🏅 Top traders

1.23

https://climate.copernicus.eu/surface-air-temperature-april-2025
April 2025 0.6 C
April 2024 0.67 C

If Gistemp also just 0.07 below last year that would be 1.25 but would suspect 1.20-1.24 per JRP is much more likely. But how much more likely 95:3 ?

@ChristopherRandles One more run tonight, but I'm guessing it will just be 1.23 (it was between 1.21-1.23 for all but the first possible run)

new ersstv5 came out a few hours ago

ghcnm.v4.0.1.20250502:

118.629

April Release: Friday, 9 May 2025, 11:00 AM EDT

@parhizj woh so close to your prediction, but I still think its gonna increase, the reason for this is that ghcnm is not yet fully filled and adjusted

@gonnarekt yeah, it usually does

ghcnm.v4.0.1.20250503:

120.779

@parhizj wow so rapid growth, is it possible to to go upper than 1.23?

@gonnarekt possibly, but it becomes less likely of course.

Anecdotally, if feels more likely it will bounce around the last temp after rising so much.

ghcnm.v4.0.1.20250504:

121.219

@parhizj what’s the fresh one?

ghcnm.v4.0.1.20250505:

122.912

or,

a reminder to always hedge:

ghcnm.v4.0.1.20250506:

122.536

ghcnm.v4.0.1.20250507:

122.626

ghcnm.v4.0.1.20250508:

122.675

ERA5 finalized for April 30.

Final point prediction is 1.183 C

(adjusted upwards from 1.173 C from own past forecast errors)

Splitting the three models still is doing the best (gefs split offset):

@parhizj looks like it will be a bit more that 1.18, i guess 1.19+ about 90%. My model shows 1.21-1.22

Forecasts for ECM vs GEFS/GEPS only slightly diverge in different directions for the last couple days of the month, so temps should be continue to be relatively flat for the last couple days of ERA5, so there shouldn't be a large impact on the forecast.

It still looks like this month's anomaly will be ~ 1.181C (has been relatively steady for several days now).

(This is adjusted upwards from 1.171 using my seeming past own prediction error bias for a very few samples)

Went through not so small amount of work after finding a major error today in the original area calculation code I wrote almost 2 years ago, and had to revise a lot of code.

Sadly I've been handicapping myself this entire time by accidentally effectively using only the southern hemisphere (some simple tests like making sure the total area was being calculated correctly passed as two bugs contributed to doubling the contribution of a single hemisphere temps) (ARIMAX AR term corrected for much of the recent error since I started using it, but now that is no longer needed, so its a simpler 1,0,0 model now).

The temperature from the tau=0 (00Z,06Z,12Z,18Z) and tau=0,6 (00Z,12Z) runs are now a tight fit to a simple linear regression to fit the ERA5 data (RMSE ~= 0.05 and a slope ~= 1 for all models):i.e.:

After some re-computing all the past temperature data I use and updating hte ARIMAX models to 1,0,0 all the models have come in much closer alignment:

GEPS & GEFS mostly agree, with last few ECM deterministic models running cool.

The long range GEFS (hindcast ARIMAX adjusted) for the last few days of the month is also now producing more reasonable temps and is consistent with the end of the latest medium range forecast.

Edit: after checking Polymarket I find they agree now with the final point estimate I come up with (~1.225 C):

I look forward to see if my final point estimates are now more or less stable now that major bugfix is done..

@parhizj how's it looking boss

@LeonardoParaiso 1.189 C is the (adjusted) point estimate I come up with as of tonight.

@parhizj hello, what do you think now? And why do you have so low expectations? In my point it’s about 1.9-1.23 now

Polymarket up to 95% on less than 1.22 but no lower bands this month so not as helpful on how much lower than 1.22C.

https://polymarket.com/event/april-2025-temperature-increase-c4?tid=1744291485550

Still % has gone up so the best est has probably moved down in last couple of days

@gonnarekt Regardless, 95% seems overconfident based on my own forecast (unless they imply very low temps, but that is impossible to discern).

From yesterday:

(All data fit to ERA5 and ARIMA 1,0,0 adjusted) Red is weighted equally between the medium range GEFS (red) (adaptive average of all runs), GEPS (green) (mean of ensemble mean for last few runs), ECM (blue) (average of recent few ECM runs)

Taking the month's mean (which a predictor of the ERA5 monthly mean) and using a linear regression from ERA5->GISTEMP leads to a point anomaly forecast of 1.179 . Then I adjusted it based on the mean errors for my last few forecast month's forecasts to 1.189.

For reference ECM has been the most accurate of this mix of data for the recent ~ week (reset on April 12):

But it is inline with roughly an equal weighting between the three data sources (called GEFS split offset).

Edit: the fact that the error has become negative with greater lead time (as the trend is downwards also) implies the debiased models other than ECM have been predicting hotter than observed temperature for the remainder of the month (i.e. yes, models have cooled off; of course the GEFS I use isn't the pure latest run either though)

Weighted by estimate of std. dev. for offset:
[ 0.          0.          0.          0.          
8.23529412 
17.64705882
40.        
34.11764706]

Have added GEPS (rolling 5 run average == 2.5 days) to medium range forecast: so I now equally weight GEFS, GEPS, ECM (ARIMAX adjused) over the medium range, and GEFS long-range and Prophet for beyond (referencing the adjusted Prophet from the post-medium range multi-model adjustment).

I found a significant bug in the ECM arimax offsets code I used earlier (resolving that abnormal discontinuity at the beginning of the forecast period). This had the effect of dropping the ECM forecast temps (they are in stark contrast to GEFS). Notably GEPS has little variation across the same number of runs (given that it has 21 members, this smoothing may be why -- as the ECM runs are single-shot deterministic).

After a small dip for the next day or so, GEFS still shows a sharp rise in a few days (a ~ 0.2 C temp increase from April 14th to April 15th).

The long range GEFS (after splitting with the adjusted Prophet from the medium-range multi-model split) is still pushing up the temps though:

Average GEFS (Split offset adjusted) - ECM_ARIMA (mean of recent runs fit with ARIMA to ERA5): 0.23
Average GEFS (Split offset adjusted) - GEPS_ARIMA (mean of recent runs fit with ARIMA to ERA5): 0.08
Multi-model Offset (Split offset adjusted) - GEFS|GEPS|ECM (mean of recent runs fit with ARIMA to ERA5): 0.09
Multi-model medium range mean offset (GEFS, GEPS, ECM): 0.0923
Long range mean offset (GEFS|Prophet): -0.1702

Multi-model mean offset (Medium and Long Range): 0.0467