Characterizing Hurricane Tracks Using Multiple Statistical Metrics

Abstract:

Historical tropical cyclone tracks reveal a wide range of shapes and speeds over different ocean basins. However, they have only been accurately recorded in the last few decades, limiting their representativeness to only a subset of possible tracks in a changing large-scale environment. Taking into account various climate conditions, synthetic tracks can be generated to produce a much larger sample of cyclone tracks to understand variability of cyclone activity and assess future changes. To evaluate how well the synthetic tracks capture the characteristics of the historical tracks, several statistical metrics have been developed to characterize and compare their shapes and movements. In one metric, the probability density functions of storm locations are estimated by modeling the position of the storms as a Markov chain. Another metric is constructed to capture the mutual information between two variables such as velocity and curvature. These metrics are then applied to the synthetic and historical tracks to determine if the latter are plausibly a subset of the former. Bootstrap sampling is used in applying the metrics to the synthetic tracks to accurately compare them with the historical tracks given the large sample size difference. If we confirm that the synthetic tracks capture the variability of the historical ones, high confidence intervals can be determined from the much larger set of synthetic tracks to look for highly unusual tracks and to assess their probability of occurrence.