Below is my attempt at answering one of the questions asked in the Good Judgment Open forecasting tournament. In full, the question reads:
How many total hurricanes will occur in the Atlantic Ocean in the 2022 hurricane season, according to the National Hurricane Center?Good Judgment (c) Open
My answer is based on three elements:
An historic record of observed hurricanes in the Atlantic basin
A trend in that record
A correlation between hurricane occurrence and sea surface temperature anomalies.
The National Hurricane Centre keeps a record of the number of hurricanes observed in the Atlantic basin. The record goes back as far as 1851. It doesn’t include the years 2016 through 2021. The numbers for these years, however, can be found on Wikipedia. These can be plotted as timeseries:
When looking at the original timeseries, however, I suspected there was a trend. This is indeed the case:
(Aside: the trend can have various reasons including an incomplete record of observation -there weren’t any satellites to identify hurricanes in 1851- and a change in climate.)
I corrected the timeseries to account for the trend. The trend line ranges from 4.26 to 5.52 to 6.79 in 1851 (start of the record), in 1936 (median year) and in 2022. I simply multiplied the entire record by the ratio of the values in 2022 and in 1936: 6.79 / 5.52 = 1.23. I multiplied the entire record by the ratio of the trendline value in 2022 and that of each of the years. The timeseries then looks like this:
There is a correlation between the number of hurricanes and the observed sea surface water temperature. The latter is expressed as an anomaly, i.e. a difference against some baseline. Plotted as a timeseries:
The correlation between the number of hurricanes and the SSTA isn’t very strong but it is all I have:
Note that sea surface temperature anomalies are available only since 1951 hence the data record has now been shortened by approx. 100 years.
Subsequently, I am going to condition the record on three conditions: SSTA in the lowest, the middle and the highest tercile, respectively. From the SSTA timeseries, I estimate that SSTA will be in the lowest tercile in 2022. The historic years where this was the case also give me the following distribution:
Expressed as fractions:
and this is the “posterior” distribution I am looking for. In tabular view: