RIP, Temperature.Global

A few years ago, I began seeing contrarians promoting data from a website called temperature.global (TG), a website that claimed to publish global temperatures. According to their description, TG "calculates the current global temperature of the Earth. It uses unadjusted surface temperatures. The current temperature is the 12M average mean surface temperature over the last 12 months compared against the 30 year mean. New observations are entered each minute and the site is updated accordingly. This site was created by professional meteorologists and climatologists with over 25 years experience in surface weather observations." The website was run anonymously; to my knowledge, nobody knows exactly who is behind it. I corresponded with at least one of the people who ran the website, and he/she used the initials TG for the person's name. In another post I documented some of the failures of the website after seeking clarification from TG. The more I corresponded with TG, the more I became convinced that this website is downright silly.

A couple months ago, the TG website announced that it would be shutting down at the end of 2024. Their data began in 2015, so this means that as of today they published 10 years of data before calling it quits. I thought it would be helpful to update my previous post with what I've learned about the site since my original post. Spoiler alert: TG did not improve anything since my February 2022 post or July 2023 post.

TG Global Mean Surface Temperature (GMST) Trends

TG published "real time data" in the sense that if you go to the website you can see the temperature reading from the last station it processed, but it did not publish any calculated means in real time. In fact, they never published any daily or monthly global mean temperatures. They only published 12 -month running means (you'll see why below). So the GMST anomaly for any month is actually the mean for the 12-month period ending in that month; i.e., the value for March 2023 is the mean from April 2022 to March 2023. I plotted all their data from 2015 to 2024, and the slope for their data was 0.763 ± 0.373°C/decade (2σ).

This slope makes the TG website easily and by far the most "alarmist" of any of the major GMST datasets. Below I plotted a comparison between TG and UAH-TLT, the darling of the contrarian movement. The slope for UAH is 0.34°C/decade, which is less than half of TG's 0.76°C/decade. 

For clarity, I should note here that, despite their claims on their website, the TG data doesn't have a properly calculated 30-year baseline (you can't make a 30-year baseline from 10 years of data). Instead, they read that NASA had estimated that their 1951-1980 baseline averaged 14°C, so they called that their baseline. All their anomaly values are actually just deviations from 14°C, which they call "normal." Since UAH only goes back to 1979, I can't set it to a 1951-1980 mean. Since I'm really only interested in the slopes, I plotted UAH with their baseline and TG with their fake "baseline" of 14°C so you can see how their slopes compare.

TG's First Two Years

Astute observers will note the wild swings in temperatures for the first two years of the TG time series. These temperature swings are clearly absurd, as they suggest that their 12-month running means vary by as much as 3.25°C, which is clearly preposterous. In an email back in 2022, I asked TG to provide the raw monthly data for their dataset (rather than 12-month running means), and I never received that, but weeks after writing my original post, I did receive those values for 2016. Here are the values I received.

A couple things should be obvious from these numbers. First, I calculated the annual mean for 2016 using these email numbers and compared it to their published value for their annual mean on their website, and the two do not agree. Their published anomaly for 2016 is 0.24°C colder than what I calculated from the values I was given in their email. Second, their monthly values show unreasonably huge swings in temperature. According to them, GMST ranges from ~4.3°C to ~22.9°C over the course of a year, a difference of over 18°C. For comparison, ERA5 shows that GMST varied by less than 4.5°C over the course of a year.

Almost certainly the wild swings in seasonal temperatures in their dataset have to do with selection and location biases. By their own admission, TG only calculates simple means of the data they collect; their averaging method is basically the same as the method adopted by Tony Heller and his followers; they do not use a grid or calculate anything resembling an area-weighted mean. Almost certainly their sample contains more NH thermometers than SH thermometers, so a simple mean of thermometer data will exaggerate the variability of temperatures over the course of a year. There are fewer SH thermometers to balance out the NH thermometers, so variability will be closer to that of the NH than the globe. And since they include sea surface temperature readings from ships (that are mobile), bias can be added by the changing shipping patterns throughout the year. These biases almost certainly explain why TG never published their monthly anomalies; this would make the flaw in their methodology obvious. By publishing only 12-month running means, they can reduce the appearance of the bias in their methodology, but it's still there. Their trends are going to be weighted towards NH trends, and specifically towards those areas in the NH that have the highest densities of thermometers.

Annual Mean vs December Mean

Given that TG publishes twelve-month running means, TG's annual published mean should equal their December mean temperature every single year, and this is usually the case but not always. There are notable exceptions in 2015, 2016 and 2023.

I asked TG about the discrepancy in 2015 and 2016 back in 2022, and TG had no idea why the discrepancy was there. Now given that they did not have 12 months of data until Dec 2015, clearly the 12-month mean temperatures for Jan 2015 to Nov 2015 can't be actual 12-month running means; they only began collecting data in Jan 2015. So TG must have fudged something (nobody knows what) to create the values for Jan to Nov 2015. But Dec 2015 should equal the annual mean for 2015. Same should go for 2016. And a similar discrepancy occurs in 2023, where the annual mean is 0.24°C but the December mean is 0.44°C. While I can't say for sure, I find it interesting that the published annual mean for 2023 in Fahrenheit is 0.44°F. I suspect what happened is TG entered their December value of 0.44°C into the Fahrenheit column of their data table for annual temperatures, then converted 0.44°F to 0.24°C. I could be wrong, but it makes the most sense of the discrepancy. 

Update: 1/13/2025. The December 2024 updates have just come in for the major GMST datasets and reanalyses. The mean trend among these for 2015-2024 was 0.27°C/decade, so I decided to plot them all on the same graph. I plotted the results of GMST datasets and reanalyses in dotted lines set to the 1951-1980 mean (left side of graph). Since TG doesn't have a real baseline, I plotted these with solid lines with the same scale (right side of graph) but shifted 1.5°C so you can see both together more easily; the TG scale is simply deviation from 14°C. I'm mostly interested in showing the slopes here, so both the left and right sides of the graph show a range of 2.5°C. Note that the TG annual and TG 12-month mean December plots should be identical, but they they differ significantly in 2015, 2016, and 2023.

Also note the differences in the slopes for these. The trendline for TG's annual mean is 0.86°C/decade, but the trendline for TG's 12-month mean December is 2.0°C/decade. Again, these should be identical. And both are larger than the slope of the 12-month running mean temperatures, which was 0.763 ± 0.373°C/decade (2σ). The mean slope of the GMST and Reanalyses was 0.27°C/decade, so that means with respect to GMST and reanalyses:

• TG's annual anomalies show 0.86/.27 = 3.2x more rapid warming.
• TG's 12-month mean in December anomalies show 2/.27 = 7.4x more rapid warming. 
• TG's 12-month running mean anomalies show 0.76/.23 = 2.8 more rapid warming.

Let's be kind, ignore the December issue, and just say that TG shows ~3x more rapid warming than GMST datasets and reanalyses.

Reconstructing Monthly Anomalies

Since TG wouldn't give me their monthly temperatures for any year except 2016, I decided to reconstruct them from the data they published and sent me by email. In my previous post I tried one method based on assuming that Jan 2015 was a one-month mean, Feb-2015 was a two-month mean, etc. up through Dec. 2015. After receiving the monthly values for 2016, I tried solving for monthly values using the 12-month running means published on their website. That is, since I know the monthly values for each month from Feb 2016 to Dec 2016 and the 12-month mean for Jan 2017, I should be able to solve for the Jan 2017 value that when averaged with Feb-Dec 2016 produces the 12-month mean for Jan 2017. Below I plotted the results of both of these methods. The red curve is my first attempt described in my previous post. The black line is my second attempt from the email data I received.

The values for these are obviously absurd, with temperatures varying by over 45°C over the course of a year, but the fact remains that if the 2016 data TG gave me is correct and their 12-month running means published on their website are correct, my calculated reconstruction represented in the black line should produce their monthly mean data. So there are two possibilities here. Either the black line correctly reproduces their monthly temperature values, and their monthly values are absurd, or the black line does not reproduce their monthly values, and there are irreconcilable discrepancies in their dataset. Given that the average of the 2016 values sent to me by email doesn't match their published annual anomaly for 2016, I strongly suspect the second possibility is correct. I suspect that their methodology is extremely flawed and/or their process is riddled with mistakes and incompetence. We can have no confidence that TG was ever able to do what they said they would do on their website.

Conclusion

TG have not given a reason why they decided to shut down after only 10 years. It's possible that TG only intended to this for a decade or that after a decade he just got tired/bored of running the website. But there are two other possibilities that are intriguing to me:

• Perhaps TG realized that their website was producing trends that at least double the temperature trends of any of the major GMST datasets.
• Perhaps TG realized that if they were to fix the flawed methodology and/or incompetence that produced the many discrepancies I described above, they would have to make adjustments to their methods, and those adjustments would also alter their published data, and then they would open themselves up to criticism that they were adjusting the data.

The last option seems the most plausible to me. TG knows about the problems with 2015-2016 because I mentioned it to them, and they acknowledged the problem. They also have this weird discrepancy in 2023 that obviously needs to be fixed. It's also obvious from even the monthly data I received in TG's email that sampling biases are causing huge problems for them given their insistence on using Heller's averaging method with no grid or area-weighting.  They can't fix these problems without altering their published data, so rather than fixing their problems, perhaps they decided to shut down. Now granted I'm just speculating about this, but whatever their motivation was, I think they made the right decision.