Cloud problems again

2016-12-04_110119

Courtesy of BBC

The graphics from the BBC weather forecast (10.58 AM) above, show just how quickly out of step the Met Office model can get within just a few hours:

  • The frontal cloud currently flooding northeast across Ireland is much more advanced than in the visible satellite image than it is in the forecast fame for 1100 UTC (underdone).
  • In the Southwest of England, although some low cloud has spread across Dartmoor and southern parts of Cornwall from the English Channel, the rest of the peninsula is having a glorious cloud free morning, although a few bits of cirrus have now started to make an appearance (overdone). I must admit when I first saw this same graphic in the 8.15 AM forecast I thought that we were in for a cloudy old day.
  • The low cloud is more extensive over the Northeast and central belt of Scotland (underdone).

I would say the Met Office have more than a few problems with their new model, especially with low cloud, but to be fair a lot of the general public will never notice these shortcomings, but this old curmudgeon did.  I Personally don’t think the BBC presenters make as much use as they should during the day of visible satellite imagery, NWP data is fine, but when it’s out of step with reality, it’s just plain misleading.

2016-12-04_111043

By the way,  the question marks in the top image are there because I’ve never clearly understood what the different shades of shadows are supposed to represent!

Why isn’t Met Office NWP data free?

the-met_landscapeview-medium

The new £20 million building to house the new supercomputer in Exeter

There’s a link on the Met Office site that gives you a menu of available NWP [numerical weather prediction] models that they produce and that you can buy, but try as I might I couldn’t find any indication of what that NWP data might cost you.

2016-11-22_073528

Courtesy of the Met Office

My simple question is: why isn’t Met Office NWP data free?

Don’t get me wrong, I don’t think that all model data should be made freely available, but certainly a subset of it should be, much like the Americans do with their Global Forecast System [GFS] model. At the same time they may like to produce a simple desktop and mobile application for the people of this fair land to view that forecast data with, because there are growing band of ‘expert users’ across the country that feel disenfranchised by the way that they are barred from being able to view what is essentially our forecast data.

Yes, I know the Met Office is a Trading Fund and part of the Department for Business, Energy and Industrial Strategy [BEIS] and expected to turn in some kind of profit each year, but with the help of a simple application they could use this free subset of NWP data to showcase what they can provide and actually generate interest and increase sales of more specific detailed forecast data from the other models that they run.

You could argue that we do see the Met Office data across all the time on our TV weather forecasts, and that’s undeniably true, but how do we know it’s the best forecast data that is being used? With that thought in mind, please take a look at the T+120 forecast for midnight (22 November 2016) from each of the three major NWP models in the world, and see how each of them have done with a forecast made 5 days ago on the 17th.

The boys from Exeter didn’t do too well did they? And they were also not too great at nailing storm Angus at T+120 either.  I didn’t cherry pick this particular example, because until an hour ago I didn’t even know I was going to write this article. Why do I use Wetterzentrale to compare NWP forecast charts? Well, they have for many years provided a website where you can examine and compare up to as many as nine different NWP models from around the world, using the same standard map projection for each. The Met Office do publish forecast fax charts out to T+120 which may be subtly different from the raw model output, but I prefer this way of doing it, after all I am comparing model data.

2016-11-22_084205I want our NWP data to be made more freely available for the sake of transparency, because I would like to know just how well our model compares with the American, French, German and European models, but as you can see that’s far from easy to do. Please don’t email with sites that do this with some fancy statistical RMS verification score because I have seen and they don’t help. My kind of verification is of the mark one eyeball – give me an analysis and any number of forecast charts and I’ll pick out the best one. I know all models perform better the higher in the atmosphere that you go, but for me it’s level zero that’s the most important one.

This article started out as another one of my Don Quixote type crusades to right what I see is a wrong – free NWP data for all. But as I wrote it, it developed more into a questioning of why are the Met Office so reluctant to free their model data, and the year on year spiralling cost of the whole thing. But that’s what blogs and blogging is all about I suppose.

If our NWP model consistently outperforms the rest, then I can see how the £97 million cost of the latest Cray XC40 supercomputer, the £20 million for the building that it sits in, and the time and effort put into its development have been fully justified, but at the moment I have my doubts that it does.

Angus – the forecast

I am still putting together an application to verify the NWP models that are available on the Internet. Namely the GFS, ECMWF and the UKMO models. I try to download the images from Wetterzentrale each day – my eternal thanks to the German guy behind that site! I am not really interested in short-term forecasts of T+72 or earlier , but in the medium range of T+96 to T+144 where things can go a little awry. And that’s how it was with Angus at T+120. Have a look at the following three solutions that were issued on the 15th of November for midnight last night (20 November), ignore the CPTEC model which is basically unstable, unstable being the technical term for crap.

2016-11-20_102159

Here’s the analysis from midnight that I use to verify the forecasts – ignore the [T+120] in the title that’s something else that needs fixing.

2016-11-20_102423

Here’s my analysis of the three models:-

  • GFS – The position is just a little too far northeast by around 1°. The minimum pressure is around +8 hPa too high and that makes Angus not quite intense enough.
  • ECMWF – The position is just too far northwest of where Angus was at midnight maybe by 1.5° in both latitude and longitude. The intensity looks just about right as I make Angus around 967 hPa at 00z.
  • UKMO – The position is too far north by around 3°. It almost looks like there is a double centre which elongates the forecast low when it really was very circular affair. The intensity was around +8 hPa too high making the center way too flabby.

The Met Office would like you to believe that forecast for 5 days ahead are as accurate as the forecasts for 2 days ahead of 20 years ago. Well on the strength of this comparison, I would say we still have some way to go. Like the Curates egg, some models are better than others, and if I had to rank the forecasts made on the 15th for last night, then I might give first place to the ECMWF model, which just nudged out the GFS model with a more accurate intensity, even though the forecast position was a little off. The Met Office came in third, which might be the reason why they prevaricated so much all last week about just how much impact this low would bring to the UK, and whether they should give it a name – who knows.

The latest 5 day forecast

This is a quick look at how four of the world’s leading NWP models are performing courtesy of Wetterzentrale. The image below is a 4 up of forecasts for each of the models for today made 120 hours ago on the 12th of November 2016.

t120-12-nov-2016

And here’s out today turned out!

2016-11-17_094024

In my new verification application they all come out with flying colours on the simple Lamb Weather Type verification, but I gave best model to the ECMWF because of some slight troughing over Sweden and south of Iceland. The Met Office model just seemed to have the pressure a little too low and the shape of the low a bit elongated in a similar way to the GFS. One up for the ECMWF. I wonder if MeteoGroup could make use of their model data when they take over at the BBC?

Anyway all the models seem in a bit of a quandary at the moment with how to handle the low that’s expected to rush across the country this weekend and exactly what track it will take. It’s no different with the forecast for T+120 forecast for next week either, here’s the latest 4-up from the midnight run for next Tuesday.

t120-17-nov-2016

The GFS and the ECMWF have a very similar solution with a quite intense low in the western part of the English Channel and the bulk of the British Isles in a very cyclonic northeasterly. The UKMO on the other have a low over Dogger with the whole of the country in a slacker northwesterly flow. Don’t ask me what’s happening to the Brazilian model [CPTEC]. I did think that it would be a good idea to include it with the other models, but having looked at some of its recent performances I am looking to replace it. The one thing you can always say about British weather is that it’s never boring.

As I said earlier, all images are courtesy of the wonderful Wetterzentrale.

The Met Office’s ‘barbecue weekend’

CnEK4dFXgAAGD4t

Courtesy of Twitter and the BBC (11 July 2016)

Numerical Weather Prediction and the social media phenomena Twitter are a wonderful thing, but if you tried to find the above image on the Met Office Twitter feed I don’t think you would be successful because it was probably deleted. The Met Office models have been swithering all week about will it or won’t it. It’s certainly going to be warm and humid in the south and that’s a fact, and that’s because the weak trailing cold front is stalled across central regions leaving the south in the dragged out remains of a warm sector. I suppose to the general public in southern areas this is exactly what was forecast, but to me it’s not the weather but the graphic, and the NWP behind the graphic that’s wrong, if this doesn’t sound too much like gobbledygook – they got it right for the wrong reasons – and how I wish now that I’d kept last weekends T+120 forecast chart!

In fact the hype and the graphics about trying to sex up a weekend from what is looking like a very normal summer is very reminiscent (in a small way) of the ‘barbecue summer‘ of 2009. The press office must look at the NWP, see that it’s looking good in the south-east for the weekend, knock up some graphics and appropriate sound bites, launch forth into the social media and TV forecasts. They may of course have been a couple of days early because it looks like that early next week the high pressure over northern France will retreat eastward and allow tropical continental air to spread up across the country for a while, before low pressure from Biscay spoils everything by mid-week.

I still maintain that NWP forecasts at T+120 and beyond are really quite poor and should always be taken with a decent pinch of salt. Perhaps even with all the computing power we have, forecasting beyond T+120 is simply just a waste of time. With that in mind I’m knocking together another application to help me verify the UKMO, GFS and ECMWF forecasts at T+120. It’ll take time to write the program, collect the data and do the verification before I can publish the results, results which although not very scientific, very few people will ever read, but that’s life!

Synops for Sat, 16 Jul 2016 at 0600 UTC

Synops for Sat, 16 Jul 2016 at 0600 UTC

UKMO T+036 0000 UTC on 16, Jul 2016

UKMO T+036 0000 UTC on 16, Jul 2016