Global environmental change has surfaced as a critical issue to both the scientific community and the general public. One aspect of particular concern involves climate change, which will exert impacts on ecosystems and economies, presenting considerable challenge to human adaptation. In Africa, a continent that is vulnerable due to multiple stressors and low adaptive capacity, climate change is expected to significantly affect both people and ecosystems. Adaptation strategies are being developed using information from studies that evaluate the impacts of climate variability and climate change in Africa. Recommendations are made for local development of adaptation strategies due to the heterogeneity of climate change and its effects on East Africa's climate. However, global climate change models are coarse in scale and mask much of the local variation in regional climate, indicating the need for higher resolution climate data. This dissertation addresses this need by comparing spatially explicit statistical methods of interpolation and prediction, both theoretically and empirically; expanding upon the method of universal kriging by incorporating complex feedback relationships that may produce simultaneity between precipitation and its covariates; and evaluating precipitation patterns over space in East Africa through a case study. Mechanisms of precipitation have been considered in detail, expanding upon many other spatially explicit applications of prediction methods to date. Further, spatially explicit inferential regression models have been developed to better understand spatial patterns and variability in East African precipitation. Predicted maps of precipitation, generated at a resolution of 1 kilometer, accurately reflect the mesoscale influences of topography and the presence of large water bodies (i.e., Lake Victoria) as well as the seasonal influences of the passing of the intertropical convergence zone (ITCZ). In terms of prediction, the spatially explicit methods considered herein clearly outperformed a global data set (i.e., the CRU TS 3.1) in terms of error and ability to reflect local variability. The method of local ordinary kriging generally outperformed the multivariate kriging techniques, indicating that precipitation patterns in areas of high topographic variability, such as East Africa, may be modeled as well or better using local search neighborhoods in the kriging process rather than using complex multivariate regression models. However, additional work to improve the multivariate regression models and overall levels of correlation are expected to yield improved prediction results. Furthermore, the case study successfully demonstrated that the newly developed method of universal kriging with instrumental variables performs similarly to other standard methods of estimation, and perhaps better in the presence of significant measurable simultaneity.
Read
