RogerBW's Blog

I tend to use the weather forecasts from the Met. Office as I find them more accurate than the BBC ones (even when the BBC is using Met. Office data). However, I find the Severe Weather Warnings rather less useful.

Some of this is presentation. There's a colour-coded scale of yellow, orange ("amber"), and red for "Be aware", "Be prepared", "Take action". ("Amber" and "yellow" should never be in the same scale!) But that scale is itself the result of a two-axis evaluation, of "likelihood" vs "impact" (each on a four-point scale). A very unlikely but highly impactful event makes for a yellow warning; so does an absolutely certain but only somewhat impactful event.

(image from the Met. Office)

Now I'm one of the few fans of prediction spreads in weather forecasting; I like the PROB30/PROB40 clauses used in aviation weather forecasts ("there is a 30% chance of this happening") and wish their use would be expanded. But I think that rolling them into a severity measure like this, in the manner of a risk assessment, is something of an abuse of the approach: if you have decided to act on a yellow warning, what you might do in response to a very high probability of a heavy rain is surely different from what you'd do if presented with a tiny chance of Noah's Flood Mk II.

(And there are "warnings" and "alerts", which seem to be separate things, but there's no definition of which is which.)

Then there's the question of what you should do. For anything short of a red warning, this comes down to "pay attention to weather forecasts, and expect trouble, but don't actually do anything unusual yourself". Actually it's worse than that, because the table claims to be based on the impact level (Very Low, Low, Medium, High) but it's indexed by alert colour – and the alert colour only tells you the composite measure of likelihood and impact level. So which is it, guys?

What you should actually do, in the unlikely event of a red warning (or a high impact level?), is listed on a separate set of pages, and there finally we have some reasonable advice. It may seem basic and obvious, but it's clearly not obvious to everyone. (If you're waiting to lay in stocks of emergency food until the emergency has been declared, you're naffed, of course, because everyone else will be trying to buy them too.)

But I have been receiving Met. Office warnings for all of south-east England (the best granularity they offer) for several years, and I have never once seen a red warning. On the other hand there are yellow warnings nearly every week (total of 772 warnings/alerts in 4.5 years, though often there are several successive warnings/alerts about a single event)… of which the vast majority, naturally, come to nothing. There have been 27 amber warnings/alerts in that time, but I think only one or two of them had a direct impact on me. I can see why they warn about everything they think might be dangerous; they're scared of the Michael Fish in 1987 effect, and these days probably of being sued too. On the other hand there certainly has been severe weather (several occasions of wind) that did affect me, and which wasn't warned about. All of this means I'm less and less likely to take the things seriously.

So how would I do it differently, without magically becoming more accurate in the forecasting? First, split likelihood and impact, and keep them split. Second, remove the yellow warnings completely: issue warnings only when there's a substantial chance of significant effects. Third, allow more granularity in the warnings: I don't need to hear about storms hitting Brighton if they aren't going to reach me in Bucks. (On the other hand I'd like to be able to check for alerts along a rough line between home and wherever I'm planning to go that day.)