The Atlantic has already produced it’s first hurricane-strength tropical cyclone for 2016 in January. However, the bulk of tropical cyclone activity comes during hurricane season which is defined by the National Hurricane Center as the period between 1 June and 30 November each year. For the past two years, I have provided an analog forecast for the upcoming hurricane season based on January and February weather in Bermuda.
This year, I hope to take a closer look and try to better quantify the correlation between Jan-Feb temperature/precipitation in Bermuda with the following hurricane season’s activity.
First, let’s take a look back at the last two year’s forecasts:
|Year||Named Storms (Actual)||Hurricanes (Actual)||Major Hurricanes (Actual)||ACE (Actual)|
|2014||9-10 (8)||5-6 (6)||2 (2)||74.4-78.6 (66.7)|
|2015||10-12 (11)||6 (4)||2 (2)||84.7-104.0 (65.3)|
^Actual activity sourced from National Hurricane Center Tropical Cyclone Reports.
Comparing the mean Jan-Feb temperatures from 1949-2015 to the number of named storms (NS), a measure of total seasonal activity, there is only a weak correlation. Similarly, there is no significant relationship between Jan-Feb total rain and NS. Using a linear model to combine the both temperature and rainfall to predict NS, there is only a weak predictive relationship that fails to describe the majority of the season-to-season variability in NS.
It appears the 2014 and 2015 forecasts were so close to reality because they happened to end with near normal activity. Neither, temperature, precipitation nor the combined model were able to produce a forecast far from the average NS and described less than 5% of the variability in NS. Further, the weak relationship between temperatures and NS is probably only a reflection of the impact of sea surface temperatures on NS. It seems that warmer and wetter winters in Bermuda tend to precede less active seasons while cooler and drier winters precede more active seasons. However, other phenomena are likely controlling the larger variability superimposed on this tendency.
So as expected, this “point analog” forecasting method really doesn’t have any skill. Like the adage goes, a broken clock is right twice a day. For what it’s worth, using this method would suggest 2016 ends with 9 named storms, 5 hurricanes, and 1 major hurricane for an ACE of 72. This is a less active than normal season, where a normal season in this dataset has 11.5 named storms, 6.2 hurricanes, and 2.6 major hurricanes with an ACE of 100.6.