Analogs Vs. Models but can both be right?
By now, those that need a winter outlook for their business have probably been exposed to more than a few thoughts, with the prevailing guidance from meteorologists leaning onto the cooler side due to a number of leading teleconnections supporting the risk of stronger blocking during winter. This can be viewed by choosing a bunch of similar years (what I/others call analogs) and plotting the resulting composite mean as a guide to what the season ahead may bring.
The other forecasting tool, which is more commonly used by the market is from computer model guidance, the favourite of which being the ECMWF, as on average it has the most skill out of all the computer models available.
As it currently stands, the latest ECMWF model guidance suggests a generally warm outlook across much of the Northern Hemisphere into this winter, except W Canada/Alaska and also more seasonal temperatures for mainland Europe/Southern Russia into E China/Korea, this is actually a fair bit 'cooler' than the previous update in September where it had seasonal temperatures in Canada and Korea but widely warm elsewhere but nonetheless the model guidance is still much warmer than the analogs which suggest there is a cool risk for especially Europe/E Asia/Siberia but depends on who you ask when looking at E USA.
Example Analog Composite mean for DJF on the left and ECWMF forecast on the right
This has led to the assumption of 'Analogs vs. Model' as a battle for this winter, or a case of 'ignoring the models' when producing a winter outlook but I believe it’s wrong to say models are often wrong lets ignore them – it’s important to note what they can tell us at this range – what has skill – and the parts of the forecast that have skill do actually match the cold analogs in terms of most likely pattern - as they did for the warmer update in September as well - let's explore that in a bit more detail.
Above is the 500GPH anomaly suggested by analog composite on the left and ECMWF forecast on the right. From first glance they do look quite different but as mentioned above, it's important to look at where the model has skill.
Below shows the Anomaly Correlation Skill in the ECMWF for DJF from the October Update:
The most obvious region of good forecasting skill is in the tropics - with a general lowering on the forecast skill the further North/South you travel from the Equator for the northern hemisphere there are some exceptions, being over Canada/Gulf of Alaska/Barents Sea but also some level of skill closer to the Equator across S USA/E China that is relevant for the Energy industry. The most obvious place with a lack of forecast skill is over Europe/Siberia but also N USA.
If we now highlight the areas of skill onto the Northern hemisphere forecast as shown below we can see that some of the key features that do have some skill is the Gulf of Alaska/Bering Sea strong ridge – N Canada low pressure with blocking to the North across Barents Sea (also Iceland has a bit of skill along with far E Siberia) with low pressure risk over Japan - which if you toggle the pictures below - you can see are the same features in the analog composite:
However the teleconnection analogs suggest that the E Asia low develops even stronger and helps to deliver stronger dose of cold here than the models can currently see – that also the NE Canada low troughs more into NE US (where there is no skill at this range) to deliver stronger doses of cold into the US and finally that there is stronger blocking over N Europe – low pressure over S Europe (also where there is no skill) that helps deliver a greater risk of cold here than the models current depict.
There were also a lot of similarities in the areas of forecast skill from the September forecast from the ECWMF (slightly different regions have skill from the Sept forecast) - which you can compare with the analog composite mean by toggling the above image left/right but again where the forecast differs most significantly is over E US and Europe where there is no forecast skill at this lead-time.
Another point not mentioned so far is a general lack of ability in the models to 'see' cold at longer lead times, this is due to over-riding global warming signal being a stronger signal than the forecast anomaly signal models can produce. Thus why using 500GPH anomaly can/does produce a better result than just surface temperatures alone.
One final note, the analog shown is a composite mean so whilst I believe it's fair to say the probability is stacked towards a cooler winter, it's somewhat like a loaded die and 2021-2022 is just one throw of the die which if thrown multiple times would likely yield a cooler average - it could just be that 2021-2022 is a throw that ends up warmer, in the same way that some of the years within the composite were warmer, and the probabilities are in constant adjustment based off the latest conditions - so it's vital to stay up to date with the latest developments if weather is of critical importance, as things could quickly change.
If weather updates are required through the winter please email: firstname.lastname@example.org