistical downscaling model is built with the predictors of Niño3 index and AEAT Index (AEATI). Rather than fixed models for every year, indices are classified into several types according to the predictor strength, and corresponding models are built for each type. The conditional statistical model avoids the influence from weak predictors for a particular year. In independent validation, the conditional TSD downscaling model improves the performance of Summer Rainfall over North China (NCSR) prediction. The correlation coefficient between observed and predicted rainfall increases from 0.61 to 0.77 and the anomaly sign consistency rate increases from 70% to 87%.