Multivariate Regression Reconstruction and Its Sampling Error for the Quasi-Global Annual Precipitation from 1900 to 2011

Abstract

his paper provides a multivariate regression method to estimate the sampling errors of the annual quasi-global (75°S?75°N) precipitation reconstructed by an empirical orthogonal function (EOF) expansion. The Global Precipitation Climatology Project (GPCP) precipitation data from 1979 to 2008 are used to calculate the EOFs. The Global Historical Climatology Network (GHCN) gridded data (1900?2011) are used to calculate the regression coefficients for reconstructions. The sampling errors of the reconstruction are analyzed in detail for different EOF modes. The reconstructed time series of the global-average annual precipitation shows a 0.024 mm day?1 (100 yr)?1 trend, which is very close to the trend derived from the mean of 25 models of phase 5 of the Coupled Model Intercomparison Project. Reconstruction examples of 1983 El Niño precipitation and 1917 La Niña precipitation demonstrate that the El Niño and La Niña precipitation patterns are well reflected in the first two EOFs. Although the validation in the GPCP period shows remarkable skill at predicting oceanic precipitation from land stations, the error pattern analysis through comparison between reconstruction and GHCN suggests the critical importance of improving oceanic measurement of precipitation.