Northern hemisphere tropical cyclones during the quasi-El Niño of late 2014

Abstract

During the second half of 2014, the tropical Pacific was in a state marginally consistent with El Niño. While oceanic indicators were indicative of a weak El Niño event, a number of atmospheric indicators were not, and a number of forecast centers did not declare an El Niño. Nonetheless, the most active tropical cyclone basins of the northern hemisphere—those of the North Atlantic and Pacific—showed tropical cyclone statistics that in some respects were consistent with El Niño. In particular, the numbers of relatively intense storms in the four basins considered—major hurricanes in the Eastern North Pacific and North Atlantic, super typhoons in the Western North Pacific, and hurricanes in the Central North Pacific—formed a pattern strongly consistent with El Niño.

Appendix: Logistic regression model

Using the standard logistic regression model, the probability that a given year is an El Niño is formulated as

$$P = [1+ \exp (-b_0-b_1 x_1-b_2 x_2-b_3 x_3 -b_4 x_4-b_5 x_5)]^{-1}$$

where $x_1, x_2, x_3, x_4$ are the number of super typhoons in the Western North Pacific, the number of hurricanes in the Central Pacific, the number of major hurricanes in the Eastern Pacific, and the number of major hurricanes in the Northern Atlantic , while $x_5$ is an interaction term computed as $(x_1 x_2 x_3)/(x_4+1)$, with the addition of 1 in the denominator to prevent division by zero in years with no major Atlantic hurricanes. The coefficients are determined in matlab by the function “glmfit” and are

$$(b_0, b_1, b_2, b_3, b_4, b_5)=(-3.1133, 0.3593, 0.4549, -0.0912, -0.2297, 0.0275),$$

if the year 1992 is included. If 1992 is excluded, we get

$$(b_0, b_1, b_2, b_3, b_4, b_5)=-3.3137, 0.2938, 0.2206, -0.2312, -0.0827, 0.0753.$$

The signs of most of the coefficients are consistent with our expectations based on our understanding of the influence of ENSO on TCs. The exception is $b_3$, the eastern Pacific coefficient, which is highly uncertain (essentially indistinguishable from zero); this can be understood from Fig. 3c, which shows that several neutral and La Niña years had large numbers of major hurricanes in the Eastern North Pacific. The wide confidence intervals on the estimates of the probability that 2014 was an El Niño year (see Sect. 3.4) reflect the uncertainties on the coefficients in the model. As this might be due to collinearity between the predictors, we examined alternate models using fewer predictors (but retaining the Western and Central North Pacific). These did not lead to differences large enough to change our interpretation, for either the central probability estimate (which remains above 90 % in all cases tested) or the width of the confidence interval.