A Poisson regression model of tropical cyclogenesis for the Australian-southwest Pacific Ocean region

This paper seeks to address some of the limitations in previous statistical forecast models of tropical cyclogenesis through the development of a series of Poisson regression models on a 2° latitude × 5° longitude spatial grid and a monthly grid in time. The "Gray" parameters [low-level relative vorticity, vertical wind shear parameter, ocean thermal energy, (saturated) equivalent potential temperature gradient, and middle-troposphere humidity] were analyzed as potential predictors of tropical cyclogenesis for the Australian –southwest Pacific Ocean region. Various predictor lead times of up to 5 months were investigated, with the most significant Poisson regression models being cross validated, and the skill of their hindcasts evaluated. The Poisson regression model incorporating a combination of saturated equivalent potential temperature gradients at various leads was found to be the most skillful in hindcasting the temporal (phase and amplitude) variability of tropical cyclogenesis for the Australian –southwest Pacific region, with a correlation coefficient between the observed and cross-validated hindcast time series of 0.54 (significant at the 99% level), and a rootmean-square error 26% better than climatology. Models using the thermal (ocean thermal energy, saturated equivalent potential temperature gradient, and middle-troposphere humidity) and all (thermal plus low-level relative vorticity and vertical wind shear parameter) predictor variables showed the most skill in hindcasting the spatial distribution of cyclogenesis in this region. The model hindcast skill in predicting individual tropical cyclone occurrences and nonoccurrences was also examined. The all-Gray parameter Poisson regression model was found to correctly hindcast up to 72.6% of cyclogenesis events and nearly 70% of nonoccurrences in the Australian—southwest Pacific region. The model design enabled the investigation of tropical cyclogenesis on subregional/subseasonal scales, with promising model hindcast skill evident. The results presented herein suggest that useful and more detailed forecasts may be possible in the future in addition to those currently provided at the basin-wide and seasonal scales.