I received an email the other day from Ty Black, a former University of Southern Mississippi professor in Economics, regarding seasonal hurricane predictions.
On the heels of the Colorado State seasonal prediction, national media outlets went running stories calling for an “above-average” year for hurricanes.
This must’ve got Ty thinking.
Page 4A of the Friday, April 6, 2018, edition of USA Today contains an article entitled “It’s Going to be Another Busy, Above Average Hurricane Season.” The piece had a chart comparing Colorado State University’s (CSU’s) hurricane predictions to actual hurricanes for the 18-year period 2000-2017 plus the 2018 forecast of 7 hurricanes. On reading the article three thoughts came to mind. First, would 7 hurricanes constitute a “busy, above average” hurricane season; second, how credible are CSU’s hurricane predictions; and third, have their predictions become more accurate over time?
For the first question I used NOAA hurricane data going back to 1980. From 1980 to 2017 there were 248 hurricanes for a 38-year average of 6.5 and a median of 7. A 7 hurricane season occurred 6 times and tied with a 4 hurricane season for the most frequent number. So a 7-hurricane season is a relatively normal occurrence, and not a particularly busy one.
Turning to the credence of the forecasts: the CSU website says, “The [forecasting] team bases its forecasts on over 60 years of historical data that include Atlantic sea surface temperatures, sea level pressures, vertical wind shear levels “…”, El Nino “…”, and other factors.” Their model is a multi-variable, complex forecasting tool.
Table 1 lists CSU’s hurricane predictions as well as the actual number of hurricanes for 2000-2017. (These are the data contained in the USA Today article.) The final column of the table shows how much actual numbers deviated from the predictions. For 8 of the 17 years predictions exceeded actual hurricanes, while for 9 of the years actual hurricanes exceeded forecast levels. The larger errors tended to be under predictions (2005, 2012, and 2017). Deviations for the entire period totaled 58 for an 18-year average deviation of 3.22.
For benchmarks against which to judge the efficacy of the CSU model, I used the average number of hurricanes for the prior 20 years. Table 2 displays the results for 2000-2018. Thus, the predicted hurricane level for 2000 is based on the average number of hurricanes for the 20 year period 1980-1999. For 2001 the forecast is based on the hurricane average for 1981-2000, and so on for the remainder of the years in the table. (Statisticians label this method a “moving average”.) I similarly made calculations using 10- and 5-year moving averages (tables not included here).
Table 3 compares the CSU results with predictions based on 20-, 10-, and 5-year moving averages. All three moving-average deviations totals were smaller than the CSU result. Correspondingly, so were the moving-average means. For example, the 20-year moving average mean deviation was 2.78 versus 3.22 for the CSU model. For the 18 years, 2000-2017, the moving average was more accurate 9 times versus 7 times for the CSU model and two ties. It is hardly a ringing validation of the complex CSU model that simple, moving averages outperformed it.
For the final question of whether the accuracy of the CSU forecasts has been improving, Table 4 divides the 18-year study period into 6-year segments and tracks the results of the CSU model. By a considerable margin the latest forecast segment, 2012-2017, underperformed the previous two segments. So improved accuracy has not occurred in recent years.
In this note I have calculated moving-average hurricane data to serve as a benchmark for evaluating the efficacy of Colorado State University’s early-season forecasts. I am not, however, arguing that moving-averages are a viable hurricane forecasting technique. I applaud CSU’s and other hurricane researchers for their efforts to enhance understanding of hurricane dynamics. At the present time, however, the forces that drive the annual level of hurricane activity appear to be more complex than the researchers’ techniques are able to capture adequately. Therefore at present it is probably wise to take annual hurricane frequency predictions with a grain of salt.
We often do down here. Historically because I’ve always subscribed to the thinking that even a slow year with a hurricane that hits you, is an active year… For you.
But now we can ignore these seasonal forecasts for another reason, too.
