A record of the accuracy of my local cloud forecasts

[Zermelo]

Incidentally, Meteoblue seems to have had a UI refresh (either that, or I've changed some setting without realizing)

[Second Time Around]

Here's the January observations, including the first for Good to Stargaze.  As you can see, taken over the whole of the month there was more or less nothing between all 7 forecasts for accuracy.

January was a very good month for clear nights, and hugely better than the preceding two.

Edited February 5, 2022 by Second Time Around

[Second Time Around]

Here's the February data.  As you can see Good to Stargaze and Meteoblue have come out ahead.

Certainly this winter has been a lot better than the autumn. 

 

[Second Time Around]

Here's the March results.

Where I've put hazy was during an anticyclone with Saharan sand in the sky.  The result was very poor transparency.  It wasn't just here, but over a very wide area as reported by others on SGL.

 

[Second Time Around]

Here's the April data.  Elsewhere there were more clear nights in the last week.

I'd forgotten how to do a print screen with my Logitech K780 keyboard.  It's fn + del.  Then the usual Ctrl + C to copy.  Now I can refer to this post if I forget again!

 

 

Edited July 1, 2022 by Second Time Around

[Second Time Around]

An unusual result for May as Wunderground was abnormally poor for the first 3 weeks of the month.  In particular, there were far too many part cloudy nights forecast.

 

 

 

 

[jjohnson3803]

So... roughly 70% accuracy for MB while WU is worst at 58% year-to-date?  Looks like April was a tough month.

 

[Second Time Around]

The accuracy over the last year has been very similar for all the forecasts.  This is the first time there's been such variation.

However, Wunderground wasn't so wrong as it first appears. 

If you look back to earlier posts you'll see that, because some sites use symbols and some percentage cloud cover, I wrote the following:

"I have to use just cloudy, part cloudy and clear on my spreadsheet.  Where total cloud cover is shown I interpret 0-25% as clear, 26 to 74% as part cloudy, and 75 to 100% as cloudy.  This is because when cloud cover is shown as a symbol on a weather map it's based on oktas, i.e. eigths.  I know it's not precise, but it's why I said last month the methodology isn't scientific."

In many cases Wunderground forecast cloud cover of just over 25 % or just under 75%, meaning that I recorded part cloudy.  A few per cent different and Wunderground would have been correct on many of these days.

One other thing I take from the data is that few of the forecasts are totally wrong, i.e. shown in red.  I think it's human nature to remember the wrong forecasts.  However, the data doesn't support this.

[Second Time Around]

I'm doubtful that until nowcasts* become available that there's likely to be little or no statistical difference between the various forecasts.  However, I don't know if and when nowcasts will become available on each of the sites.  I'm therefore going to keep on recording the data.  If nothing else it'll be a record of clear nights here in East Kent that may be useful to others.

* See my post of 31/12/21

[Second Time Around]

 

Here's the June data.  As you can see, from being easily the worst in May, Wunderground was easily the best in June.

From today onwards I'm adding an 8th cloud forecast.  This is from Ventusky that was recommended on Cloudy Nights.

[Second Time Around]

Here in East Kent there were a lot of clear nights in June.  However, as you can see in July there was a very high proportion of part cloudy nights.  This was despite many days being largely sunny.

This is the first time I've recorded Ventusky's forecasts, so I've started adding a total for the last half year so the data can be compared.

 

[MercianDabbler]

Thanks for this. CO's incorrect promise of clear skies during the non-silly hours tonight prompted me to find some other sites for use in future and I found your thread.

Skies looked crystal clear here before sunset. Fortunately I went on previous evening's 'form' and interpreted the sight of some wispy clouds around 9PM as a bad omen and started a non astronomy activity for the evening. Sure enough the clouds rolled in. FWIW meteoblue and metcheck both predicted cloud. Not statistically significant I know but they are now on my list. I like Zoom Earth a lot too - I've been looking for a satellite cloud thing that keeps working after dark.

[Second Time Around]

Here's the August data.  I thought I'd missed the 31st August forecasts but had misfiled them.  Phew!

As you can see below many of the forecasts were less accurate this month.  This was especially so with the BBC, who recorded by far the least accurate forecasts since I started, frequently being overly optimistic.  In the last third of the month I recorded no clear nights at all, even when there was plenty of sunshine by day and further inland it was mainly clear.

 

 

 

 

Edited September 2, 2022 by Second Time Around

[Zermelo]

The forecasts here seem to have been particularly bad this last week. Unusually, they have been too  pessimistic.

[Second Time Around]

Reluctantly, I've had to stop the project as I found it difficult to keep checking forecasts and making observations every night.  It came to a head recently when we were on holiday and time was at a premium.

Anyway, there are 17 months observations without missing a day to look back on. 

The main conclusions are firstly that there's no significant difference between the various forecasts, and secondly that they're totally wrong much less than I expected.  You can confirm this by seeing how few red colours there are on the observations.

I reckon that it's human nature to remember the wrong forecasts far more than the correct ones.  I think there's a scientific term for this but can't recall it.

Anyone know?

Edited October 2, 2022 by Second Time Around

[Zermelo]

13 hours ago, Second Time Around said:

I reckon that it's human nature to remember the wrong forecasts far more than the correct ones.  I think there's a scientific term for this but can't recall it.

In my case, "confirmation bias", as I have a low opinion of weather forecasters, but you were probably thinking of a more specific phrase.

[Second Time Around]

According to Google it seems to be called "isolation effect".

"The Isolation Effect, also known as the Von Restorff Effect, proposes that one item that differs from multiple similar objects that are present, the one item that differs will be more likely to be remembered. When the item in question stands out less, the likelihood of it being remembered similarly decreases."

[Thondwe]

Just found this as noted that  DarkSky API was dying soon, so ClearOutside I expect to also have issues? 

From this much appreciated effort, seems all much of much ness across the sources, though presentation matters I guess.  Anyway, seems the Window and the "Sky Windows" - Satellite feeds are the best way to extrapolate viewing conditions!  Tend to use zoom.earth but sometimes a little hard to interpret night time cloud cover.  Met Office %age cloud cover map also looks good.

As ever, micro climate (Gower Peninsular - sticking out into Bristol channel), has a big bearing!

Thanks for the analysys!