Change Detection with Climate Data Application

Abstract

This thesis has two major components: Change Detection and Hurricane Trajectory Prediction. In Change Detection, we generalize the scope of the existing change detection framework, and propose change detection algorithms in the context of exponential family distribution, generalized extreme value distribution and so on. We also propose a novel and efficient approach to detect extreme change through order statistics. Performance of the proposed methodologies are assessed through simulation study. For Hurricane Trajectory Prediction, we propose a collection of models(neighborhood methods, time series etc), and ensemble them to predict the hurricane's upcoming locations. We analyze the prediction performance through historical hurricanes, and discuss its advantages and disadvantages.