Date of Award

2015

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Homaifar, Abdollah

Abstract

Providing advance notice of rare events, such as a cloud cluster (CC) developing into a tropical cyclone (TC), is of great importance. Having advance warning of such rare events possibly can help avoid or reduce the risk of damages and allow emergency responders and the affected community enough time to respond appropriately. Considering this, forecasters need better data mining and data driven techniques to identify developing CCs. Prior studies have attempted to predict the formation of TCs using numerical weather prediction models as well as satellite and radar data. However, refined observational data and forecasting techniques are not always available or accurate in areas such as the North Atlantic Ocean where data are sparse. Consequently, this research provides the predictive features that contribute to a CC developing into a TC using only global gridded satellite data that are readily available. This was accomplished by identifying and tracking CCs objectively where no expert knowledge is required to investigate the predictive features of developing CCs. We have applied the proposed oversampling technique named the Selective Clustering based Oversampling Technique (SCOT) to reduce the bias of the non-developing CCs when using standard classifiers. Our approach identifies twelve predictive features for developing CCs and demonstrates predictive skill for 0 - 48 hours prior to development. The results confirm that the proposed technique can satisfactorily identify developing CCs for each of the nine forecasts using standard classifiers such as Classification and Regression Trees (CART), neural networks, and support vector machines (SVM) and ten-fold cross validation. These results are based on the geometric mean values and are further verified using seven case studies such as Hurricane Katrina (2005). These results demonstrate that our proposed approach could potentially improve weather prediction and provide advance notice of a developing CC by using solely gridded satellite data.

Share

COinS