Open AccessA Moving Updated Statistical Prediction Model for Summer Rainfall in the Middle-Lower Reaches of the Yangtze River Valley
Author(s) -
Yan Guo,
Jianping Li,
Jun Zhu
Publication year - 2017
Publication title -
journal of applied meteorology and climatology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.079
H-Index - 134
eISSN - 1558-8432
pISSN - 1558-8424
DOI - 10.1175/jamc-d-16-0376.1
Subject(s) - climatology , anomaly (physics) , environmental science , yangtze river , sea surface temperature , correlation coefficient , forecast skill , mean squared error , geology , statistics , mathematics , geography , physics , archaeology , china , condensed matter physics
Because summer rainfall in the middle-lower reaches of the Yangtze River valley has remarkable interannual and decadal variability and because the precursors that modulate the interannual rainfall change with the decadal variation of the background state, a new model that employs a novel statistical idea is needed to yield an accurate prediction. In this study, the interannual rainfall model (IAM) and the decadal rainfall model (DM) were constructed. Moving updating of the IAM with the latest data within an optimal length of training period (20 yr) can partially offset the effect of decadal change of precursors in IAM. To predict the interannual rainfall of 2001–13 for validation, 13 regression models were fitted with precursors that change every 4–5 yr, from the preceding winter North Atlantic Ocean sea surface temperature anomaly (SSTA) dipole to the Mascarene high, followed by the East Asia sea level pressure anomaly (SLPA) dipole and the preceding autumn North Pacific SSTA dipole. The moving updated model demonstrated high skill in predicting interannual rainfall, with a correlation coefficient of 0.76 and a hit rate of 76.9%. The DM was linked to the April SLPA in the central tropical Pacific Ocean, and it maintained good performance in the testing period, with a correlation coefficient of 0.77 and a root-mean-square error (RMSE) of 7.7%. The statistical model exhibited superior capability even when compared with the best forecast by the Climate Forecast System, version 2 (CFSv2), initiated in early June, as indicated by increased correlation coefficient from 0.62 to 0.75 and reduced RMSE from 12.3% to 10.7%.