Social Channels


Additional Options

Date Range

Receive a Daily or Weekly summary of the most important articles direct to your inbox, just enter your email below:

NASA Satellite orbiting Earth Credit: Nasa / Unsplash
21 June 2017 16:44

Study: Why troposphere warming differs between models and satellite data

Zeke Hausfather

Zeke Hausfather

Zeke Hausfather

Zeke Hausfather

21.06.2017 | 4:44pm
Global temperatureStudy: Why troposphere warming differs between models and satellite data

The most common measure of global temperature rise is here on the Earth’s surface, but scientists also gather data on how temperatures in the atmosphere high above us are changing.

Of particular interest is the troposphere – the lowest layer of the atmosphere where almost all of our weather occurs. To track temperatures, scientists use satellites, which have been providing data since they were first launched in the late 1970s.

Since around the start of the 21st century, the tropospheric warming recorded by satellites has been slower than the rate projected by climate models. In a new study, published in Nature Geoscience, researchers find that these differences are outside the range of what we would expect from natural variability.

Instead, they say the differences could be down to recent changes in greenhouse gas emissions, solar output, volcanic eruptions and air pollution that weren’t anticipated in the assumptions made by climate modellers.

Overall, the study suggests that while tropospheric warming has not accelerated to the extent that models have predicted in recent years, there’s little evidence that it has slowed down.

Temperatures from satellites

Much of our historical temperature data comes from weather stations, ships, and buoys on the Earth’s surface. Since 1979 temperature records of the atmosphere are also available from satellite-based microwave sounding units (MSU). These measure the “brightness” of microwave radiation bands in the atmosphere, from which scientists can estimate air temperatures.

However, the bands measured by the satellite instruments cannot easily provide the temperature of a specific layer of the atmosphere. Researchers have identified particular sets of bands that correspond to the temperature of the lower troposphere (TLT) spanning roughly 0 to 10 km, the middle troposphere (TMT) spanning around 0 to 20 km and the lower stratosphere (TLS) spanning 10 to 30 km.

Unfortunately, these bands tend to overlap a bit. For example, TMT estimates will include part of the lower stratosphere, while TLT estimates will include some surface temperature. These overlaps matter because different parts of the atmosphere are expected to react very differently to climate change.

When greenhouse gases trap incoming solar radiation, they tend to increase the temperature of the surface and lower atmosphere, and decrease the temperature of the upper atmosphere as less solar radiation is escaping. We see this in satellite observations and data from weather balloons, where the lower stratosphere is cooling while the underlying troposphere and surface are warming.

Because the tropospheric temperature estimates from satellites overlap with part of the stratosphere, they end up combining a bit of stratospheric cooling with tropospheric warming and can underestimate the true rate of warming. To avoid this issue, the new study applies a correction to remove some of the stratospheric cooling from the TMT series. The approach they use for this is described in a previous paper published in the Journal of Climate.

Correcting errors in the data

Dealing with stratospheric contamination is not the only challenge when working with satellite data. Unlike on the surface where there are tens of thousands of individual observation stations, there are only around two to three MSU satellites taking measurements at any given time, and the satellites only last about five-to-ten years before they need to be replaced.

While the satellites are designed to pass over the same part of the earth at the same time every day, that changes as their orbits decay. A satellite that once took the temperature over London at 2pm ten years ago, for example, might now be taking it at 8pm. Changing the observation times has a big effect on the temperatures measured, and researchers need to correct their measurements for this.

Similarly, a replacement satellite might measure temperatures slightly differently from its predecessor. Around the year 2000, for example, the instrument in the satellites was changed to an upgraded version of the sensor. All of these can potentially introduce bias into measurements that need to be addressed.

There are two main groups that process the same underlying MSU data to estimate atmospheric temperatures: the University of Alabama, Huntsville (UAH) and Remote Sensing Systems (RSS). Each group has a different set of assumptions to correct for various issues in the data, and they end up with fairly different results. You can see how the UAH (yellow line) and RSS (blue) figures differ in the chart below – particularly after the year 2000.

Annual global mean middle tropospheric temperatures from RSS and UAH from 1979 through 2016, covering from 82.5 N–82.5 S. No stratospheric adjustments are included as an adjusted UAH dataset is not available. Chart by Carbon Brief using Highcharts.

While RSS generally agrees with the rate of warming seen globally in surface temperature records, UAH shows much less warming – including a more pronounced slowdown in temperature rise after 1998. The differences between satellite records are much larger than those between different surface temperature estimates. Co-author Dr Carl Mears, the co-founder of RSS, suggests that:

“In general, I think that the surface datasets are likely to be more accurate than the satellite datasets. The between-research spreads are much larger than for the surface data, suggesting larger structural uncertainty.”

These large uncertainties between satellite datasets somewhat complicate any comparison of tropospheric temperatures with climate models, as it makes it unclear if the disagreement is due to issues in the models or in the observations, and leaves open the possibility that additional corrections to the data may happen in the future.

Comparisons with climate models

In their paper, the researchers employed a number of different statistical tests to compare climate models and observations of TMT. They corrected both models and observations for stratospheric cooling influence, and compared the two over the period from 1979 through 2016 as shown in the figure below. The upper chart shows the model output and the lower chart shows the observations from RSS.

Top panel shows stratosphere-corrected RSS TMT compared to the CMIP5 multimodel average TMT. Bottom panel shows the difference between the two over time. Source: Santer et al. (2017)

While the rate of warming in the models and observations is pretty close prior to the year 2000, the differences after 2000 are much larger. Some of these differences are explained by short-term natural variability, such as El Niño events, which do not necessarily occur at the same time in the models as in the observations and tend to average out. However, even with this removed from the observations, the researchers find that notable differences remain.

To explain these differences, the researchers tested a number of different possible factors. First, they looked to see if the difference could be explained by longer-term multi-decadal natural variability from El Nino and ocean temperature oscillations that was not captured in the model average.

They found that while natural internal variability can explain most of the relatively small differences between modeled and observed tropospheric warming in the last two decades of the 20th century, but can’t fully explain why model tropospheric warming is larger than in the satellite data during much of the early 21st century.

Second, they looked to see if the difference might be caused by models being too sensitive to CO2. They found no discernable relationship between model sensitivity and their ability to accurately predict tropospheric temperatures over this period.

The conclusion the researchers came to was that the model-observation discrepancy isn’t down to a single factor, but a combination. Specifically, they posit that it is due to a combination of internal variability and that models got some climate forcings wrong in recent years.

Climate models used historic data for factors like greenhouse gas concentrations, solar output, volcanic eruptions, air pollution, and other factors that can affect the climate through 2005 or so, but after that point made assumptions of how these would change in the future. Recent research has suggested that a series of moderate volcanic eruptions, a long and unusually low minimum in the sun’s energy output during the last solar cycle, and an uptick in particulate pollution from Chinese coal-fired power plants have all changed these forcings in ways unanticipated by the modelers.

These forcings will be updated in current modeling effort, called CMIP6, being done in preparation for the next Intergovernmental Panel on Climate Change report. This new generation of models, featuring forcings closer to observations in recent years, will likely show better correspondence with tropospheric temperature observations, but may not be any more or less sensitive to CO2 than the prior generation of models (CMIP5).

According to Dr Gavin Schmidt, director of NASA’s Goddard Institute for Space Studies, who was not involved in the paper, there are even plans to rerun the older CMIP5 generation of climate models with updated forcings to see what happens if those are updated in isolation without changing other factors.

Ultimately, the paper finds that while there is a mismatch between climate models and observations in the troposphere since the year 2000, there is little evidence to-date that the model/observation differences imply that the climate is less sensitive to greenhouse gases. The results suggest that while these short-term differences between models and observations are a subject of great scientific interest, it does not diminish the reality of long-term human-driven warming.

Santer, B. D. et al. (2017) Causes of differences in model and satellite tropospheric warming rates, Nature Geoscience, doi:10.1038/ngeo2973

Sharelines from this story
  • Study: Why troposphere warming differs between models and satellite data
  • Just wanted to mention a few points.

    First, in addition to RSS and UAH, there are at least two other groups looking at tropospheric temperature. There’s an NOAA/STAR group that looks at global temperatures (including in the tropics) and a UW group that looks at the tropics. Both of these groups were mentioned in the paper you’re discussing.

    Second, the NOAA group has been using the stratospheric cooling correction since at least 2004, and the UW has used it as well. RSS now uses the correction, leaving UAH as the only group that refuses to use it. Here are two examples of when this correction was used:
    NOAA, 2004: “Contribution of stratospheric cooling to satellite-inferred tropospheric temperature trends”
    UW, 2015: “Removing diurnal cycle contamination in satellite-derived tropospheric temperatures: understanding tropical tropospheric trend discrepancies”

    Third, there are at least 4 explanations for the divergence between models and tropospheric observations.
    1) Error in inputted forcings
    2) Observational uncertainty
    3) Internal variability
    4) Model physics error
    There explanations are covered at:
    pages 3-4:
    section 4:
    These explanations are not mutually exclusive. The previous “Journal of Climate” paper you mentioned, showed evidence of 2. The current paper you’re discussing points towards 1.

    • waxliberty

      thanks for the comment.

  • Greg King

    Nice article. The work by Mears and colleagues, the NOAA/STAR and Univ of Washington groups will be know to the UAH group (Christie and Spencer). Did you contact them to get their response? I expect they are doing their own evaluation and comparisons as they are the odd man out and it is important that they respond.

  • Mark Pawelek

    Zeke Hausfather wrote:

    “When greenhouse gases trap incoming solar radiation …”

    According to Zeke’s own AGW hypothesis, GHG does not trap radiation, and is, pretty much, transparent to solar radiation, incoming or outgoing. He should’ve said “When greenhouse gases slow down outgoing black body radiation”. It’s only 2 more words, and far more accurately reflects the AGW hypothesis. When alarmists don’t believe their own hypothesis why should I trust what they say?

    • DennisHorne

      GHGs trap the incoming solar energy on the way out. But I can see this ambiguity is totally wrong and has also thrown doubt on evolution, plate tectonics, quantum theory, relativity and probably Archimedes Principle as well. 😉

      • Mark Pawelek

        It’s not the “same radiation”. The black body radiation emitted by warm matter is not solar radiation. Nor do GHGs trap it. GHG slows the loss of LWIR to space.

        It’s a shame activist scientists are abusing their position as scientists to overstate their case. They are bringing all science into disrepute. A shame other scientists are supporting these people too.

        • DennisHorne

          Where does the energy radiated back to space come from?

          To all intents and purposes no scientist is abusing or overstating the case or bringing science into disrepute.

          You are just another common or garden-variety denier in camouflage and your ammo is all blank.

  • “When greenhouse gases trap incoming solar radiation …”
    Show us the equation , ie : differential , by which an electromagnetic , ie : spectral , phenomenon asymmetrically “traps” heat . Or an experimental demonstration of the effect .

    You can’t . Neither exists .

    • Any IR spectrum of the sky prove you wrong.

      • Give me a spectrum of the planet as seen from the outside and I will give you the equilibrium temperature computed for it . That’s the major thrust of my website at . ( Spectral table please ; I don’t have time or interest in trying to convert a graph . )

        That spectrum , I believe , computes a temperature perhaps 20K lower than the ~ 279 of a gray ball in our orbit .

        I asked for equations , not data . What you can’t show is either equation or experiment demonstrating the spectral phenomenon which eg : James Hansen claims is responsible for the difference between that computed “radiative balance” measurement and our 288K estimated surface temperature .

        • Remove the atmosphere and you have a no IR from the atmosphere hitting the ground. At the atmosphere, you have the IR hitting the ground and heating it. Off course, overall energy conservation apply at the equilibrium.

          • Equations please .
            All else is untestable word waving .

          • Heu, this is obvious. There is no need for equations. Without atmosphere the sky is colder than with an atmosphere. Hence the ground is also colder.

            If you want math readn anything on radiate transfer. This theory has been tested and apply for 70 years.

          • Ther absolutely IS need for equations . The variance is much larger but the mean only depends on the absorption=emission spectrum as seen from the sources and sink .

            Your claim , and the major GHG claim is that atmospheres “trap” heat via some spectral effect causing , in the extreme case , Venus’s surface temperature to be 2.25 times the temperature of a gray ball next to it .

            That’s a quantitative absurdity when you run the equations for radiative balance .

            So I say again , show us your equations . Mine are at . Show us where they are wrong .

          • This is trivial. Ground is hotter with an atmosphere that without, because anything is larger than zero. You need no more to understand. If you want to do the math to get a precise number, you have to integrate the flux for every absorption lines in 40 layer of the atmosphere.

            Off course, overall the conservation of energy still apply and you end up with a top of the atmosphere temperature in equilibrium.

            I suggest you by a book about radiative tranfers. There are full of equation.

          • Equations please .
            All else is untestable word waving .

            Do you in fact believe Venus’s extreme surface temperature 2.25 times the gray body temperature in its orbit , or even our surface temperature 1.03 times the gray body temp in our orbit or closer to 1.12 times the radiative balance temperature calculated for our spectrum as seen from space is due to an optical , ie : spectral , effect ?

            I suggest you review the constraints of the Divergence Theorem .

          • Try that: Proven true for 57 years and counting.


            I think you have missed some very basic courses.

          • I have the book , a true classic . It’s good to know it’s now online since my copy is on probably permanent loan to my niece since she was in the physics program at Boulder .

            Show me where it makes any claim that some stack of spectral filters can trap a higher energy density on the side away from a radiant source .

            Because that is your claim . It just takes one differential equation . And an example setup demonstrating the trapping would be nice .But if you can do it , let’s go into business together because we will have ourselves a perpetual heat engine .

            I take it you disagree with my computations at . Could you please post your corrections .

          • Thermal gradients still exist. This simply reduce the heat transfer rate from the surface to space. Upper atmosphere is colder by the way.

            So the thermodynamic is preserved.

          • I think this is your least responsive reply yet .
            Yes the thermal gradient exists . That is what needs to be explained . And if you can’t explain Venus you can’t explain any .
            Have looked at my Heartland YouTube ?
            I present equations and computations . You so far have only presented qualitative verbiage .

            All it would take is 1 equation .
            But such an equation doesn’t exist .

            Hint : as you will see , some time after the Heartland presentation , it got thru to me that the reasons atmospheres , whatever their optical properties , are hotter at their bottoms than that calculated for their radiative equilibrium is that other macroscopic force so inexcusably absent from the computations of total energy balance , gravity . And several people have worked some of those computations out with striking agreement with observed “lapse rates” .

          • Well, gravity is include in atmospheric model.

            And indeed, you will observed the adiabatic gradient at first order, but the radiation transfer is also important.

            If you increase the optical opacity in IR, this increase the thermal gradient (you reduce the heat transfer coefficient).

  • Your are wrong. Gravity is always include in climate model. Saturation is only true in then center of the CO2 band at low altitude. Also, you have to add the additional contribution of the water vapor.

    • Simply false . Gravity is included to explain pressure but not temperature . Here Hansen explicitly describes Venus’s “runaway” greenhouse effect being due to CO2 optical properties , not weight : .

      This is why I continue to demand equations . You have yet to offer any to explain how the interior of a ball , ie : the bottom of a planet’s atmosphere , can be hotter than that calculated for its radiative equilibrium from its spectrum as seen from space — given the violation of the Divergence Theorem , essentially basic thermodynamics that heat flows from hot to cold .

      Here does provide equations and computations and remarkable agreement with observation . As you can see from my my driving interest is executable notation to express these computations including any mappings of them over surfaces or volumes , so I won’t work thru HockeySchtick’s or any of the other similar derivations until motivated by someone else’s interest .

      But here on the gravity side , you have explicit quantitative equations tested against observation while on the GHG spectral side you have NONE . NADA , NIL . And glaring violation of basic physical principles .

      So I repeat : Show Us Your Equation .

Related Articles


Expert analysis directly to your inbox.

Get a Daily or Weekly round-up of all the important articles and papers selected by Carbon Brief by email.


Expert analysis directly to your inbox.

Get a Daily or Weekly round-up of all the important articles and papers selected by Carbon Brief by email.