AbstractsMathematics

R reconstruction of global precipitation, error analysis and El Nin??o interpretation

by Lars Johannes Laemmlein




Institution: San Diego State University
Department:
Year: 2016
Posted: 02/05/2017
Record ID: 2065885
Full text PDF: http://hdl.handle.net/10211.3/172185


Abstract

This thesis further developes a multivariate linear regression model for the reconstruction of the global precipitation anomaly through empirical orthogonal functions (EOF). These EOFs are computed from the Global Precipitation Climatology Project (GPCP) data. The Global Historical Climatology Network (GHCN) station data is used as the dependent variable. The reconstruction is done on a resolution of 1??1?? and methods of efficiency reduction are described. A detailed description of the whole reconstruction process is provided in this thesis. An additional intercept for the model is significant. The results of the downgrading process, especially the assumptions on the error distribution, are analyzed. The error of the reconstruction model is not normally distributed and has a variable variance. Robust or Cross-Validation estimators can still provide realistic confidence and prediction intervals. The latter estimator is also a good tool to determine the optimal number of EOF modes. The trend of the global average annual precipitation from 1900 to 2015 is 0.022 (mm/day)/100a which coincides with the trend of other models. The goodness-of-fit measures indicate that reconstruction with the high resolution is a powerful model. Patterns in the movement of the center of gravity of El Ni??no events are examined in this thesis. Advisors/Committee Members: Shen, Samuel, Bailey, Barbara, De Sales, Fernando.