Trom

When control and opp differ all it really tells you is that the difference is due to high resolution given the two runs have the same initial data for their starting points. So control is there to give us an idea of how reliable the opp run is. If the two mirror each other then the model is arriving at the same conclusions regardless of resolution. So using the data you posted you can argue that control supports the operational run throughout (not of course to say they are correct but it does increase confidence in the opp run). When they diverge though opting for the control over the opp is really saying you are betting on lower resolution vs higher resolution.

Nice to see the operational and control in relative tandem. Gives you a bit more confidence when the op is tracked by its low resolution brother.

It's worth noting a couple of points regarding verification stats. 1. They are for the entire northern hemisphere. The polar areas and the equator are far more consistent over time and as a result the models have very high verification stats for these areas (especially the continent). The mid latitude areas are far more changeable and the verification stats are far lower. There's a huge variation in model performance in different areas of the northern hemisphere. Catacol posted some really good data on this last January. 2. Not only are we at mid latitude we are also on the boundary of the Atlantic and continental Europe which makes our weather even more variable and difficult to forecast. So verification stats for the UK would be far far lower than those plots suggest. I found Catacols post from last year and have copied it below (the second image is really enlightening) Here's the link to Catacol's original post

Root Mean Square Error is the standard deviation of observed values around forecast values for out of sample data. So if the models were perfect they would have a RMSE of zero. Are you confusing this with correlation which some verification data uses?

The models are medium range models so typically lack the resolution for a T0 rain forecast. Better sticking to higher resolution short range modelling if you want a higher level of accuracy.

Hello Mushy just trying to get my head around the anomalies. I can interpret the charts but was a little uncertain regarding the source of the data. They're essential 4 and 6 day rolling means taking data from GFS and ECM are they not? How many days of model runs goes into each anomaly chart and if it's more than one is it weighted in anyway? So do they show an averaging effect due to the use of multiple models, the fact that they span 4 and 6 days and the impact of multiple model runs? Presumably being means over 4 and 6 days periods they are less likely to be influenced by a model showing run to run variation/volatility. I guess of the flip side if a model correctly picks up on a directional change they will also be slower to show it. Is that a correct interpretation?

How exactly is median calculated. I know it's the middle observation when observations are ordered from lowest to highest but how would this work with ensembles? What defines lowest and highest value?

The idea that the GFS verification stats are way behind the other models just is not true. It does perform marginally worse than ECM but not by a lot. Also you need to be aware that when looking at the verification stats you are seeing the entire Northern Hemisphere from the equator to the poles and we know these areas are easy to forecast than the mid latitudes. So when you see those 0.9 to 0.8 correlations they are not true of the UK. Also as Catacol showed in a post the other day the UK and Atlantic to our west has the worst correlations of all. So I would say all the models struggle with the UK and the verification stats do not show the true picture for the UK which would be much much lower.

The standard deviation on the 06Z ensembles showed the North West to be the area of greatest uncertainty. So no surprise the 12Z op shows differences in this region. The good news was that despite the variability the ensembles showed in the North West many still produced cold for the UK. The 12Z ensembles still show the greatest variability at +168 in the Greenland area.

Yes at T84 (3.5 days) the uncertainty/volatility in the ensembles increases significantly over the NE US and Canada. That volatility transitions into Greenland around day 6.

Interesting to see such agreement at 168hrs across the suite. Op has support from control and large proportion of ensembles. Ensemble mean: Standard deviation in the ensembles: So across the ensemble members we have the most variability to our North West and relatively little to our East. It will be interesting to see if it garners any support from other models on the 12Z or even itself in later runs.

It's worth remembering that the models are pretty good at forecasting the poles and equator but much worse at modelling the mid latitudes. The models accuracy for the UK being next to a big pool of water and mid lat was woeful in a post Catacol made prior to Christmas. It was a bit of an eye opener for me. ECM had a -0.2 to 0.2 correlation of anomalies to actual for the UK, which is more or less random. So those verification stats are skewed heavily by the equator and poles. Here's a link to Catacol's post Basically the models are better in winter than the summer - this seems to be due to greater confidence in predicting the polar regions (presumably due to the vortex). The models are great for the equator and decent for the polar regions in winter. Poor for mid lats and woeful for the UK.

Interesting post. The RMSE clearly shows the models are better in summer and worse in winter. Low RMSE meaning a good match between forecast and actual and higher being poorer. I wasn't sure I was interpreting the anomaly correlations correctly. The UK appears to be in a grey zone indicating -0.2 to 0.2. Those correlations look woeful but I wondered if I was missing something. It appears the models are great at the equator and reasonable at the poles but horrible in the mid latitudes. Is this the correct interpretation?

But the control is just a low res version of the operating run. Surely it's purpose is to support or not support the operating run.

Yes, still the effects of last years SSW? We really haven't seen a zonal regime since last winter.

Whilst I was leafing through the data on it I came across the statement that ICON is higher resolution than ECM at 7km rather 14km for Europe (globally I think it's 13km). Is the 14km still true of ECM?

That and control and op follow a similar path with the same initial data and only different resolution. Does the op still drop in resolution later in the run, or is it now run in the higher resolution for the entire run? Still if you look at the story from tonight's models then it's a minor fly in the ointment. Still looks a little dicey at time regarding the evolution on all models with a few near misses and tomorrow is of course another day but overall you'd say the pattern looks epic. Of course we are a small island next to a large body of water so difficult to model when the Atlantic is as uncharacteristically quiet as it has been - not the norm. I know the argument is the models just run algorithms based on Boyles law and a sample of initial starting data but they do retro fit using empirical data and of course there is less precedent for these set ups so you would think that would impact accuracy.

Well said Mike. I always wonder why it gets posted in isolation. Just the operating run initial data run at low res. If you back it over the op then you are really saying the higher resolution of the operating run is leading to less accuracy. Surely its main purpose is to support the operating run if it shows a similar evolution.

Encouraging that ECM showed a similar direction earlier too.

Very decent uppers edging in at 300 hrs

No doubt that the direction of travel has been positive today. We've gone from predominantly SW/W weather directions across all models out into deepest FI to height rises on both GFS and ECM. Also over the day each run has looked to build on the previous (GFS). Still all really out in FI but significantly better than the last 6 days of model watching if you want a cold blast. Lets hope we can see these heights still being modelled on tomorrows runs. 18Z looking like further improvements at early dates

The models are showing the opposite of what you are suggesting. Until today (since the cold snap) all have shown constant southerlies and westerlies with the occasional north westerlies caused by lows passing to the north. Today is the first day where they've suggested height rises and possible cold encroachments from the North East - East. It's all still a long way out and therefore subject to all the volatility and variability attached to that but the first signs of potential are showing.

GFS and parallel give variations on a theme of South to West based winds throughout the run. UKMO similar. Not a lot of cheer if you are looking for a cold spell.

Are you talking about the low pressure features from Canada that the models show phasing with the main low in the mid Atlantic? Or something else?

Just the South East and Northern Ireland seeing falling snow tonight.