An examination of the potential governing factors of tropical cyclone intensity

Abstract

This study has investigated methods for predicting the intensity of tropical cyclone (TC). The investigation was conducted through three interrelated research projects. The first project has re-examined the relationship between empirical maximum potential intensity (MPI) and sea surface temperature (SST). The result indicates that the empirical MPI-SST relation is basin specific and sensitive to both the study period and data source. Moreover, the analysis using bootstrap random sampling shows that the observed maximum potential intensity is not robust, because the true maximum is never observed if the number of observations in a SST bin is small. The 99th percentile intensity, on the other hand, is a robust quantity. Hence, the 99th percentile intensity-SST relation is proposed as the replacement of empirical MPI-SST relation. The second project has investigated the asymmetry parameter of TC. The existing asymmetry parameter, Deviation Angle Variance (DAV), which was calculated using local gradient of the brightness temperature field in infrared satellite images within a predefined radius of calculation (ROC), is shown to be sensitive to fluctuations within the brightness temperature field. Furthermore, in the presence of fluctuations, DAV cannot differentiate between symmetric and asymmetric TCs. Consequently, a different asymmetry parameter, Galaxy Asymmetry parameter (GASYM), is proposed. The results show that GASYM is less sensitive to fluctuations when compared to DAV. Moreover, to address the problem of ROC, the cluster identification (CI) method is proposed, which uses density-based spatial clustering algorithm. The CI method has the ability to identify TC cloud cluster from infrared satellite images. Thus, calculations of the asymmetry parameter can be achieved without using the ROC method. Nevertheless, the relationship between asymmetry and intensity has the shape of sigmoid function as well as large scatters, which are in agreement with results in the literature. This suggests that using asymmetry alone to determine intensity would not be accurate. The third project has assessed the performance of model is constructed using partial least squares (PLS) regression for the application of intensity projection for future climate. PLS is one of the methods that properly handle multicollinearity in the predictors. Common important predictors are identified from the intensity estimation model. Furthermore, predictors that are calculated using large area averages were shown to be not suitable to use because useful information could be removed and hence leading to poor model outputs. The performance of the intensity projection model was evaluated by two sets of data. The first set contains observations in the period of 1979-1988, and the other set contains observations in the period of 1999-2008. The statistical model was trained by one of the sets and tested by the other set. The maximum uncertainty of this statistical model is found to be around 12 m/s. This suggests that although statistical model could be a useful tool to assess the intensity of TC in a different climate state, the uncertainty could be large and it should be used with caution.