Temperature Point Comparison : GFS v Polarstern

[Niall Dollard]

There have been discussions in the main melting thread about comparing output from forecasting models and actual point data over the Arctic.

Personally, I look at two main models, the ECMWF and the GFS to give me an idea of conditions over the whole Arctic Basin, Data from GFS is more widely available and they make more runs available per 24 hours.

We currently have data from the Polarstern ship drifting in the Arctic at circa 83 N 13 E. Using the GFS data displayed on Nulschool, it is possible to do a point comparison with the data from Polarstern. (I am not aware if the GFS model assimilates these hourly synops from the Polarstern ship).

This is not intended to single out the GFS model -I have only used that model for comparison as it was most readily available on Nullschool. I am not making a statement on which model is the best.

I have noticed there can be a huge difference with some models when it comes to plotting surface temperatures over the Arctic Basin. Example the French Arpege model quite recently was showing temperatures circa -14 C over a wide area near the pole whereas both GFS and ECMWF were showing temperatures much closer to zero.

The first comparison attached is for the 48 hours on 14th May to 12th May. I hope to do some other days also - to give a longer idea over time.

. 

[oren]

Nice analysis.
Two ignorant questions:
* Is there a link for the PS synop reports? Do they have a WMO code?
* At what elevation above the ocean surface is their temp measurement taken?

[Niall Dollard]

Nice analysis.
Two ignorant questions:
* Is there a link for the PS synop reports? Do they have a WMO code?
* At what elevation above the ocean surface is their temp measurement taken?

Those are pertinent questions.

1) Link to PS data :

https://www.awi.de/fileadmin/user_upload/MET/PolarsternCoursePlot/psobsedat.html

2) I'm not certain about a WMO code. But I have seen the PS data plotted from time to time (not every hour) on https://www.weatherobs.com/

3) I don't know how high above the surface the temperature is measured. Wind measurement for weather stations should be 10m above surface on a mast. Given that Arctic is so exposed I think so long as the sensor is well removed from any manmade factors/engines would be the important thing. Given a proper exposure, for temperature, the difference at 2m, 5m or 10m would be very small (from an adiabatic lapse rate point of view). See my post below

[Niall Dollard]

Here are the previous 2 days (12th noon back to 10th noon). 

(edit: I've also added a plot of previous 5.5 days to 0 UTC on 9th May. With the temperature data labels removed).

[Niall Dollard]

More on the Polarstern data  here:

https://www.awi.de/nc/en/science/long-term-observations/atmosphere/polarstern.html

Polarstern is well equipped for meteorological research as well as for routine meteorological services. The meteorological observatory is permanently manned with a weather technician/-observer from the German Weather Service (DWD) who performs the routine 3-hourly synoptic observations and the daily upper air soundings. The measurements are coded and transferred directly to the Global Telecommunication System (GTS) by email where they contribute to weather forecasting.

[Niall Dollard]

* At what elevation above the ocean surface is their temp measurement taken?

I reckon from this, it is at a height of 29m. (I forgot it is a massive ship !)

Typical dry adiabatic lapse rate is 1 C per 100m.
 

[Freegrass]

Sea Ice Temperature on those days was approximately -20°C. Maybe GFS is discounting ice temperature, while Polarstern measurements do get influenced by that?

[Frivolousz21]

With it being 29M versus the GFS 2M temps wont that typically cause a cool bias I'm situ?

Unless an inversion keep the lowest reaches of the atmosphere stratified.

But that is still 84 feet difference roughly

[uniquorn]

Thanks Niall. If it's not too much trouble feb23rd when PS was nearest the pole might also be interesting, possibly the farthest away from any land stations.

[Niall Dollard]

With it being 29M versus the GFS 2M temps wont that typically cause a cool bias I'm situ?

Unless an inversion keep the lowest reaches of the atmosphere stratified.

But that is still 84 feet difference roughly

A typical cooling by altitude for dry air would be about 0.3 C (when the air is more saturated it would be a bit less).

Sheltering shielding by topography would be a factor over land. ie a temperature screen in a field would be expected to heat up more than one 84 ft up on a mast because of trees/hedges etc.

But of course this is the Arctic and from the pictures of the Mosaic project, the surface offers very little shelter and surface heating so I expect the difference should be no more than about 0.3 C cooler at the higher altitude.

But in the case of shallow inversions, in calm conditions it is possible to have a reverse situation whereby mist or fog cling to the surface of the ice and it could be a little warmer at 84 ft (if the fog was that shallow - allowing some sunshine to come through).

[Niall Dollard]

Thanks Niall. If it's not too much trouble feb23rd when PS was nearest the pole might also be interesting, possibly the farthest away from any land stations.

Will be next on the list. 

Another thing I would like to explore is spot forecast accuracy. Take an arctic station, not necessarily the Polarstern, say a well exposed land station like Barrow (Utqiagvik) or Alert and compare model forecasts for that Arctic spot, 2 to 7 days out. Ideally I would like to pick a station that is little influenced by topography/mountains/glaciers. 

[Phoenix]

Thank you for doing this Niall. This is a great example of resourcefulness.

[interstitial]

The test set up on the ship might introduce a bias? The ship is metal and the surface you place a temperature sensor above on land ie vegitation or blacktop effects those temperature readings. I wonder if anyone has looked for systemic bias on the polarstern sensors. Comparing temperature readings from the ship and on the ice would do it.

[uniquorn]

All of the Tbuoys are measuring surface and near surface temperatures. Pick one in the central cluster, which should be close to Polarstern, (accurate position not provided) to check for bias.

[interstitial]

I dont know but I thought there was some discussion on buoy temperature bias. I think it would bias to water temperature but I am no expert.

[interstitial]

<snip>
Tldr,
The thermistor numbers for the ice/snow (air) boundary are essentially stable over the season, at least since their deployment in October- there have been no top-melt temperatures yet

From the maximizing asif skills...   thread

[uniquorn]

I dont know but I thought there was some discussion on buoy temperature bias. I think it would bias to water temperature but I am no expert.

Perhaps you can provide us with some data on which to make our own judgements

[oren]

interstitial, the upper thermistors above the snow line should be measuring air temperatures with no significant bias (obviously those in the snow, the ice, and the water below the ice, are measuring other things). But it would be nice to compare them with the PS measurements.

[Niall Dollard]

I found this page from ESRL/PSLab on Near Real-Time Verification of Short-Term Forecasts During MOSAiC.

https://psl.noaa.gov/people/amy.solomon/MOSAiC_NRV.html

The goal of MOSAiC-NRV is to evaluate the skill of fully-coupled short-term forecasts after each leg (approximately every 2-3 months) of the MOSAiC campaign at the Polarstern location. Multi-model diagnostics focus on process-based evaluation of the coupled system to identify systematic biases that limit the skill of Arctic forecasts.

The attached image highlights the difficulties with forecasting 2m temperatures in the high Arctic even at only 2 days out.

For example at day 44 (Nov 13th?) the obs was reporting - 18 C (probably from Polarstern) with NOAA - CAFS predicting -26.5 C and ECMWF IFS predicting -23 C.

There is quite a lot of variability on these charts, comparing the models to the obs. Especially when you get to a 10 day lead.   

[Stephan]

Thank you Niall for the work to put this data altogether. To me it looks like a wild up and down and like random which model is tendentially higher or lower than others. Would you mind to calculate monthly or three-months averages to find out which model may be an outlier?

[Niall Dollard]

It's work from the Mosaic people, Stephan. Yes it would be nice to see more of this performance analysis.

As per request I did a comparison on 23rd/24th February, when Polarstern was nearest the pole (88.6 N).

Compared with the more recent comparison checks, the GFS performance wasnt that bad. Of the 17 checks, it never deviated by more than 2.5 C, and on 7 occasions differed from the Polarstern reading by only 1 C or less. 

Perhaps this was weather related, as the weather over this 48 hour period was quite similar with low pressure over Franz Josef Land steering stiff winds from a NE direction (winds were between 30 and 50 km/hr). 

[blumenkraft]

Thanks a lot, Niall. Great work.

[uniquorn]

Thanks Niall. Not really set up here for static charts but here are the buoy temps for ~feb23-24 for comparison. I thought there might be more of a 'pole hole' effect. click to play
The timings don't quite match but T56, T62 and T66 are probably closest to PS.
A more forensic mind than mine might make a better analysis.

[Niall Dollard]

I thought there might be more of a 'pole hole' effect.

By being furthest from land/other land stations ? The basin is so large with only remote sensing or buoys in the depths of polar winter to give us an idea what the 2m temperatures are.

But I think maybe it is when the wind is blowing right over "the pole hole" is the problem (and not how close the sensor is to the pole).

I wonder what meteorological conditions cause/are associated with the biggest deviation/problems ?   

The period from 9am on 10th May to 9am 12th May GFS was showing temperatures between 3 C and 5 C warmer than the Polarstern. During this period the wind was coming from a north or northwest direction with an airmass feeding across from the Chukchi Sea. That is a long track over the pole and perhaps the modelling/sensing was not good.

When the wind direction changed & backed more to the east with the wind now coming more from northern Russia/Franz Josef, the accuracy improved again (during May 13th).

Might be worth checking another period when the wind direction was coming from NNW right over the pole.

[Niall Dollard]

A graph on my longest run of data comparing GFS to Polarstern to help illustrate what my last post intended to say.

The blue bars are the deviation GFS temp minus Polarstern temp. (GFS is usually warmer, scale is on the left axis).

The red dots show the wind direction (compass points 0 to 360 deg on right axis)

In summary:

9th May : Wind is NNE. Deviation is small
10th/11th May : Wind is NNW. Deviation is lot larger. Wind is coming from over the pole. (Pole Hole Effect ? )
12th/13th May : Wind veers to the NE. Deviation lessens
14th May : Wind veers further to the east. Deviation decreases and reverses. GFS gets colder than PS for a short time.

[uniquorn]

In case anyone wants to take the comparison further, here are the mosaic Pbuoys with active surface temperature sensors. These are on the ice surface so will be biased to ice temperature.
click to play.
Tbuoy temperature animation is too tricky. Lat/Lon is messed up
Not sure I trust p151 and p152

[Niall Dollard]

@ Uniquorn (or anyone else) What do you make of my graph and the "pole hole" effect. Is it conclusive in showing that the GFS shows more temp deviation from Polarstern when the air mass is coming over the pole (ie from a NNW direction) ?

Maybe a second check at a different time/month when the synoptic situation was similar (ie with wind coming right across the pole to the Polarstern) and see if the GFS showed a similar large deviation ?

Re the Pbuoys, I don't know what to compare them with. The GFS temperature is about 2m, is it not ? The surface temps could be a lot different. I have often measured a difference of 3 C to 5 C between 2m temperatures and temperatures just on the top of a snow surface. Bigger differential when winds are calm.       

[Glen Koehler]

      Yes, temperatures in the GFS forecast are for 2M.

      FWIW - I've bee working on an algorithm to translate forecast 2M temps into temperatures closer to the surface for agricultural short-crop frost prediction.  The near to ground temperatures are much more dynamic - lower at night, higher during the day, about the same during mid-morning and early evening.  The extremity of variation depends on cloud cover, wind, relative humidity.  Winds over about 7mph (12 km/h) are enough to mix up the vertical air profile enough to overcome temperature stratification.  High cloud cover returns radiant heat to the ground at night, thus keeping near-ground temp closer to 2M temp.  High RH and moisture in the air also works against temperature stratification by height.  The most extreme differences are zero wind and clear skies.

     Of course, the ocean and ice environment is completely different than crops in soil.  I have no idea how much of this is relatable to temperature profiles in the Arctic Ocean and over the ASI.  But some of the same principles may apply.

[uniquorn]

Reasonable assumption on 'pole hole' effect and yes, needs a few more examples to be conclusive.
The thermistor buoy string starts maybe 50cm above ice from this image and data. Maybe best to stick to analysing them. I just thought some surface temps would be useful.


The first 25 thermistors of 10 or more Tbuoys provide a lot of data to calculate bias. Maybe someone else will help.(still not me   )
https://journals.ametsoc.org/doi/pdf/10.1175/JTECH-D-13-00058.1
More Tbuoy animations here. Clearly they are not all deployed at the same height.

[anthropocene]

With it being 29M versus the GFS 2M temps wont that typically cause a cool bias I'm situ?

Unless an inversion keep the lowest reaches of the atmosphere stratified.

But that is still 84 feet difference roughly

I have always argued and will continue to argue that the main cause is temperature inversion. The ice creates its own micro-climate, especially close to ground (ice) level. Friv, this is why the waves of "torches" of hot air temps hitting the ice which you regularly forecast never cause the amount of ice melt expected. A large proportion of the warm air will just be forced upwards away from the ice by the area of blocking colder air sat on the ice surface.

As further evidence for this I present the importance of melt ponds - especially early in the melting season. Melt ponds will tend to generate local sources of rising thermals which will cause the temperature inversion to break down. Therefore a large amount of melt ponds early in the melt season encourages more snow and ice melt and makes the ice more susceptible to invasions of warm air. 

My proposition is that the weather forecasting tools (e.g. GFS) do not properly take this inversion into account when presenting 2m temperature values.


 

[Niall Dollard]

My proposition is that the weather forecasting tools (e.g. GFS) do not properly take this inversion into account when presenting 2m temperature values.
 

Utqiagvik (Barrow) located as it is near Point Barrow the most northern point of the USA, sticks out into the Arctic Ocean and is probably as close as can be to a lower Arctic Ocean climate. Fogs and  stratus/low cloud are very common here especially in summer. A casual check at the Utqiagvik webcam around this time of year will often show that.

From wiki " Owing to the prevailing easterly winds off the Arctic Ocean, Utqiaġvik is completely overcast slightly more than 50% of the year. It is at least 70% overcast some 62% of the time. Cloud types are mainly low stratus and fog; cumuli forms are rare. Peak cloudiness occurs in August and September when the ocean is ice-free. Dense fog occurs an average of 65 days per year, mostly in the summer months".

With fog you can often have a situation where the temperature is lower at surface than up several hundred metres in the atmosphere.

It is a significant challenge for the models to handle fogs and inversions. I imagine, given we have so many years of previous data around the whole north slope of Alasaka, that the models would be better equipped to create 2m temps for Utqiagvik rather than locations far removed in the centre of the Arctic. Perhaps a GFS vs Utqiagvik temperature point comparison would be  a worthwhile exercise.   

I find this interesting paper on inversions by Mark Serreze et Al (currently director of NSIDC) The paper was published in 1991 and so some of the timing events may be a bit different now but I'm sure a lot of the data on inversions is still relevant to today's Arctic. This snippet shows how the frequency of inversions changes by month, but note at the drifting stations the frequency remains high over the summer months compared to stations more inland (Zhigansk)

https://journals.ametsoc.org/doi/pdf/10.1175/1520-0442%281992%29005%3C0615%3ALLTIOT%3E2.0.CO%3B2

[uniquorn]

https://www.nature.com/immersive/d41586-020-01446-x/index.html

During the winter, the atmosphere in the Arctic doesn’t get colder with height, but warmer. To study these temperature inversions, MOSAiC scientists released four weather balloons per day, allowing them to measure the temperature (among other parameters) at one-second intervals as the balloons climbed high into the atmosphere. They also built a meteorological tower that constantly monitored the atmosphere’s temperature at 2, 6, 10 and 23 metres above the surface.
The extreme layering surprised researchers. On several different occasions they saw inversions as big as 5 °C in the lowest 10 or 20 metres of the atmosphere. Yet scientists don’t know how often these inversions occur, or their typical strength.