Surface air temperature is monitored using thermometers. Temperature above the surface can also be measured with thermometers using radiosondes, radio transmitters which relay weather information to the ground from instruments usually attached to balloons and allowed to rise through the atmosphere. However, there are some distinct problems with balloon-borne radiosonde data: there’s poor coverage, especially over the oceans, and both instruments and procedures have changed over the years, making it difficult to determine long-term trends with accuracy. But since late 1978, temperature in various layers of the atmosphere has been measured by satellites carrying a microwave sounding unit (MSU), or its more advanced cousin the advanced microwave sounding unit (AMSU). The satellite missions were intended to aid weather studies rather than climate studies, but the data they have returned has been examined for its implications relating to climatic change in earth’s atmosphere.
Of course the satellite measurements are not without their difficulties too. The data come not from a single satellite but from more than a dozen, starting with TIROS-N in late 1978 and extending through NOAA-17 and AQUA. This leads to the tricky problem of pieceing the various data sets together into a continuous time series. Also, the instruments age over time, and more important the orbit changes. Not only have the satellite orbits slowly decayed (getting a bit lower as the years pass), they have also drifted. Drift affects the time at which the satellite crosses the equator, and hence the time of day at which measurements apply; since temperature depends strongly on the diurnal cycle (it’s hotter at noon than midnight!) it’s necessary to correct carefully for satellite drift in order to have a good representative measure of the atmosphere’s temperature which is not affected by the day-night cycle.
And as if that weren’t complicated enough, satellites don’t measure the temperature in a thin atmospheric layer but in a large region. The two microwave “channels” which have been most studied for climate information are MSU channels 2 and 4. Channel 2 (giving temperature T2) measures mostly the troposphere but is also affected by the surface and the stratosphere, while channel 4 (giving temperature T4) is almost entirely reflective of temperature in the stratosphere.
The initial studies of MSU data indicated that the troposphere wasn’t warming nearly as fast as the surface, in fact it wasn’t clear that the troposphere was warming at all. But by 2000 some of the problems with handling the data had been identified and attempts were made to compensate for them. Much of this work was carried out by a team at the University of Alabama at Huntsville, the UAH team. They attempted to correct for some of the effect of orbit decay and drift, and of stratospheric influence on the tropospheric T2 channel, by combining data taken at different viewing angles. Using information from observations straight downward (nadir) and those taken near the horizon, they devised a method to compensate some of the influence which was stratospheric rather than tropospheric (Christy et al. 2000). They also created a time series designed to represent the temperature, not in the mid-troposphere, but in the lower troposphere, creating the “T2LT” data set, which is not directly observed by the satellite instruments but is derived by transforming other satellite observations. By 2000 they estimated that the trend in tropospheric temperature from 1979 onward was 0.04 K/decade while the lower-troposphere trend was 0.06 K/decade (the surface trend from that time is about 0.17 K/decade).
The warming trends found were less than expected. Computer models of global climate predicted that the troposphere should warm faster than the surface; for the globe as a whole the tropospheric warming should be about 20% larger than surface warming (Hansen et al. 2002, J. Geophys. Res. 107, doi:10.1029/2001JD001143), and in the tropics the tropospheric warming should be about 50% larger (Hegerl & Wallace 2002, J. Clim. 15, 2412–2428). It’s important to emphasize that this is not a prediction which depends on global warming being due to man-made greenhouse gases. Enhanced tropospheric warming is common to all causes of global warming, be it due to greenhouse gases, solar variability, or whatever. Hence enhanced tropospheric warming cannot be used to determine the cause of global warming.
A complete re-analysis of MSU/AMSU data was done by a team at Remote Sensing Systems, leading to a rival of the UAH data set, the RSS time series (Mears et al. 2003). Using what they consider to be superior corrections for orbital changes, and a better way of merging the various satellite data sets, by 2003 the RSS data showed considerably more tropospheric warming than the UAH data, but still not as much as expected from computer models. Mears et al. summarize the differences thus:
|Diurnal adjustment: origin||Derived using hourly output from the CCM3 climate model, and a radiative transfer model||Derived using observed cross-scan differences measured by the MSU instruments|
|Diurnal adjustment: methodology||Adjustments all made to a single reference time (noon, local time)||Adjustments made to different reference times for each satellite (local time of first observation for each moth of instrument operation)|
|Determination of target factors||All valid 5-day averages with simultaneous observations for two or more satellites used in least squares regression procedure||Only long-term periods of satellite overlap used; some target factors set to zero if their improvement to the intersatellite differences are insufficient to warrant use|
|Determination of intersatellite offsets||Determined in a unified way during the regression procedure for determining target factors||Determined using a single path or ‘‘backbone’’ that links together the various satellites|
|Smoothing before target factor determination||5-day averages||60–120-day averages|
All the differences in methodology lead to differences in the result, but the primary difference arises from the different methods of merging different satellite data sets:
A more important difference between our methodologies is the way in which we determine the intersatellite merging parameters. We use a unified approach where each overlapping pentad average is treated with equal weight to determine both the target factors and the intersatellite offsets. The equal weighting of each 5-day overlap serves to deemphasize periods of short overlap without ignoring them altogether. Christy et al. (2000, 2003) impose a minimum time period over which an overlap must occur before it can be taken into account to help determine the merging parameters. This leads CS to discard the TIROS-N–NOAA-6, NOAA-7–NOAA-9, NOAA-8–NOAA-9, NOAA-9–NOAA-10, and NOAA-10–NOAA-12 overlaps when determining their target factors. Their intersatellite offsets are then determined by evaluating the mean difference between coorbiting satellites utilizing a single path that connects all the satellites in question.
Writing in 2003, Mears et al. estimated the global tropospheric warming trend at 0.097 +/- 0.02 K/decade; by that time the UAH group had updated their procedures such that their estimated trend was 0.09 K/decade, higher than their previous estimate but still lower than that of the RSS group and that expected from computer model simulations.
Meanwhile Vinnikov & Grody, in yet another analysis, suggested that the method used by the UAH group was insufficient to compensate for the impact of the diurnal cycle on the observations. As they say, “Christy’s group used the MSU measurements by themselves to evaluate the diurnal cycle. Unfortunately, however, the diurnal cycle cannot be validated with in situ data because of insufficient observations.” Using simultaneous determination of the instrument calibrations and effect of the diurnal cycle to create yet another data set which I will call the VG data. Vinnikov & Grody estimated the global tropospheric temperature trend at 0.22 to 0.26 deg.C/decade, actually larger than predicted by climate models. In more recent work, they apply modified least squares rather than ordinary least squares, which reduces the estimated trends to 0.21 K/decade for the tropics and 0.20 K/decade for the globe as a whole.
But it was still not clear that these procedures remove all the influence of the stratosphere from the T2 data, to give a time series which truly represents the troposphere without stratospheric influence. For this reason, Qiang Fu and colleagues at the University of Washington produced yet another version of estimated tropospheric temperature based on satellite observations (Fu et al. 2004). They estimate the stratospheric contribution to MSU channel 2 temperatures using MSU channel 4, which records only stratospheric temperatures, calibrating the correction using radiosonde measurements. As they say in the abstract, “The resulting trend of reconstructed tropospheric temperatures from satellite data is physically consistent with the observed surface temperature trend. For the tropics, the tropospheric warming is ~1.6 times the surface warming, as expected for a moist adiabatic lapse rate.” Hence a time series of tropospheric temperature was published which did not contradict computer model predictions of faster warming in the troposphere than at the surface (slightly faster globally, much faster tropically).
The University of Washington analysis can be applied to both the RSS and UAH data sets, leading to two further time series of tropospheric temperature, the UW-RSS and UW-UAH data sets. The UW method can also be applied to the analysis of Vinnikov & Grody, yielding a very large estimated tropospheric warming trend of 0.33–0.37 K per decade.
Most of the T2 data sets are available online, including RSS, UAH, UW-RSS, UW-UAH, and VG. I decided to determine the trend rates from each with the most up-to-date data sets, and simply to examine the data to see how they portray temperature evolution in the troposphere. The monthly averages have so much noise that to plot the different data sets together simply creates a confusion of dots, but by smoothing the data (on a 1/2-year time scale) we can get a better idea of their time evolution:
The most obvious thing is that all the data series tend to rise and fall together, and show considerable short-timescale detail. Hence they give a very good picture of the changes which are related to weather rather than climate, so fulfilling their intended purpose admirably. But in terms of long-term trends, there is considerable disagreement, with the VG series showing stronger warming than the others and the UAH series showing considerably weaker warming.
Linear regression (on the data rather than the smoothed series) gives us an idea of the trend rates using the most up-to-date data, and their probable errors. I emphasize that these errors (which are two standard deviations) reflect only the uncertainty due to the regression alone, they do not reflect any uncertainty in the data values themselves. The global trends (in addition to the surface warming rate from NASA GISS data) are:
The trends for the tropics only are:
All the analyses except UAH are compatible with computer model projections of tropospheric warming; the error ranges include the values expected from model simulations. The UAH analysis, however, is incompatible with model simulations, showing warming which is just too little to accord with model results.
Which of the data sets it to be believed? Frankly, I don’t know. I would say that the UW analysis seems to me to have a much better way of compensating for the stratospheric influence on the T2 channel than its predecessors, and I would also say that in my opinion, the VG method for computing the diurnal effect seems by far the most logical. However, the RSS and UAH analyses are more often referred to in the scientific literature; whether this is due to their having been first, and having existed in the literature longer, I don’t know. I also don’t know enough about the details of how the instrument biases are corrected, or how the different satellite data sets are combined, to say with confidence which analysis is closer to the truth.