March 2026 market

Weeee

14/2 - 1.15 with full forecast

15/2 - 1.17

@Rhhgdd 16/2 - 1.17

@Rhhgdd 17/2 - 1.193

@Rhhgdd these are raw figures without adjustment. GISS was normally +0.015 in the last 3 months of 2025 compared to my model.

In January we've had a lot more coverage and the normally anomalous South Africa stations haven't come in and the giss-model diff was -0.013.

Even if the hot SF stations come in for February they won't have that big of an impact as after all we're expecting a lot more stations upon release. So for the time being, I'm expecting 1.180-1.198 as the range, we'll see how the forecast changes over the next few days but all ensembles have started aligning now so I'm expecting little divergence.

edit://decimal point mistake

@Rhhgdd I'm fairly sure you meant to write +0.015 and -0.013 (I've made that mistake myself a couple times). Here is my own GISTEMP->ERA5 errors (i.e. what I call the unadjusted number around the 2nd/3rd of the month): note though that for March 2025 onwards I use my own gistemp run's unrounded data, using the ghcnm run's data used for the official release to calculate the error, with the notable exception of August and September 2025 when I think the coverage was fairly poor.

For reference though, I stuck with a final bias adjustment of +0.014 from August 2025 onwards given the above concerns. For this month I am going to stick with a subjective adjustment as I mentioned though (for now, -0.005 C).

~

You said, "after all we're expecting a lot more stations upon release" but in terms of absolute number of stations I really do not think this will be the case as based on the shift in ghcnm that happened (some time around) after February 2nd (and continues until now) -- if the current practice continues we will have GREATER coverage but LESS stations.

With that in mind I don't think the August-Dec 2025 runs on will be the best comparison in terms of coverage when early runs come out; you can look at the numbers now though to see that Oct-Dec 2025 actually has less stations in the most recent ghcnm data than from runs weeks after the GISTEMP release date (about 500 less stations) (this is why I'm still using my gistemp run with the ghcnm from February 2nd for the overall historical statistics in my ERA5->GISTEMP model).

For example, taking a common release for the Oct-Dec LOTI (i.e. ghcnm nightly releases dated from 20251107, 20251207, 20260107) we had ~6700-6800 usable stations (by some ad hoc way I check it: ghcnm stations >=10 years in the 1991-2020 baseline period) available for that 8th day of the month. By comparison, for January (the ghcnm release dated 20260207) had 5450 stations (and even last night's ghcnm only had 5969).

@parhizj My model doesn't utilize ERA average temps, I find it way too inaccurate. But if i know which stations will be available upon release I am pretty accurate, the more the stations the smaller the error too.

The only thing that's just been catching me off guard these last few months is how unreliable GHCNm is when it comes to data availability. In this day and age it really shouldn't be that difficult to just maintain stability when other reanalysis tools are able to do so.

Generally we have an increasing trend in the number of stations. And some of the data came in a lot earlier in February for Jan. however, I haven't actually looked into "usable stations"

@Rhhgdd That's fine if you don't use ERA5, but if you use something better though I'm pretty sure your error wasn't THAT large for the last 3 months ... (an error of 0.15 is 3 bins -- of course my error was also very abnormally large for August/October but those were exceptional -- and the monthly model for August has poor correlation); I checked your posts for January for instance and the last one I think you posted 1.1 or 1.11 so thats only a -0.02 or -0.03C error

@parhizj oh yeah I editted that, I meant 0.015 ofc 😅 I got 1.087 raw then adjusted up to 1.1-1.11 as that's what the adjustment was in winter 2025.

And then January came along, the stations available on the 11th were completely different and the highly anomalous South African stations haven't come in at all.

Ultimately, I don't even know what to expect for this month until GHCNm data comes out more consistently. It's guesswork this month but I'm expecting either -0.015 lower or +0.05 higher.

Unless of course South African 100°C stations come in then we're looking at the higher end. On the other hand with more coverage as you've previously mentioned, they won't make as big of a difference.

@chaitea Model unadjusted: 18/2 - 1.176.

And because we don't know whether to adjust up or down yet current range is 1.16-1.19 which is not ideal and not very accurate... If it is anything like January coverage then we are adjusting down so my money is on the lower range, subject to forecast change but it's stabilizing rn

@chaitea Model unadjusted: 19/2 - 1.176

Interesting to see the same exact results, toss of a dice anyway

@chaitea 20/2 - 1.172

@chaitea 21/1 - 1.179,

I think safe to say this should be adjusted downwards after doing some station reconnaissance

@chaitea 22/1 - 1.171

@chaitea 23/2 - 1.177

@chaitea 24/2 - 1.185

@chaitea A small rise in your data? Mine has trended marginally downwards for the last several days....

@parhizj looks like a subtle change in NH, the temperature drop isn't as severe so some subboxes are expected to be warmer. this could just be due to a forecast overcompensating so i normally look at the average of the last few days from my model rather than a single daily value

my final adjustment will be -0.008. still expecting 1.16-1.17

@chaitea i am expecting it to be trending upwards from this point on however

@parhizj 25/2 - 1.190 (adj 1.18)

@chaitea 26/2 - 1.208 (adj 1.20)

@chaitea 👀 Well if warmer ssts have buffered all that cool air in the North I could I understand since I only track 2m air temp if thats the reason its been diverging a good deal from my unadjusted temp since the dip.

@parhizj 27/2 - 1.189 (adj 1.18)

the increase was a small bleep, but like i said before looking at average, expecting 1.17-1.18

@chaitea 27/2 - 1.184 (adj 1.17-1.18, final figure)