Perspectives on online change-detection techniques deployed to model atmospheric anomalies

The menacing rate at which strong hurricanes get formed and the damages they inflict on coastal regions have posed a headache to governments around the globe. Studies track this and offer speculations as to the cause. Locating the time stamps around which frequencies begin to fluctuate has, however, proved to be quite complex. We promote and investigate the applicability of a recent sequential testing approach to pin-point locations of such structural breaks. Deviations from a null assumption of stationarity (in favour of a self-exciting class) will be checked through a novel statistic constructed through time-reversal. We will demonstrate reliable classification and estimation power, in addition to the controlling of false alarms, through extensive simulations, and will characterise intensities under which our proposal outperforms its traditional competitors. Brownian-bridge based goodness-of-fit tests will confirm model justifiability. Clustering of oceanic basins through metrics such as the Hausdorff will offer crucial insights on how the North Atlantic’s evolution closely mimics the South Indian’s or the South Pacific’s. We explain how change-point-induced partitions can be exploited to offer forecasts such as seventeen probable hurricanes and tropical storms in the North Atlantic or thirty in the West Pacific next year. Bootstrapped intervals are offered to quantify volatilities and uncertainties. Generalisations to non-hurricane contexts will be proposed.