By Andy May
Comments on my recent posts concerning HadCRUT5 surface and UAH Lower Troposphere temperatures often degrade to a comparison of the accuracy of RSS and UAH satellite temperatures. Some seem to believe that RSS is more accurate than UAH, when radiosonde data shows the reverse is true. So, this is a short post to briefly cover the issue.
My main sources are a blog post by Roy Spencer and a peer-reviewed paper by John Christy, Roy Spencer, William Braswell, and Robert Junod in 2018 here. The 2018 paper, herein called CSBJ18, presents a comprehensive analysis of four satellite mid-troposphere temperature records. Three are global datasets: UAH (version 6), RSS (version 4), and NOAA (version 4). The fourth is the newer UW (University of Washington, version 1) series, which only covers the critical tropical region from 30°S to 30°N.
Comparisons of the satellite datasets are difficult, because while they share the same data, they use very different procedures to produce their respective temperature records. In particular, the UAH procedure is quite different from the other three. While the traditional NOAA dataset is called “STAR,” they now have a new one, called RMTMT for the middle troposphere that we will discuss near the end of the post.
CSBJ18 compares all the satellite datasets to 564 stations of the Integrated Global Radiosonde Archive (IRGA). The stations utilized have data from 1979 to 2016. The satellite monthly data was compared to radiosonde (weather balloon) monthly averages using a global grid. CSBJ18 explains the details of the comparisons they made. Their procedure and methods were very thorough. Nearly all radiosonde records are over land, so comparisons to weather reanalysis datasets were also made since warming rates are different over land and ocean.
In every comparison, both globally and for the tropics, the UAH satellite temperature record correlated to the radiosondes best. In addition, the UAH global temperature trend is lower than the trends of the other datasets from 1979 to 2015 globally and for the tropics. The radiosonde data is not perfect, it has erroneous data as well, but it is independent of the satellite records and provides a neutral, unbiased check on the various methods of processing the satellite data. There is no such check for the various surface temperature datasets, they all share the same data and mostly use the same methods to process it.
One of the reasons that UAH has a lower warming trend than the other datasets is UAH has corrected clearly spurious data in the older NOAA-11 to NOAA-14 satellite instruments and the other datasets have not. These satellites used an earlier MSU instrument and had orbital problems that needed to be corrected for. NOAA-14 overlapped with NOAA-15 for three years and NOAA-15 had a much more advanced MSU (the AMSU), plus it had a better orbit during the overlap period. Comparing the data from NOAA-15 to NOAA-14 demonstrated the problem with the NOAA-14 instrument, the details of how UAH corrects for this problem and several others are explained in a 2017 paper by Spencer, Christy, and Braswell. After all corrections were applied to the NOAA-14 and NOAA-15 data, NOAA-14 still showed +0.2°C/decade more warming than NOAA-15. After only the basic diurnal correction was applied, the two satellites differed by as much as 0.34°C/decade.
The full story of all the required corrections and adjustments to produce a satellite tropospheric temperature record is too involved to be explained here but is well documented in the references cited. Here we will only show the results of Christy, et al.’s comparison of the satellite datasets to the radiosonde data. Figure 1 compares the global radiosonde station data to the satellite date in the same locations. In this comparison the UAH data correlates best with the unadjusted radiosonde data. The Y axis is the correlation between the satellite data and the radiosonde data, and higher is better. The solid bars represent the unadjusted data, the stippled bars use the same satellite data, but the adjusted radiosonde data includes only those radiosondes that match the respective satellite dataset with a correlation coefficient of 0.7 or better. No changes were made to the satellite datasets.
As Christy, et al. explain, while the UAH dataset correlates better than the other two satellite datasets, the difference in correlation approaches statistical significance, but does not reach that level. Figure 2 compares the temperature trends from the IGRA radiosondes to the global satellite datasets. The trend of the unadjusted radiosonde data, from 1979 to 2015, is shown in gray. The satellite data, for the radiosonde locations, is shown in green. The adjusted radiosonde data is shown in red, and the full global grid trend is shown in pink.
The UAH dataset matches the radiosonde warming trend better than the RSS and NOAA datasets. The better match applies both to the adjusted data and the unadjusted data. At the radiosonde locations, the UAH trend is within 0.01°C/decade of the unadjusted radiosonde trend. The RSS and NOAA datasets show much more warming.
The view from NOAA
It is interesting and revealing that CSBJ18 has not been rebutted in any journal articles or blog posts that I could find. In fact, they seem to avoid discussing the earlier satellites, that is NOAA-14 and prior satellites that have the problems discussed in CSBJ18 and above. Cheng-Zhi Zou, who helps to maintain the NOAA satellite dataset, wrote a paper that compares RSS, NOAA and UAH temperature records, but he begins the comparison in 2002, well after NOAA-15 was launched with the more advanced AMSU microwave unit in it. This avoids comparing the satellite records from NOAA-14 and previous satellites that contain the more primitive MSU data. His Figure 4 (our Figure 3) compares the records from 2002 to 2020.
The left plots in Figure 3 compare the generated temperature anomalies and shows that from 2002 to 2020 the various temperature series overlay quite well, with occasional NOAA STAR outliers in red. The right-hand plots show the difference between the three listed satellite records and the new NOAA RFTMT dataset introduced by Zou, et al. NOAA’s new RFTMT is warmer than either RSS or UAH over both land and ocean, and by almost the same amount. It is remarkable how close RSS and UAH are. Zou, et al. want RFTMT to be a new standard satellite dataset and claims it is more accurate than the others, but I’m not on board with that.
Zou, et al. point out that NOAA-15 had problems late in its life and STAR uses it until 2015. This probably causes some of the NOAA STAR problems. UAH stops using NOAA-15 in 2007 and RSS stops using it in 2010. Zou, et al. recommend that their new RFTMT satellite record be used as a standard, but it seems to be just another NOAA attempt to increase the warming trend with unwarranted adjustments and extensive cherry-picking of satellite data. Zou, et al. estimate that a satellite with the most advanced AMSU instrument, in a perfectly stable orbit and under ideal conditions, has a trend uncertainty less than 0.04°C/decade over the period from 2002-2020, which is in line with previous estimates. Orbital characteristics add to the uncertainty and the total global warming trends he reports are UAH: 0.17°C/decade, STAR: 0.17°C/decade, RSS: 0.18°C/decade, and RFTMT: 0.20°C/decade. The total uncertainties associated with all the warming trends are roughly ±0.1°C/decade. The uncertainty by satellite is listed in Figure 3, you may need to click on Figure 3 to get it to full scale to read the values.
Computing climate sensitivity with UAH
In addition to verifying that the radiosonde temperature data compares better to UAH, than either RSS or NOAA’s datasets, Christy and McNider use the data to compute lower tropospheric climate sensitivity to greenhouse forcing. This is best characterized as a TCR (transient climate response) estimate and Christy and McNider call it TTCR for tropospheric transient climate response.
They accepted the IPCC assumptions that the only significant influences on climate since 1979 are volcanic eruptions, human greenhouse gas emissions, other human activities, and ENSO activity. They removed the volcanic and ENSO effects from their lower troposphere UAH record and the underlying trend, absent these effects, was about +0.1°C/decade.
They used the IPCC AR5 values for GHG forcing since 1979 (1.45 W/m2) and total forcing (1.24 W/m2). Given the corrected warming of 0.1°C/decade they have a total warming for their period of 0.368°C. The period is 38 years and the warming they computed is 0.096°C, which we rounded off to 0.1°C. Applying the values to compute TTCR, Christy, et al. compute 1.1°C/2xCO2 ±0.26, where “2xCO2” means doubling CO2. This should be a more accurate estimate of TCR than can be obtained from any surface record, as more of the atmosphere is used in the calculation.
Satellite temperature measurements are more useful than surface measurements in climate studies for several reasons:
- The data used is all collected the same way and with similar instruments.
- More atmospheric mass is included.
- The temperature measured is mostly above the chaotic boundary layer of the atmosphere and is more stable.
- Radiosonde data is available as an independent check on the calculations.
Regarding #4 above, the UAH temperature calculations correlate best with the radiosonde data, suggesting that it is the best satellite temperature record. The RSS decision to include the clearly flawed NOAA-14 data in their calculations is highly questionable when the clearly superior NOAA-15 data is available for the period in question. NOAA-15 develops problems later and both RSS and UAH drop it early, but NOAA-STAR continues to use it for many more years.
Since the UAH temperature record is probably the superior record, it is reasonable to estimate TCR from it. Christy and McNider did this and derived a TCR of 1.1°C/2xCO2. While this value is much less than the AR6, AR5 and AR4 value of 1.8°C/2xCO2 (AR6: 1.2 to 2.4), it falls in line with estimates from Lewis and Curry, Lindzen and Choi, Alexander Otto and colleagues, and others.
A reduction of 39% in TCR is significant, particularly since Christy, et al.’s estimate falls below both the AR6 likely range (1.4 to 2.2) and their very likely range (1.2 to 2.4). It seems very likely to me that AR6 is deliberately ignoring valid analysis and data to sell their political agenda.
The references can be downloaded here.