Files
Abstract
In this dissertation, we aim to simulate realistic daily climate data at 23 locations across Puerto Rico through the year 2100 via statistical downscaling. To accomplish this, we first propose and examine a multivariate infilling algorithm that is a combination of a vector autoregressive model (Sims, 1980) and the expectation-maximization algorithm (Moon, 1996). In order to improve predictions from this algorithm, we define and derive a new vector auto regressive model which is allowed to incorporate contemporaneous observations as predictors. We call this model the adapted vector autoregressive model. This iterative interpolation routine is then used to populate missing data from a broken historical climate network within Puerto Rico.
Next, we utilize the multivariate downscaling methods of Jeong et al. (2012a) and Jeong et al. (2012b) to generate site-specific climate projections at 23 locations in Puerto Rico from four general circulation models at two carbon emission scenarios: RCP4.5 and RCP8.5. Finally, we propose a bias correction method that combines elements from localized regression and quantile mapping. This extension to quantile mapping allows GCM bias to vary as a function of season (Julian day). Results tend to favor a warmer and wetter climate in Puerto Rico in the coming century. While low magnitude rainfall events may decrease in frequency, high magnitude rainfall events are projected to increase in frequency in most models. Furthermore, and perhaps related to the projected increase in precipitation, the average day in Puerto Rico is projected to see a reduction in solar radiation.
Next, we utilize the multivariate downscaling methods of Jeong et al. (2012a) and Jeong et al. (2012b) to generate site-specific climate projections at 23 locations in Puerto Rico from four general circulation models at two carbon emission scenarios: RCP4.5 and RCP8.5. Finally, we propose a bias correction method that combines elements from localized regression and quantile mapping. This extension to quantile mapping allows GCM bias to vary as a function of season (Julian day). Results tend to favor a warmer and wetter climate in Puerto Rico in the coming century. While low magnitude rainfall events may decrease in frequency, high magnitude rainfall events are projected to increase in frequency in most models. Furthermore, and perhaps related to the projected increase in precipitation, the average day in Puerto Rico is projected to see a reduction in solar radiation.