Description
This study provides a spatial prediction for the monthly temperature anomaly field on a 5°×5° grid of the quasi-globe from 75°S-75°N, using land only data between January 1880 and December 2015. Predictions for the quasi-globe are also done utilizing only data from oceanic regions. A multivariate regression method is used for the prediction, which employs Empirical Orthogonal Functions (EOFs) as regressors. The following datasets are used as response variable: (i) the Global Historical Climatology Network (GHCN) data for the reconstruction from land only data; (ii) the National Oceanic and Atmospheric Administration (NOAA) Global Surface Temperature data over the ocean for the reconstruction from ocean only data. The EOFs are computed from the National Centers for Environmental Prediction (NCEP) Reanalysis data for monthly air temperature means from January 1948 to December 2015. EOF patterns are interpreted for El Niño effects and temperature trends. Tests of detrending the NCEP data are analyzed for the EOF computation and the spatial predictions. This thesis includes new results on the error analysis and model validation of the multivariate regression predictions generated from land data. As a criterion for model selection, it is found that the Akaike Information Criterion (AIC) can contribute to identify important climate patterns in the EOF modes but does not necessarily improve the predictive performance. Prediction errors are estimated by both theory and comparison with the complete NOAA Global Surface Temperature dataset. The time series of the global average temperature of the reconstruction from land data shows a trend of 1.02 °C/100a, which is identical to the trend of NOAA global land data. For evaluation of the reconstructed temperature patterns, spatial correlations are computed between the reconstructed temperature field and the NOAA Global Temperature data. All spatial correlations are positive with maximum correlation of 0.824 over the ocean, and 0.915 over the land. Comparisons to the reconstruction from ocean only data (ii) reveal an effect of oversmoothing between ocean and land temperature in the NOAA Global Temperature data. It is conjectured that the NOAA Global Temperature data systematically underestimate sea surface temperature. Spatial prediction examples of January 1983 and January 1998 demonstrate that the oceanic El Niño Southern Oscillation (ENSO) can be accurately predicted from the land data only.
