Tropical Cyclogenesis Potential in Climate Change: An Application of Artificial Neural Networks

One of the most important consequences of warming climate is the potential change in tropical cyclogenesis. To estimate this change, we use General Circulation Model (GCM) output to obtain changes in Emanuel's revised Genesis Potential Index (GPI) for present and future climate into the 21st century. The GPI is an empirical relation representing the contribution of the following 6 environmental variables to the genesis of tropical cyclones: 850 hPa absolute vorticity, 600 hPa relative humidity, 600 hPa specific humidity at saturation, 600 hPa temperature, 850-200 hPa wind shear, and potential intensity (PI).

The self-organizing map (SOM), a non-linear classifier, is used to classify the monthly Austral Summer GCM output of the six environmental variables into different patterns that dominated the period 1960 to 2099 over the tropical North Atlantic, and to construct corresponding patterns of GPI. The analysis of the SOM results, using the monthly output of five IPCC AR4 climate models (SRES scenario A2), projected for next century a decrease in tropical cyclogenesis potential in all parts of the North Atlantic basin between 2N to 46N. The relative contribution of each environmental variable to the projected trends is determined using the SOM output as input to a forward-feeding back propagation neural network (NEVPROP4). The results indicate the increase of vertical wind shear and decrease of PI are the two major factors that contribute to the projected decrease of cyclogenesis potential in the North Atlantic basin. While the changes in vertical shear and PI contribute most significantly to the projected trends, the increase in 600hPa temperature also plays a role in the decrease of cyclogenesis potential in a zonal band between 20N to 30N in North Atlantic.