A Report Produced by The CO2 & Climate Team

The Data Weigh In

Every climate model that is run with increasing atmospheric concentrations of greenhouse gases produces some degree of warming at earth’s surface and even greater warming above the surface, especially in the atmospheric layer between 5,000 to 30,000 feet in altitude (the troposphere). Models calculate this warming to be especially strong in the tropical half of the planet and weaker in a very small region around both poles. Observations of real world temperature trends in the lower atmosphere don’t confirm these model results and instead show that, generally, warming trends decline with altitude.

     Why is this important? The atmosphere is an integrated whole. Temperature aloft is an important determinant of temperature at the surface. If the models have the “upstairs” wrong but have it right “downstairs” in the area near the surface, they’ve been pretty lucky. Some might say, pretty “adjusted.”

     The discrepancy between models and observations is the crux of one of the major arguments against the models and over reliance on them to anticipate future climate. If the models can’t accurately portray present observations, they cannot be relied upon to predict the future.

     The hypothesis that models continue to get it wrong is strongly supported by results from a research effort led by The University of Rochester’s David Douglass and published in a pair of articles in Geophysical Research Letters (online on July 9, 2004). Two other scientists involved in the effort were Patrick Michaels and Paul Knappenberger, chief editor and technical supervisor these of World Climate Alerts.

     In the first paper, Douglass et al. compare temperature trends from the surface upward through the lower atmosphere as projected by three state-of-the-art climate models using several different sets of actual temperature observations made at various elevations in the atmosphere. Each of the three climate models (Hadley CM3, DOE PCM, and GISS SI2000) were run using historic values of a combination of natural climate forcing agents (solar variability and volcanic eruptions) and anthropogenic ones (greenhouse gases and aerosols). Results from the past two decades — the period of greatest anthropogenic influence on atmospheric chemistry — were compared with observations of temperature made during the same time. The observations come from sets of radiosonde data (collected by instruments attached to weather balloons ascending through the atmosphere) and temperature data collected by satellites.

     Douglass finds that while modeled and observed trends match well at the surface, everywhere else there is considerable discrepancy —the observed trends exhibit much less warming than model projections and, in some cases, the observations actually indicate a cooling trend (see Figure 1).

Figure 1. Modeled temperature trends (dotted lines) and observed temperature trends (solid lines) represented as 10-3K/decade in the lower atmosphere during two recent decades in four different regions. While the modeled and observed trends match well at the surface, in nearly every case, the modeled trends above the surface are greater than the observed trend (source: Douglas et al., 2004a).

   This comparison provides clear evidence that climate models fail to capture the actual workings of the atmosphere. Weather systems largely derive their characteristics (size, strength, precipitation efficiency, etc.) from temperature contrasts in the atmosphere. The models fail to accurately portray observed trends and also fail to accurately portray observed weather patterns. This is true not only of current climate patterns but also of modeled projections of our climate future.

     In the second paper, Douglas and his research team investigate whether the same set of observations can shed light on the hypothesis that satellite measurements of temperature trends in the earth’s lower atmosphere are contaminated by temperature trends at higher altitude, in the stratosphere. Highly publicized research by Fu et al. (2004) reported in Nature magazine suggests a reason satellite temperature measurements of the lower troposphere (as compiled by researchers at the University of Alabama-Huntsville) show only about half the warming trend as observations collected at the surface during the past twenty-five years. According to Fu, in formulating the trends of the lower atmosphere the UAH researchers fail to take into account a cooling influence from the stratosphere (the atmospheric layer just above the troposphere). Fu believes that if the stratospheric data were properly handled, the temperature trends in the lower atmosphere would match the surface (and modeled) trends much more closely.

     An earlier World Climate Alert argues why this is invalid — Fu’s solution involves a physically unrealistic combination of multiple temperature derivations (see http://www.co2andclimate.org/wca/2004/wca_17a.html for details). Douglass and his fellow researchers provide further demonstration of the inadequacies of this solution. They employed two completely independent temperature measuring systems to monitor temperature in the troposphere (stretching from about between 5,000 to 30,000 feet in altitude) including direct observations made by thermometers carried aloft by weather balloons and indirect observations made by satellite-borne instruments that record the temperature-dependent microwave emissions from oxygen molecules.

     Use of these two datasets provides the ability to crosscheck results, one against the other.

     If temperature trend calculations are performed thoroughly and carefully using the best possible accounting for all known data quality issues, the results are remarkably similar. The observed warming in the lower troposphere is about half that observed at the surface. As Douglass notes, this result runs contrary to climate model projections.

     The fact temperature trends recorded by weather-balloon instrumentation and satellite instruments match well isn’t news. We’ve highlighted this fact for years (e.g. http://www.co2andclimate.org/wca/2004/wca_15f.html). However, no one really knows why lower atmospheric trends differ from those measured at the surface or why observed lower atmospheric trends differ from model projections. Fu suggests it results from “unaccounted for” stratospheric influences. But that isn’t physically reasonable. Douglass’ work points to other possible causes.

     Douglass’s team focuses on surface temperatures and moisture derived from the balloon-based lower atmospheric observations. They compare the derived surface temperatures with observed surface temperatures and find the trends in the surface temperatures derived from the weather balloons more closely match the satellite temperature trends than do observed surface temperature trends from ground-based weather stations (Figure 2).

     Because the derived surface temperatures largely are free of local effects (e.g., urbanization, industrialization and land-use changes) and many other data quality issues known to plague surface measurements (see http://www.co2andclimate.org/wca/2004/wca_18c.html for details), Douglass concludes non-climatic influences likely play a large role in the discrepancy between observed surface temperature trends and observed lower atmospheric temperature trends. This also suggests climate models overestimate the warming from changes in atmospheric composition and confirm similar work focusing only on temperatures in the United States (Kalnay and Cai, 2003 — see http://www.co2andclimate.org/wca/2003/wca_1b.html for more details).

Figure 2. Temperature trends by 5º latitude bands (where data are available) in three different data sets: observed surface data (red circles), derived surface data (light blue circles), satellite-based lower tropospheric data (darker blue circles). All of these data are for the most part independent of one another. Notice how the trends from the derived surface data more closely match trends from satellite data than they do observed surface data (source: Douglass et al., 2004b).

    In sum, the results of research presented in Douglass’s two papers provide strong evidence for three important points:

1) The discrepancy between temperature trends measured at the earth’s surface and those measured in the earth’s lower atmosphere is real.

2) A large part of this discrepancy likely is caused by local, non-climatic influences on surface thermometers not by stratospheric contamination of the lower tropospheric data.

3) Climate models that include observed changes to known climate forcing agents (both natural and anthropogenic) are unable to replicate the observed behavior of the temperatures in the lower atmosphere. Furthermore, if local, non-climatic influences are largely responsible for the surface temperature trends, then the climate models are getting the surface trends right for the wrong reasons — indicating their failure at that level as well.

     Such findings should give pause to anyone who relies on climate model output to inform their decision-making.


Douglass, D.H., Pearson, B.D., Singer, S.F., 2004a. Altitude dependence of atmospheric temperature trends: Climate models versus observation. Geophysical Research Letters, 31, doi:10.1029/2004GL020103.

Douglass, D.H., et al., 2004b. Disparity of tropospheric and surface temperature trends: new evidence. Geophysical Research Letters, 31, doi:10.1029/2004GL020212 .

Fu, Q., et al., 2004. Contribution of stratospheric cooling to satellite-inferred tropospheric temperature trends. Nature, 429, 55-58.

Kalnay, E., and Cai, M., 2003. Impact of urbanization and land-use change on climate. Nature, 423, 528-531.