A Multivariate Regression Reconstruction of the Quasi-Global Annual Precipitation on 1-Deg Grid From 1900 To 2015

Abstract

This paper presents a multivariate linear regression reconstruction for the near-global annual precipitation anomalies with 1-deg latitude–longitude resolution from 1900 to 2015. The regression’s explanatory variables are the empirical orthogonal functions (EOFs), computed from the Global Precipitation Climatology Project (GPCP) dataset. The data for the regression’s dependent variable are from the station dataset of the Global Historical Climatology Network (GHCN). The data for the explanatory variables are the EOF data at the GHCN data locations. Compared to the earlier work of reconstruction at [Formula: see text] latitude–longitude resolution, our current reconstruction has two contributions. First, the spatial resolution is reduced to [Formula: see text] latitude–longitude. The finer resolution allows the data to be more useful in applications, such as historical drought assessment for a given region. Second, the multivariate regression is directly computed from linear regression models and hence includes the intercept term, which is not a coefficient of an EOF. The intercept enables a more realistic detection of the long-term trend of the spatial average. The trend of the global average annual precipitation from 1900 to 2015 is 0.133 (mm/day)/100a for the reconstruction with an intercept, and is 0.022 (mm/day)/100a without an intercept. The latter agrees with the trends of other models. The reconstruction error is assessed by a time-varying standard deviation.