Identification of human-induced changes in atmospheric moisture content

doi:10.1073/pnas.0702872104

Identification of human-induced changes in atmospheric moisture content

^aProgram for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, Livermore, CA 94550;
^cRemote Sensing Systems, Santa Rosa, CA 95401;
^dNational Center for Atmospheric Research, Boulder, CO 80307;
^eScripps Institution of Oceanography, La Jolla, CA 92037;
^fInstitut für Unternehmensforschung, Universität Hamburg, 20146 Hamburg, Germany;
^gClimatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, United Kingdom;
^hNational Institute for Environmental Studies, Tsukuba 305-8506, Japan;
ⁱHadley Centre for Climate Prediction and Research, United Kingdom Meteorological Office, Exeter EX1 3PB, United Kingdom; and
^jLawrence Berkeley National Laboratory, Berkeley, CA 94720

Edited by Inez Y. Fung, University of California, Berkeley, CA, and approved July 27, 2007 (received for review March 27, 2007)

Abstract

Data from the satellite-based Special Sensor Microwave Imager (SSM/I) show that the total atmospheric moisture content over oceans has increased by 0.41 kg/m² per decade since 1988. Results from current climate models indicate that water vapor increases of this magnitude cannot be explained by climate noise alone. In a formal detection and attribution analysis using the pooled results from 22 different climate models, the simulated “fingerprint” pattern of anthropogenically caused changes in water vapor is identifiable with high statistical confidence in the SSM/I data. Experiments in which forcing factors are varied individually suggest that this fingerprint “match” is primarily due to human-caused increases in greenhouse gases and not to solar forcing or recovery from the eruption of Mount Pinatubo. Our findings provide preliminary evidence of an emerging anthropogenic signal in the moisture content of earth's atmosphere.

Footnotes

^bTo whom correspondence should be addressed. E-mail: santer1{at}llnl.gov
Author contributions: B.D.S., K.E.T., T.M.L.W., T.P.B., S.A.K., and D.W.P. designed research; B.D.S., K.E.T., P.J.G., J.S.B., and P.A.S. performed research; C.M., F.J.W., K.E.T., T.M.L.W., W.B., and N.P.G. contributed new reagents/analytic tools; B.D.S., P.J.G., and M.F.W. analyzed data; and B.D.S., C.M., F.J.W., K.E.T., P.J.G., T.M.L.W., T.P.B., J.S.B., W.B., N.P.G., S.A.K., G.A.M., T.N., D.W.P., P.A.S., W.M.W., and M.F.W. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
This article contains supporting information online at www.pnas.org/cgi/content/full/0702872104/DC1.
↵ k The domain for spatial averaging was 50°N–50°S. This domain was chosen to minimize the effect of model-vs.-SSM/I water vapor differences associated with inaccurate simulation of the latitudinal extent of ice margins. All data points over land were masked out before calculation of spatial averages. Averages were area-weighted, properly accounting for both complete and fractional grid-cells within the region.
↵ l Ensembles of the 20CEN simulations were performed with 15 of the 22 models analyzed here (see SI Text). Each ensemble contains multiple realizations of the same experiment, differing only in their initial conditions but with identical changes in external forcings. This approach yields many different realizations of the climate “signal” (the response to the imposed forcing changes) plus climate noise. Averaging over multiple realizations reduces noise and facilitates signal estimation. Here, we calculate averages over ALL and ANTHRO 20CEN runs. In each case, x̄ is the arithmetic mean of the individual model results, i.e., x̄ = 1/N Σ_j=1 ^N x̄_j, where N is the total number of ALL or ANTHRO models (12 and 10, respectively) and x̄_j is the ensemble mean signal (or individual realization) of the jth model. This averaging method avoids undue emphasis on results from a single model with a large number of realizations.
↵ m The total changes in 〈W _o〉 over 1900–1999 are 1.37 kg/m² and 1.52 kg/m² for the ALL and ANTHRO models, respectively. These represent increases of 4.8% and 5.2% relative to the ALL and ANTHRO model climatological annual mean values of 〈W _o〉 over 1900–1909. Changes are defined as b × n, where b is the slope parameter of the linear trend (in kilograms per square meter per month) fitted by the standard least-squares method over a specified period of n months. Note that 〈W _o〉 is sometimes expressed in millimeters (12). The conversion factor between kilograms per square meter and millimeters is 1.
↵ n Note that Fig. 3 shows weak evidence of a relationship between the mean state and temporal variability of 〈W _o〉 (Fig. 3 A) and stronger evidence that higher-amplitude variability of Niño 3.4 SSTs leads to greater coherence between SST and 〈W _o〉 fluctuations (Fig. 3 C). The observed values of these quantities are reasonably well captured by the ALL and ANTHRO model averages. The model averages are also close to the observed climatological mean 〈W _o〉 and to the SSM/I water vapor trend over 1988 to 1999 (Fig. 3 A and B).
↵ o The centered (spatial means removed) pattern correlations between the observed W _o changes in Fig. 4 E and the ALL and ANTHRO model fingerprints in Fig. 4 A and B are 0.50 and 0.52, respectively. The corresponding values for the correlation between the observed W _o changes and the leading ALL and ANTHRO model noise modes in Fig. 4 C and D are 0.19 and 0.28. The noise modes were estimated by calculating EOFs from two pooled data sets (see SI Text) consisting of concatenated control run W _o data from the ALL and ANTHRO models (26). Because the signs of the EOFs are arbitrary, only absolute values of the pattern correlation are given.
↵ p One further sensitivity test involved repeating our entire fingerprint analysis with patterns of percentage changes in W _o. Anomalies in each individual data set (observations, 20CEN runs, and control integrations) were defined relative to the overall climatological annual mean of the data set and then converted to percentage changes. This procedure reduces the possible impact of model moisture biases on the estimated signals and noise. Fingerprint patterns are more uniform, because per degree Celsius increase, the percentage change in W _o is much closer to being a constant than is the actual change in W _o, which increases rapidly with increasing temperature. When spatial means are included, the use of percentage changes yields positive and consistent detection of an anthropogenic fingerprint, with detection times similar to those shown in Table 1 for actual changes in W _o. Because the “percentage change” fingerprint is spatially more uniform than the fingerprints shown in Fig. 4 A and B, it is less meaningful to explore the detectability of a mean-removed fingerprint.
↵ q Mount Pinatubo's influence on simulated W _o trends depends on such factors as the length of the W _o time series, the proximity of the volcanic effect on W _o to the midpoint of the time series, the amplitude of the maximum cooling and drying, and the volcanic signal decay time.
Abbreviations:

CMIP,

Coupled Model Intercomparison Project;

EOF,

empirical orthogonal function;

GHG,

greenhouse gas;

S/N,

signal-to-noise ratio;

SSM/I,

Special Sensor Microwave Imager;

SST,

sea surface temperature.
Freely available online through the PNAS open access option.