Because water vapor is the principal greenhouse gas, variations

Upper-tropospheric moistening in response to anthropogenic warming Eui-Seok Chung a, Brian Soden a,1, B. J. Sohn b, and Lei Shi c a Rosenstiel School of Marine and Atmospheric Science, University of Miami, Miami, FL 33149; b School of Earth and Environmental Sciences, Seoul National University, Seoul 151-747, Korea; and c National Climatic Data Center, National Oceanic and Atmospheric Administration, Asheville, NC 28801 Edited by John H. Seinfeld, California Institute of Technology, Pasadena, CA, and approved June 27, 2014 (received for review May 23, 2014) Water vapor in the upper troposphere strongly regulates the strength of water-vapor feedback, which is the primary process for amplifying the response of the climate system to external radiative forcings. Monitoring changes in upper-tropospheric water vapor and scrutinizing the causes of such changes are therefore of great importance for establishing the credibility of model projections of past and future climates. Here, we use coupled ocean atmosphere model simulations under different climate-forcing scenarios to investigate satellite-observed changes in global-mean upper-tropospheric water vapor. Our analysis demonstrates that the upper-tropospheric moistening observed over the period 1979 2005 cannot be explained by natural causes and results principally from an anthropogenic warming of the climate. By attributing the observed increase directly to human activities, this study verifies the presence of the largest known feedback mechanism for amplifying anthropogenic climate change. detection attribution long-term monitoring Because water vapor is the principal greenhouse gas, variations in its concentration strongly influence the climate s response to both anthropogenic and natural forcings (1). Changes in the amount of water vapor in the upper troposphere play a particularly important role because the trapping of outgoing terrestrial radiation is proportional to the logarithm of watervapor concentration (1, 2), and climate models predict enhanced moistening in the upper troposphere compared with the boundary layer (3). Although short-term fluctuations of upper-tropospheric water vapor are consistent among reanalysis datasets, decadal variations show substantial discrepancies even in sign (4, 5). Hence, long-term monitoring of upper-tropospheric water-vapor changes, and understanding causes responsible for such changes are essential for enhancing confidence in the prediction of future climate change (4, 6). Changes in upper-tropospheric water vapor have been examined based on satellite-observed radiances of 6.7-μm water-vapor channels (3, 7, 8), which are closely related to the layer mean relative humidity in the upper troposphere (9). Decadal trends in upper-tropospheric relative humidity exhibits distinct regional patterns associated with changes in the atmospheric circulation, but the decadal trends over larger domains are small due to opposing changes at regional scales (8). Analyzing the globalscale changes in 6.7-μm water-vapor radiances reveals little change over the past three decades. However, when the 6.7-μm radiances are examined relative to microwave radiance emissions from oxygen, a distinct radiative signature of upper-tropospheric moistening can be revealed (3). Although the presence of a moistening trend has been detected in the satellite record, the cause of this moistening has not been determined. Thus, it remains unclear whether the observed moistening could result from natural fluctuations in the climate system, or whether human activities have significantly contributed to the trend. Because climate feedbacks can behave differently in response to natural climate variations compared with anthropogenic warming (10), fully validating the presence and strength of this feedback ultimately requires the detection of a change in upper-tropospheric water vapor that is directly attributable to human activities. Given the importance of uppertropospheric water vapor, a direct verification of its feedback is critical to establishing the credibility of model projections of anthropogenic climate change. A new set of coordinated climate change experiments have been conducted for the fifth phase of the Coupled Model Intercomparison Project (CMIP5; ref. 11). One of the climate change scenarios included in the CMIP5 is a historical experiment in which coupled ocean atmosphere models are integrated with historical changes in forcing agents over the period 1850 2005. Climate variability produced from the historical experiment can then be analyzed in more detail in combination with two related experiments: one integrated with only anthropogenic forcings from well-mixed greenhouse gases, and the other integrated with only natural forcings from volcanoes and changes in solar activity. These two experiments can help identify the causes for recent changes in climate, provided the historical experiment with all forcings is capable of reproducing the observed trends. In this study, we use the historical climate change experiments from CMIP5 to demonstrate that the satellite-observed changes in upper-tropospheric water vapor are inconsistent with naturally forced variability and can only be explained by anthropogenic forcing. Temporal Variations and Trends of Upper-Tropospheric Water Vapor The National Oceanic and Atmospheric Administration (NOAA) operational polar-orbiting satellites have been taking measurements of the 6.7-μm water-vapor channel (channel 12) radiances from Significance The fact that water vapor is the most dominant greenhouse gas underscores the need for an accurate understanding of the changes in its distribution over space and time. Although satellite observations have revealed a moistening trend in the upper troposphere, it has been unclear whether the observed moistening is a facet of natural variability or a direct result of human activities. Here, we use a set of coordinated model experiments to confirm that the satellite-observed increase in upper-tropospheric water vapor over the last three decades is primarily attributable to human activities. This attribution has significant implications for climate sciences because it corroborates the presence of the largest positive feedback in the climate system. Author contributions: E.-S.C. and B.S. designed research; E.-S.C. and B.S. performed research; E.-S.C., B.S., B.J.S., and L.S. analyzed data; and E.-S.C., B.S., and B.J.S. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. Freely available online through the PNAS open access option. 1 To whom correspondence should be addressed. Email: b.soden@miami.edu. This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1409659111/-/DCSupplemental. EARTH, ATMOSPHERIC, AND PLANETARY SCIENCES www.pnas.org/cgi/doi/10.1073/pnas.1409659111 PNAS Early Edition 1of6

High-Resolution Infrared Radiation Sounder (HIRS) version 2 (HIRS/2) since November 1978. Because climate monitoring was not the primary purpose of the HIRS mission, various attempts have been made to correct for biases, and to minimize intersatellite discrepancies, to make the HIRS record more suitable for climate study (8, 12). The bias-corrected, intercalibrated HIRS water-vapor channel radiance dataset (13) is used to examine the decadal timescale variability of uppertropospheric water vapor. Unfortunately, the continuity of the 6.7-μm water-vapor record ends in 2005 due to the shift of central wavelength from 6.7 μm (HIRS/2) to 6.5 μm (HIRS/3), which also coincides with the end of the CMIP5 historical experiment. We therefore limit our observational analysis to the 27-y period 1979 2005. A time series of global, monthly mean brightness temperature anomalies of HIRS channel 12 (T12) is given in Fig. 1A (red line). Brightness temperature anomalies are computed relative to the mean seasonal cycle over the period 1980 2004. For the period 1979 2005, the brightness temperature anomalies vary within ±0.4 K, with only a very small positive trend over this period. The time series of HIRS channel 12 brightness temperature anomalies simulated from the CMIP5 historical experiment of 20 coupled ocean atmosphere models (Materials and Methods) is also presented in Fig. 1A. The multimodel ensemble mean is shown by the blue line, with vertical bars denoting the intermodel spread. Note that multimodel averaging dampens the amplitude of the monthly variability compared with that of the satellite observations. Nevertheless, the CMIP5 models capture the observed decadal variability despite substantial biases in climatological mean distribution (14). The observed linear trend in T12 (for more details about uncertainties in estimated linear trends, see SI Materials and Methods) is similar to that computed from the multimodel mean and lies near the center of the distribution of trends from the individual models (Fig. 1, Right). The small magnitude of the trend shown in both the satellite observations and the model ensemble confirms that global-mean upper-tropospheric relative humidity remained nearly constant on decadal timescales (3). In addition to HIRS instruments, the NOAA operational polar-orbiting satellites are equipped with a microwave sounding unit (MSU) that provides weighted-average temperature information for deep atmospheric layers between the surface and the stratosphere, by way of four channels located in the 60-GHz oxygen absorption band. The remote sensing systems (RSS) reprocessed the brightness temperatures from the MSU and its follow-on, advanced MSU (AMSU), to construct a bias-corrected, intercalibrated MSU/AMSU dataset (15). Fig. 1. Time series of global-mean brightness temperature anomaly from satellite observations and CMIP5 historical experiment simulations for: (A) T12, (B) T2, and (C) T2 T12. Error bars in light blue represent ±2 SD of the multimodel ensemble mean. The corresponding histograms of decadal trend from CMIP5 historical experiment simulations are given on Right with dashed lines in blue denoting a decadal trend of multimodel ensemble mean (T12: 0.05 ± 0.01 K decade 1, T2: 0.20 ± 0.10 K decade 1,T2 T12: 0.15 ± 0.04 K decade 1 ). The bin size of histograms is 0.02 K decade 1. The decadal trend of multimodel ensemble mean for which model-simulated changes in T2 T12 were computed under a constant water-vapor scenario is represented by a green dashed line ( 0.08 ± 0.10 K decade 1 ). Vertical lines in red represent a decadal trend from satellite observations (T12: 0.04 ± 0.04 K decade 1, T2: 0.17 ± 0.06 K decade 1, T2 T12: 0.13 ± 0.04 K decade 1 ). Horizontal error bars denote ±2 SE of the linear trend (±2 SE of the linear trend are computed using the method in ref. 20). 2of6 www.pnas.org/cgi/doi/10.1073/pnas.1409659111 Chung et al.

We use the MSU/AMSU channel 2 brightness temperatures (T2) in which the stratospheric contribution is removed using a combination of different viewing angles (16, 17). The time series of the observed T2 anomalies indicates sporadic warming and cooling associated with El Niño La Niña events with a distinct warming trend over this period. Although the amplitude of this interannual variability is not captured in the multimodel mean (blue line in Fig. 1B) because El Niño La Niña events do not occur simultaneously among the models, the multimodel ensemble mean of the historical experiment does show decadalscale warming that is consistent with the MSU/AMSU observations. The observed trend in T2 (Fig. 1, Right) is slightly smaller than that predicted by the multimodel ensemble mean although it lies well within the distribution of individual model trends and is consistent with previous studies (18, 19). The water-vapor channel radiances are influenced not only by variations in water-vapor concentration, which alter the atmospheric opacity, but also by atmospheric temperature variations, which alter the Plank emission. Although spatiotemporal variations in water vapor are significant, changes in atmospheric oxygen concentrations, which determine T2 emissions, are negligible (3). Thus, the difference T2 T12 measures the divergence in emission levels between upper-tropospheric water vapor and oxygen. This divergence provides a direct measure of the extent of upper-tropospheric moistening; i.e., the increased concentration of water vapor elevates the emission level for T12 and offsets the warming evident in T2, which experiences no change in emission level (3). Based on these properties, a time series of the brightness temperature difference, T2 T12, is constructed to quantify the global-scale changes in upper-tropospheric water vapor for both satellite observations and CMIP5 historical simulations (Fig. 1C). El Niño La Niña events dominate subdecadal-scale variations in the satellite observations but are not evident in the CMIP5 ensemble mean due to multimodel averaging. However, on decadal timescales both the satellite observations and the coupled ocean atmosphere model simulations exhibit a distinct increase in global-mean upper-tropospheric water vapor. Moreover, the observed linear trend in T2 T12 is very similar to that predicted by the multimodel mean and lies near the center of the distribution of individual model trends. To demonstrate that the difference T2 T12 is a measure of the concentration of upper-tropospheric water vapor, the modelsimulated changes in T2 T12 are computed by holding the concentration of water vapor constant over time. The resulting trend (green line in Fig. 1C) is near zero and lies well outside both the model simulations with changing water vapor and the observed trend. Both calculations use the same sets of temperature profiles, indicating that the increase in T2 T12 is due to the increased concentration of water vapor in the upper troposphere and not due to changes in temperature. EARTH, ATMOSPHERIC, AND PLANETARY SCIENCES Fig. 2. Time series of global-mean brightness temperature anomaly of (A) T12, (B) T2, and (C) T2 T12, simulated from CMIP5 preindustrial control experiment for a 27-y period. Each line denotes an individual coupled ocean atmosphere model. The corresponding histograms of decadal trend are given on Right with the bin size of 0.02 K decade 1. The decadal trend of multimodel ensemble mean for which model-simulated changes in T2 T12 were computed under a constant water-vapor scenario is represented by a green dashed line (0.00 ± 0.01 K decade 1 ) with a horizontal error bar denoting ±2 SE of the linear trend. Vertical lines in red represent decadal trends from satellite observations over the period 1979 2005. Chung et al. PNAS Early Edition 3of6

Fig. 3. Decadal trends of multimodel ensemble mean brightness temperature simulated from CMIP5 historical experiment (red circles), HistoricalNat (blue triangles), and HistoricalGHG (green triangles) for five 30-y periods: (A) T12, (B) T2, and (C) T2 T12. Error bars denote ±2 SEofthe linear trend. Detection and Attribution of the Moistening Trend To examine whether internally generated variability could produce the moistening trend, we analyze output of the corresponding CMIP5 preindustrial control run (11), which contains only unforced, internal climate variability. In contrast to the historical experiment, none of the simulated brightness temperature records (T12, T2, or T2 T12) show a significant trend over a 27-y period (Fig. 2). The histograms presented on the right further demonstrate that decadal trends with a magnitude equal to that observed do not occur in any of the unforced experiments. A constant water-vapor scenario results in nearzero trends in T2 T12. These results suggest that the uppertropospheric moistening observed during the satellite era does not result from internal variability but from a combination of historical changes in anthropogenic and natural forcings. To examine different forcing contributions, we assess the relative contribution of anthropogenic greenhouse gases to historical changes in the upper-tropospheric water vapor by analyzing two additional CMIP5 experiments linked to the historical experiment. In these experiments, the coupled ocean atmosphere models are integrated with anthropogenic greenhouse gases (i.e., historicalghg), and with natural forcing sources (i.e., historicalnat), respectively. For 12 out of 20 models in which output is available for all three experiments, the decadal trends are computed for the five 30-y periods. Fig. 3 compares decadal trends for the multimodel ensemble mean with horizontal error bars denoting ±2 SE of the linear trend (±2 SE of the linear trend are computed using the method in ref. 20). Decadal trends of the model-simulated T12 show both positive and negative values for the historicalnat experiment, but signs are predominantly positive for the historicalghg experiment. Although the influences of changes in aerosols and land use cannot be ruled out, an increase in anthropogenic greenhouse gases seems to be responsible for the decadal trend over the satellite era, because trends from the historical and historicalnat experiments lie clearly outside each other s range. For decadal trends of T2, the range of estimated decadal trends is generally wider for historicalnat than historicalghg (Fig. 3B), indicating that subdecadal variability could be more significant in the former. The increase of anthropogenic greenhouse gases consistently leads to a warming trend for all periods. Although the impact of natural forcing sources can negate greenhouse-gas-induced warming signals (e.g., for the period 1946 1975; refs. 21 23), it is mostly weaker and more variable. Given these characteristics, the warming trend over the satellite era is primarily attributable to the increase of anthropogenic greenhouse gases. For the T2 T12 (Fig. 3C), decadal trends for historicalghg and historicalnat fall within each other s range for the first two periods, but become significantly different from each other in later periods. The magnitude of decadal trend due to natural forcings is generally small, whereas the contribution of anthropogenic greenhouse gases is always positive, and is amplified throughout the whole period. Comparisons with the historical experiment indicate that decadal trends for the historical experiment are affected by changes in natural forcing sources, as well as anthropogenic greenhouse gases. For example, a negative (thus drying) trend for the historical experiment over the period 1946 1975 is mainly induced by natural forcing sources, because increases in anthropogenic greenhouse gases induce a significantly positive (moistening) trend. Concerning the satellite-derived moistening trend in recent decades, the relations of trend and associated range among three experiments lead to the conclusion that an increase in anthropogenic greenhouse gases is the main cause of increased moistening in the upper troposphere. Discussion and Conclusions To illustrate the importance of the observed upper-tropospheric moistening in amplifying the climate sensitivity, radiative kernels (24 26) are used to quantify the strength of the water-vapor feedback from all levels with that obtained for the upper troposphere alone (SI Materials and Methods). The histogram in Fig. 4 compares the distribution of model-simulated water-vapor feedback during the historical scenario with and without historical forcings. Simulations with anthropogenically induced warming Fig. 4. The histogram shows a distribution of the water-vapor feedback strength computed using a radiative kernel for two 10-y periods (i.e., 1979 1988 and 1989 1998) of the historical scenario with a red line denoting a multimodel mean (1.92 ± 0.99 W m 2 K 1 ). The bin size of the histogram is 0.2 W m 2 K 1. A blue dashed line indicates a multimodel mean of the water-vapor feedback strength for which water vapor in the troposphere would change under natural forcing alone (i.e., HistNat), and the case that the evolution of upper-tropospheric water vapor was not modified by HistNat is represented by a green dashed line (i.e., Hist UTWV-only). The multimodel mean values for the HistNat and Hist UTWV-only are 0.08 ± 0.99 Wm 2 K 1 and 1.53 ± 0.87 W m 2 K 1, respectively. Horizontal error bars represent ±2 intermodel SD. A violet dashed line denotes the observational estimate of the water vapor feedback for the period 2000 2010 ( 1.2 W m 2 K 1 )(27). 4of6 www.pnas.org/cgi/doi/10.1073/pnas.1409659111 Chung et al.

Table 1. A list of CMIP5 climate models used in this study Modeling center Model name Model expanded name Commonwealth Scientific and Industrial Research Organization and Bureau of Meteorology ACCESS1-0 Australian Community Climate and Earth System Simulator, version 1.0 Beijing Normal University BNU-ESM Beijing Normal University Earth System Model National Center for Atmospheric Research CCSM4 Community Climate System Model, version 4 Centre National de Recherches Météorologiques CNRM-CM5 CNRM Coupled Global Climate Model, version 5 (CNRM) NOAA/Geophysical Fluid Dynamics Laboratory (GFDL) GFDL-CM3 GFDL Climate Model, version 3 NOAA/GFDL GFDL-ESM2G GFDL Earth System Model with Generalized Ocean Layer Dynamics component NOAA/GFDL GDFL-ESM2M GDFL Earth System Model with Modular Ocean Model 4 (MOM4) component National Aeronautics and Space Administration GISS-E2-R GISS Model E2, coupled with the Russell ocean model Goddard Institute for Space Studies (GISS) Met Office Hadley Centre HadGEM2-ES Hadley Centre Global Environment Model, version 2-Earth System Institute for Numerical Mathematics (INM) INM-CM4 INM Coupled Model, version 4 Institut Pierre-Simon Laplace (IPSL) IPSL-CM5A-LR IPSL Coupled Model, version 5, coupled with Nucleus for European Modelling of the Ocean (NEMO), low resolution IPSL IPSL-CM5A-MR IPSL Coupled Model, version 5, coupled with NEMO, mid resolution Japan Agency for Marine Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo); and National Institute for Environmental Studies MIROC-ESM-CHEM Model for Interdisciplinary Research on Climate Earth System Model, atmospheric chemistry coupled version Japan Agency for Marine Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo); and National Institute for Environmental Studies Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, and Japan Agency for Marine-Earth Science and Technology MIROC-ESM Model for Interdisciplinary Research on Climate Earth System Model MIROC5 Model for Interdisciplinary Research on Climate, version 5 Max Planck Institute for Meteorology (MPI-M) MPI-ESM-LR Max Planck Institute Earth System Model, low resolution MPI-M MPI-ESM-P Max Planck Institute Earth System Model, paleo version Meteorological Research Institute (MRI) MRI-CGCM3 MRI Coupled Atmosphere-Ocean General Circulation Model, version 3 Norwegian Climate Centre (NCC) NorESM1-M Norwegian Earth System Model, version 1 (mid resolution) NCC NorESM1-ME Norwegian Earth System Model, version 1 (version with carbon cycle) EARTH, ATMOSPHERIC, AND PLANETARY SCIENCES simulate large positive feedbacks from water vapor and are distinctly different from generated from natural forcing alone (blue dashed line). To highlight the importance of the upper troposphere, the feedback calculations are repeated using only watervapor changes in the troposphere above 600 hpa from the historical simulation (green dashed line in Fig. 4). Approximately 80% of the total water-vapor feedback results from water vapor in the upper troposphere. Although the absolute increase in water vapor is small at these levels, the absorptivity scales with the fractional changes in water vapor, which are typically 2 3 times larger in the upper troposphere compared with the surface (SI Materials and Methods). Note that the observational estimate for the period 2000 2010 (27) lies within the distribution of model simulations only when anthropogenic forcing is included, further indicating that the observed changes in upper-tropospheric water vapor are a direct result of anthropogenic warming. Bias-corrected, intercalibrated satellite observations produce a radiative signature, suggesting that moisture in the upper troposphere has increased over the past 30 y (3). When integrated with historical changes in forcing agents, coupled ocean atmosphere models are found to produce decadal trends consistent with satellite observations. In contrast, coupled ocean atmosphere models fail to capture observed trends in the preindustrial control experiment, suggesting that upper-tropospheric moistening over the satellite era is not an internally generated variability. Two additional model experiments, integrated with anthropogenic greenhouse gases and natural forcing sources separately, further indicate that the observed moistening trend is mainly induced by an increase in anthropogenic greenhouse gases. As a result, it is expected that the influence of a projected increase in anthropogenic greenhouse gases will amplify upper-tropospheric moistening, and is thus likely to amplify global warming via enhanced water-vapor feedback. Materials and Methods Decadal trends of upper-tropospheric water vapor determined from the satellite observations are compared with those simulated from CMIP5 coupled ocean atmosphere climate models, to ascertain whether the satellitedetermined decadal-scale variations are due to anthropogenic forcing agents. In doing so, the historical experiment output from 20 climate models (ACCESS1-0, BNU-ESM, CCSM4, CNRM-CM5, GFDL-CM3, GDFL-ESM2G, GFDL-ESM2M, GISS-E2-R, HadGEM2-ES, INMCM4, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC- ESM-CHEM,MIROC-ESM,MIROC5,MPI-ESM-LR,MPI-ESM-P,MRI-CGCM3, NorESM1-M, and NorESM1-ME; see Table 1 for information about climate models) is contrasted with the corresponding preindustrial control run results (i.e., picontrol) that represent an unforced climate variability. The output of ocean atmosphere coupling experiments is analyzed, as suppressing the ocean atmosphere interactions could inhibit the internally generated variability that might not be in phase with externally forced variability (28, 29). Chung et al. PNAS Early Edition 5of6

Forcing agents included in the historical experiment are: well-mixed greenhouse gases, tropospheric and stratospheric ozone, land use, volcanoes, solar forcing, sulfate, black carbon, organic carbon, dust, and sea salt, and their detailed prescriptions may vary depending on models. The CMIP5 includes two additional experiments designed to investigate the response of the climate system to changes in anthropogenic sources (i.e., historicalghg), and natural sources (historicalnat). Ref. 11 provides detailed information on the CMIP5 experiments. To avoid uncertainties inherent to the inversion processes of satellite-observed radiances, atmospheric profiles of temperature and specific humidity produced from the CMIP5 experiments are inserted into a fast radiative transfer model (30) to compute synthetic brightness temperatures that would be observed by satellites for given atmospheric conditions. ACKNOWLEDGMENTS. We thank two anonymous reviewers and the editor for their constructive and valuable comments, which led to an improved version of the manuscript. We acknowledge the World Climate Research Programme s Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modelling groups (listed in Materials and Methods) for producing and making available their model output. For CMIP, the US Department of Energy s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This research was supported by grants from the National Aeronautics and Space Administration and the National Oceanic and Atmospheric Administration Climate Program Office. B.J.S. was supported by the Korea Meteorological Administration Research and Development Program under Grant CATER 2012 2061. 1. Held IM, Soden BJ (2000) Water vapor feedback and global warming. Annu Rev Energy Environ 25:441 475. 2. Allan RP (2012) The role of water vapour in Earth s energy flows. Surv Geophys 33(3-4):557 564. 3. Soden BJ, Jackson DL, Ramaswamy V, Schwarzkopf MD, Huang X (2005) The radiative signature of upper tropospheric moistening. Science 310(5749):841 844. 4. Paltridge G, Arking A, Pook M (2009) Trends in middle- and upper-level tropospheric humidity from NCEP reanalysis data. Theor Appl Climatol 98:351 359. 5. Dessler AE, Davis SM (2010) Trends in tropospheric humidity from reanalysis systems. J Geophys Res 115(D19):D19127. 6. Sherwood SC, Roca R, Weckwerth TM, Andronova NG (2010) Tropospheric water vapor, convection, and climate. Rev Geophys 48(2):RG2001. 7. Bates JJ, Wu X, Jackson DL (1996) Interannual variability of upper-tropospheric water vapor band brightness temperature. J Clim 9(2):427 438. 8. Bates JJ, Jackson DL (2001) Trends in upper-tropospheric humidity. Geophys Res Lett 28(9):1695 1698. 9. Soden BJ, Bretherton FP (1993) Upper tropospheric relative humidity from the GOES 6.7 μm channel: Method and climatology for July 1987. J Geophys Res 98(D9): 16,669 16,688. 10. Colman RA, Hanson LI (2013) On atmospheric radiative feedbacks associated with climate variability and change. Clim Dyn 40(1-2):475 492. 11. Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5 and the experiment design. Bull Am Meteorol Soc 93(4):485 498. 12. Jackson DL, Soden BJ (2007) Detection and correction of diurnal sampling bias in HIRS/2 brightness temperatures. J Atmos Ocean Technol 24(8):1425 1438. 13. Shi L, Bates JJ (2011) Three decades of intersatellite-calibrated High-Resolution Infrared Radiation Sounder upper tropospheric water vapor. J Geophys Res 116(D4):D04108. 14. Tian B, et al. (2013) Evaluating CMIP5 models using AIRS tropospheric air temperature and specific humidity climatology. J Geophys Res Atmos 118(1):114 134. 15. Mears CA, Wentz FJ (2009) Construction of the Remote Sensing Systems V3.2 atmospheric temperature records from the MSU and AMSU microwave sounders. J Atmos Ocean Technol 26(6):1040 1056. 16. Spencer RW, Christy JR (1992) Precision and radiosonde validation of satellite gridpoint temperature anomalies. Part II: A tropospheric retrieval and trends during 1979-90. J Clim 5(8):858 866. 17. Mears CA, Wentz FJ (2009) Construction of the RSS V3.2 lower-tropospheric temperature dataset from the MSU and AMSU microwave sounders. J Atmos Ocean Technol 26(8):1493 1509. 18. Santer BD, et al. (2005) Amplification of surface temperature trends and variability in the tropical atmosphere. Science 309(5740):1551 1556. 19. Fu Q, Manabe S, Johanson CM (2011) On the warming in the tropical upper troposphere: Models versus observations. Geophys Res Lett 38(15):L15704. 20. Weatherhead EC, et al. (1998) Factors affecting the detection of trends: Statistical considerations and applications to environmental data. J Geophys Res 103(D14): 17,149 17,161. 21. Ammann CM, Meehl GA, Washington WM, Zender CS (2003) A monthly and latitudinally varying volcanic forcing dataset in simulations of 20th century climate. Geophys Res Lett 30(12):1657. 22. Meehl GA, et al. (2004) Combinations of natural and anthropogenic forcings in twentieth-century climate. J Clim 17(19):3721 3727. 23. Trenberth KE, Dai A (2007) Effects of Mount Pinatubo volcanic eruption on the hydrological cycle as an analog of geoengineering. Geophys Res Lett 34(15):L15702. 24. Soden BJ, et al. (2008) Quantifying climate feedbacks using radiative kernels. J Clim 21(14):3504 3520. 25. Shell KM, Kiehl JT, Shields CA (2008) Using the radiative kernel technique to calculate climate feedbacks in NCAR s Community Atmospheric Model. J Clim 21(10):2269 2282. 26. Vial J, Dufresne J-J, Bony S (2013) On the interpretation of inter-model spread in CMIP5 climate sensitivity estimates. Clim Dyn 41(11-12):3339 3362. 27. Dessler AE (2013) Observations of climate feedbacks over 2000 10 and comparisons to climate models. J Clim 26(1):333 342. 28. Loeb NG, et al. (2012) Observed changes in top-of-the-atmosphere radiation and upper-ocean heating consistent within uncertainty. Nat Geosci 5(2):110 113. 29. Meehl GA, et al. (2013) Externally forced and internally generated decadal climate variability associated with the Interdecadal Pacific Oscillation. J Clim 26(18): 7298 7310. 30. Hocking J, et al. (2011) RTTOV v10 Users Guide. NWP SAF (European Organisation for the Exploitation of Meteorological Satellites, Darmstadt, Germany). 6of6 www.pnas.org/cgi/doi/10.1073/pnas.1409659111 Chung et al.

Supporting Information Chung et al. 10.1073/pnas.1409659111 SI Materials and Methods Uncertainty in Linear Trends from Satellite Observations. The influences of time span on the computation of trends are examined for the satellite observations of High-Resolution Infrared Radiation Sounder (HIRS) channel 12 and microwave sounding unit (MSU) channel 2. Fig. S1 presents least-squares linear trends as a function of the time length denoted by the end year of a period staring from 1979. Vertical error bars represent ±2 SE of the estimated trend computed following ref. 1, indicating that the uncertainty in a least-squares linear fit diminishes with the increase in time span. The changes in linear trend with respect to time span are smaller for MSU channel 2 compared with HIRS channel 12. Vertical Profiles of Water-Vapor Changes and their Implication for the Water-Vapor Feedback. Water-vapor changes in the troposphere and their implication for the water-vapor feedback are examined for the historical and historicalnat scenarios using output from 12 models (BNU-ESM, CCSM4, CNRM-CM5, GFDL-CM3, GFDL-ESM2M, GISS-E2-R, HadGEM2-ES, IPSL-CM5A- LR, MIROC-ESM-CHEM, MIROC-ESM, MRI-CGCM3, and NorESM1-M; see Table S1 for information about climate models). Fig. S2 shows the absolute and fractional changes in specific humidity as a function of pressure over the period 1979 2005 with circles and horizontal error bars denoting multimodel mean and ±2 intermodel SD, respectively. Despite noticeable intermodel spread, the increase in water vapor is evident, particularly, in the lower troposphere for the historical scenario. Anthropogenic warming results in relatively smaller water-vapor increases in the upper troposphere, however fractional changes are greatest there. In contrast, fractional changes do not show the amplified upper-tropospheric moistening under natural forcing alone. Because the trapping of terrestrial radiation by water vapor is proportional to the logarithm of its concentration, the implication of the amplified upper-tropospheric moistening for the watervapor feedback was assessed by quantifying the feedback strength over two 10-y periods (i.e., 1979 1988 and 1989 1998) for three cases. First, the feedback strength was determined for the historical scenario-simulated water-vapor changes using a radiative kernel method (2 4). The feedback strength computations were repeated for the case that water vapor would change under natural forcing alone. In so doing, the water vapor profiles were modified using the trend determined from the historicalnat scenario. To quantify the part of the total water-vapor feedback related to the upper-tropospheric moistening, the feedback computations were also conducted for the modified water-vapor profiles but with retaining historical scenario-simulated watervapor changes in the upper troposphere. The feedback strengths for the three cases are compared in Fig. 4. 1. Weatherhead EC, et al. (1998) Factors affecting the detection of trends: Statistical considerations and applications to environmental data. J Geophys Res 103(D14): 17,149 17,161. 2. Soden BJ, et al. (2008) Quantifying climate feedbacks using radiative kernels. J Clim 21(14):3504 3520. 3. Shell KM, Kiehl JT, Shields CA (2008) Using the radiative kernel technique to calculate climate feedbacks in NCAR s Community Atmospheric Model. J Clim 21(10):2269 2282. 4. Vial J, Dufresne J-J, Bony S (2013) On the interpretation of inter-model spread in CMIP5 climate sensitivity estimates. Clim Dyn 41(11-12):3339 3362. Chung et al. www.pnas.org/cgi/content/short/1409659111 1of4

Fig. S1. Decadal trends of observed brightness temperatures as a function of time span for (A) HIRS channel 12 (T12), (B) MSU channel 2 (T2), and (C) MSU channel 2 HIRS channel 12 (T2 T12). Years specified on abscissa denote the end year of time period starting from 1979. Error bars denote ±2 SE of the linear trend. Chung et al. www.pnas.org/cgi/content/short/1409659111 2of4

Fig. S2. Changes in water vapor in the troposphere. (A) The absolute changes in specific humidity (unit: gram per kilogram per decade) as a function of pressure for the historical (red) and historicalnat (blue) scenarios over the period 1979 2005. (B) Same as in A, but for fractional changes in specific humidity (unit: percent). Circles and horizontal error bars denote multimodel mean and ±2 intermodel SD, respectively. Chung et al. www.pnas.org/cgi/content/short/1409659111 3of4

Table S1. CMIP5 climate models listed in the Supporting Information Modeling center Model name Model expanded name Beijing Normal University BNU-ESM Beijing Normal University Earth System Model National Center for Atmospheric Research CCSM4 Community Climate System Model, version 4 Centre National de Recherches Météorologiques (CNRM) CNRM-CM5 CNRM Coupled Global Climate Model, version 5 National Oceanic and Atmospheric Administration GFDL-CM3 GFDL Climate Model, version 3 (NOAA)/Geophysical Fluid Dynamics Laboratory (GFDL) NOAA/GFDL GDFL-ESM2M GDFL Earth System Model with Modular Ocean Model 4 component National Aeronautics and Space Administration Goddard GISS-E2-R GISS Model E2, coupled with the Russell ocean model Institute for Space Studies (GISS) Met Office Hadley Centre HadGEM2-ES Hadley Centre Global Environment Model, version 2-Earth System Institut Pierre-Simon Laplace (IPSL) IPSL-CM5A-LR IPSL Coupled Model, version 5, coupled with Nucleus for European Modelling of the Ocean, low resolution Japan Agency for Marine Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo); and National Institute for Environmental Studies Japan Agency for Marine Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo); and National Institute for Environmental Studies MIROC-ESM-CHEM MIROC-ESM Model for Interdisciplinary Research on Climate Earth System Model, atmospheric chemistry coupled version Model for Interdisciplinary Research on Climate Earth System Model Meteorological Research Institute (MRI) MRI-CGCM3 MRI Coupled Atmosphere-Ocean General Circulation Model, version 3 Norwegian Climate Centre NorESM1-M Norwegian Earth System Model, version 1 (mid resolution) Chung et al. www.pnas.org/cgi/content/short/1409659111 4of4