No abstract

The author would like to thank Dr. John N. Rayner for introducing him to the principles of numerical modelling and for providing advice and encouragement throughout this project. He would also like to thank Dr. A.J. Arnfield for his assistance in the development of the techniques used to simulate the radiative fluxes and for his helpful comments concerning the manuscript. Thanks are also due to the Department of Geography and to the Instruction and Research Computing Center for providing the computational resources necessary to complete this project. In addition, the author wants to express his appreciation to Betty Parker for typing the manuscript. Finally, he would like to thank all his friends for the words of encouragement that they gave him during this project.

A new dataset of tropical cloud clusters, which formed or propagated over the Atlantic basin during the 1998-2000 hurricane seasons, is used to develop a probabilistic prediction system for tropical cyclogenesis (TCG). Using data from the National Centers for Environmental Prediction (NCEP)-National Center for Atmospheric Research (NCAR) reanalysis (NNR), eight large-scale predictors are calculated at every 6-h interval of a cluster's life cycle. Discriminant analysis is then used to find a linear combination of the predictors that best separates the developing cloud clusters (those that became tropical depressions) and nondeveloping systems. Classification results are analyzed via composite and case study points of view. Despite the linear nature of the classification technique, the forecast system yields useful probabilistic forecasts for the vast majority of the hurricane season. The daily genesis potential (DGP) and latitude predictors are found to be the most significant at nearly all forecast times. Composite results show that if the probability of development P Ͻ 0.7, TCG rarely occurs; if P Ͼ 0.9, genesis occurs about 40% of the time. A case study of Tropical Depression Keith (2000) illustrates the ability of the forecast system to detect the evolution of the large-scale environment from an unfavorable to favorable one. An additional case study of an early-season nondeveloping cluster demonstrates some of the shortcomings of the system and suggests possible ways of mitigating them.

A binary neural network classifier is evaluated against linear discriminant analysis within the framework of a statistical model for forecasting tropical cyclogenesis (TCG). A dataset consisting of potential developing cloud clusters that formed during the 1998–2001 Atlantic hurricane seasons is used in conjunction with eight large-scale predictors of TCG. Each predictor value is calculated at analysis time. The model yields 6–48-h probability forecasts for genesis at 6-h intervals. Results consistently show that the neural network classifier performs comparably to or better than linear discriminant analysis on all performance measures examined, including probability of detection, Heidke skill score, and forecast reliability. Two case studies are presented to investigate model performance and the feasibility of adapting the model to operational forecast use.

A one-way coupled atmospheric-lake modeling system was developed to generate short-term, mesoscale lake circulation, water level, and temperature forecasts for Lake Erie. The coupled system consisted of the semioperational versions of the Pennsylvania State University-National Center for Atmospheric Research threedimensional, mesoscale meteorological model (MM4), and the three-dimensional lake circulation model of the Great Lakes Forecasting System (GLFS). The coupled system was tested using archived MM4 36-h forecasts for three cases during 1992 and 1993. The cases were chosen to demonstrate and evaluate the forecasts produced by the coupled system during severe lake conditions and at different stages in the lake's annual thermal cycle. For each case, the lake model was run for 36 h using surface heat and momentum fluxes derived from MM4's hourly meteorological forecasts and surface water temperatures from the lake model. Evaluations of the lake forecasts were conducted by comparing forecasts to observations and lake model hindcasts. Lake temperatures were generally predicted well by the coupled system. Below the surface, the forecasts depicted the evolution of the lake's thermal structure, although not as rapidly as in the hindcasts. The greatest shortcomings were in the predictions of peak water levels and times of occurrence. The deficiencies in the lake forecasts were related primarily to wind direction errors and underestimation of surface wind speeds by the atmospheric model. The three cases demonstrated both the potential and limitations of daily high-resolution lake forecasts for the Great Lakes. Twice daily or more frequent lake forecasts are now feasible for Lake Erie with the operational implementation of mesoscale atmospheric models such as the U.S. National Weather Service's Eta Model and Rapid Update Cycle.

No abstract

An alternative 24-h statistical hurricane intensity model is presented and verified for 13 hurricanes during the 2004-05 seasons. The model uses a new method involving a discriminant function analysis (DFA) to select from a collection of multiple regression equations. These equations were developed to predict the future 24-h wind speed increase and the 24-h pressure drop that were constructed from a dataset of 103 hurricanes from 1988 to 2003 that utilized 25 predictors of rapid intensification. The accuracy of the 24-h wind speed increase models was tested and compared with the official National Hurricane Center (NHC) 24-h intensity forecasts, which are currently more accurate on average than other 24-h intensity models. Individual performances are shown for Hurricanes Charley (2004) and Katrina (2005) along with a summary of all 13 hurricanes in the study. The average error for the 24-h wind speed increase models was 11.83 kt (1 kt ϭ 0.5144 m s Ϫ1 ) for the DFA-selected models and 12.53 kt for the official NHC forecast. When the DFA used the correctly selected model (CSM) for the same cases, the average error was 8.47 kt. For the 24-h pressure reduction models, the average error was 7.33 hPa for the DFA-selected models, and 5.85 hPa for the CSM. This shows that the DFA performed well against the NHC, but improvements can still be made to make the accuracy even better.

No abstract