Feasibility of Using Classification Analyses to Determine Tropical Cyclone Rapid Intensification
Master's thesis Jun 2003-Mar 2004
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING AND MANAGEMENT
Pagination or Media Count:
Tropical cyclone intensity techniques developed by Dvorak have thus far been regarded by tropical meteorologists as the best identification and forecast schemes available using satellite imagery. However, in recent years, several ideologies have arisen which discuss alternative means of determining typhoon rapid intensification or weakening in the Pacific. These theories include examining channel outflow patterns, potential vorticity superposition and anomalies, tropical upper tropospheric trough interactions, environmental influences, and upper tropospheric flow transitions. It is now possible to data mine these atmospheric parameters thought partly responsible for typhoon rapid intensification and weakening to validate their usefulness in the forecast process. Using the latest data mining software tools, this study used components of NOGAPS analyses along with selected atmospheric and climatological predictors in classification analyses to create conditional forecast decision trees. The results of the classification model show an approximate R2 of 0.68 with percent error misclassifications of 13.5 for rapidly weakening typhoon events and 21.8 for rapidly intensifying typhoon events. In addition, a merged set of suggested forecast splitting rules was developed. By using the three most accurate predictors from both intensifying and weakening storms, the results validate the notion that multiple parameters are responsible for rapid changes in typhoon development.