Winter 2013/14 - A short (sympathetic) note on seasonal forecasting

The Met Office recently announced that it's already - with February still unfinished - the wettest winter (December, January and February) on record. A whopping 486.8mm that beats the previous record of the 1995 winter. I plan to come back to these two winters in particular in future as both were wet without a particularly strong windstorm present that might mark out a stormy winter, but both were more notable for their wetness than their extreme windiness (like, say, the winters of 1990 and 1999).

What interested me in particular is the Met Office's contingency forecast for December, January and February for rainfall for the forthcoming winter from late November. This isn't a seasonal forecasting model-bashing exercise, it's more to highlight not only what happened - but throwing the question out there on just how on earth you can forecast the really extreme winters - be in cold, wet or windy - the ones that have a real impact. The ones that (re)insurers would benefit knowing about if there is some predictability.

I've put the November forecast below - any annotations I've made are in red. I should add that I'm a huge fan of how the Met Office go about presenting this data - it's great way of seeing the spread of data for the forthcoming period for a particular variable as well as seeing what has gone before to compare what the models are predicting.

Met Office Dec/Jan/Feb forecast for precipitation

How did the forecast do? On the face of it, pretty awfully. There is no hint from the distribution compared to previous years that an overly-wet winter was likely - if anything the hints were for a spectacularly average winter. It goes to show how far we have to come when it comes to winter forecasting.

I'm not going to pontificate about what may have gone wrong and what might not have been resolved by the Met Office's global forecasting models, but it highlights how difficult it is - but also how dangerous it might be to rely on longer-range forecasts for business decisions.

Ultimately I have sympathy for the people producing such products as for every good forecast (e.g. the negative NAO winters of the late 2000s/early 2010s were quite well signposted, albeit in hindcasts) there always the extreme spell that it'll be ridiculously difficult to expect such models to point to. I do end on a cheery note though - look at that one ensemble member on the right hand. One model run pointed to a record wet winter. Somebody give that ensemble member an award !

Data shown here was taken from the Met Office website.