A solid hit on seasonal forecasting

[Shane Young]

Those annual predictions? They're usually not worth the paper they're printed on. And now a large paper has hit them hard.
http://online.wsj.com/article/SB124096550135766317.html

Bout time :)

[Wes Carter]

But the greatest irony is their timing. For whatever reason, last year they all pretty much nailed it.

[IAN GIAMMANCO]

I do think they have some value, and especially in the risk analysis industry. Now the should the general public know about them?, probably not and it may cause more harm than good. 2007 is a good example, with an above average activity but only 1 US landfall (Humberto) thus the impression from the general public in the US seemed to be a below average year. These forecasts do have some statistical skill and are better (if only a little) than a coin flip. The media has latched onto seasonal forecasts so I don't think they are going away.

I think the future of this work maybe in full numerical simulations of an entire season rather than statistical modeling as our modeling and computational capabilities continue to improve.

[richhorodner]

I do think they have some value, and especially in the risk analysis industry. .

CSU's seasonal forecast for many years did NOT contain probabilities for landfall.
Then, all of a sudden they did. In fact, the CSU narrative used to state that predicting landfall probabilities was not reasonable and prone to huge error as the steering currents are too subject to chance.

Then, CSU's main source of research money started to come from the insurance industry, and still does; and all of a sudden landfall probabilities started being included in their seasonal forecasts. I did not feel that was a coincidence.

I found and still find this suspicious; as it seemed the insurance lobbies in each state began using these forecasts as a means to convince the deciding entities (legislatures, state insurance commissioners) that increases were "in order" for their states insurance rates.
---

So, I guess, in
The way CSU's forecasts lump huge sections of coastline in their forecasts also appears suspicious as a means of bolstering the insurance industry's arguments. Instead of giving probabilities for each state (which would be a low number) they say "Gulf Coast probabliity---52%, Atlantic coast probability, 62 %. If they used each state separately, the probabilities would be like 15% or 20%, at most. And in states like New York or Mass. the %s would be like 5- 10% at most, even in busy years.

These smaller %s would not "sound" convincing as an EXCUSE to raise rates, so they opted for the longer stretches of coastline; ie, numbers like 45% to 60% are easier to convince the non-scientific knowledged legislative decision makers to raise rates.

Politics and funding; great bedfellows.

A good example of how useless (except as the excuse to raise rates) these landfall probabilities are is the last three years. Hurricane numbers have been high, but there has not been a single major (cat #3 or higher) landfalling hurricane in those three years.

The number of landfalling hurricanes in a given year, or even a given hurricane, is more dependent on pure luck: on where the long wave steering currents just happen to be at the time each of the season's few hurricanes develop.

I am not saying CSU fudges its numbers. I am agreeing with their old statements that trying to pin down landfall areas or probabilities is way too subject to error because of the ever changing and variable positioning of the long wave pattern during the two months of the real hurricane season (August 15-Oct 20th) when 80+% of major hurricanes make landfall.

So, I guess, I really do agree with your statement that these forecasts do have a value for the risk analysis industry:

$ $ $ $ $ $ $