On the Surface Temperature Sensitivity of the Reflected Shortwave

Download PDF
1 downloads 0 Views 1MB Size Report
Comment
Oct 1, 2012 - The global-mean top-of-atmosphere incident solar radiation (ISR) minus the outgoing longwave radiation. (OLR) and the reflected shortwave ...
1 OCTOBER 2012

6585

AUMANN ET AL.

On the Surface Temperature Sensitivity of the Reflected Shortwave, Outgoing Longwave, and Net Incident Radiation HARTMUT H. AUMANN, ALEXANDER RUZMAIKIN, AND ALI BEHRANGI Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California (Manuscript received 15 October 2011, in final form 29 February 2012) ABSTRACT The global-mean top-of-atmosphere incident solar radiation (ISR) minus the outgoing longwave radiation (OLR) and the reflected shortwave radiation (RSW) is the net incident radiation (NET). This study analyzes the global-mean NET sensitivity to a change in the global-mean surface temperature by applying the interannual anomaly correlation technique to 9 yr of Atmospheric Infrared Sounder (AIRS) global measurements of RSW and OLR under cloudy and clear conditions. The study finds the observed sensitivity of NET that includes the effects of clouds to be 21.5 6 0.25 (1s) W m22 K21 and the clear NET sensitivity to be 22.0 6 0.2 (1s) W m22 K21, consistent with previous work using Earth Radiation Budget Experiment and Clouds and the Earth’s Radiant Energy System data. The cloud effect, 10.5 6 0.2 (1s) W m22 K21, is a positive component of the NET sensitivity. The similarity of the NET sensitivities derived from forced and unforced models invites a comparison between the observed sensitivities and the effective sensitivities calculated for the Fourth Assessment Report models, although this requires some caution: The effective model sensitivities with clouds range from 20.88 to 21.64 W m22 K21, the clear NET sensitivity in the models ranges from 22.32 to 21.73 W m22 K21, and the cloud forcing sensitivities range from 10.14 to 11.18 W m22 K21. The effective NET and clear NET sensitivities derived from the models are statistically consistent with those derived from the AIRS data, considering the observational and model derivation uncertainties.

1. Introduction Changes in the global climate are reflected in changes in the global-mean net incident radiation: NET 5 ISR 2 (RSW 1 OLR), where ISR is the incident solar radiation at the top of the atmosphere (TOA), RSW is the reflected shortwave radiation, and OLR is the outgoing longwave radiation at the TOA. The nominal day/night global averages of RSW 5 107 W m22, OLR 5 235 W m22, and ISR 5 342 W m22 were closely balanced to NET 5 0 (Kiehl and Trenberth 1997). However the global balance is not static: 1) the ISR varies seasonally due to the eccentricity of Earth’s orbit, 2) the RSW changes due to changes in cloud cover and surface albedo; and 3) the OLR is influenced by the presence of clouds. The increase in the greenhouse gases creates long-term changes in the OLR and possibly in the RSW. The question of

Corresponding author address: Hartmut H. Aumann, Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Dr., Pasadena, CA 91109. E-mail: [email protected] DOI: 10.1175/JCLI-D-11-00607.1 Ó 2012 American Meteorological Society

how the earth climate system responds to these longterm changes initiated an extensive research effort that included numerical modeling and analyses of observational data [see Bony et al. (2006) for a comprehensive review]. Here, we focus on the observational part of the problem and evaluate a global-mean NET sensitivity using leff 5

DNET DTs

(1)

in units of watts per meter per kelvin, where Ts is the surface temperature. If the earth were a 255-K blackbody, that is, the influence of its atmosphere and any shortwave reflected component were ignored, its NET sensitivity would be the derivative of the Planck function, lPlanck 5 23.6 W m22 K21. A negative NET sensitivity, that is, when the NET decreases as the surface temperature increases, implies a stable system. The OLR and the RSW are influenced by the presence of clouds. The paper by Schneider (1972) was among the earliest publications pointing out the effect of clouds on

6586

JOURNAL OF CLIMATE

the NET. Clouds and their effects were discussed in terms of the ‘‘thermostat hypothesis’’ (Ramanathan and Collins 1991), ‘‘self-regulation’’ (Waliser and Graham 1993), the ‘‘iris hypothesis’’ (Lindzen et al. 2001; Hartmann and Michelsen 2002; Lindzen and Choi 2009; Dessler 2010; Murphy 2010; Dessler 2011, and references therein). However, clouds are still identified as a key uncertainty in climate models (Solomon et al. 2007). Here, we define the cloud effect as the sensitivity of the NET radiation with clouds minus the NET sensitivity if the clouds are removed without otherwise changing the state of the atmosphere or the surface, that is, we replace (RSW 1 OLR) with [(RSW 2 ClearRSW) 1 (OLR 2 ClearOLR)]. This approach is known as the ‘‘cloud forcing’’ analysis, which was introduced by Cess et al. (1990), who defined the cloud forcing as the difference between the all-sky and clear-sky radiation at TOA. The advantage of this approach is that the cloud forcing is directly observable and hence it may be used for validating climate models. The weakness of this approach is that the changes in the cloud forcing cannot be directly interpreted as being caused solely by the clouds or caused by changes in the other variables (temperature, lapse rate, water vapor, and albedo) that occur under the transformation from clear to cloudy conditions (Bony et al. 2006). To infer the feedbacks produced by each variable, one needs to calculate the radiative kernels (partial derivative of the radiation relative to a specific variable) and the change of each variable relative to the surface temperature (Soden et al. 2008). However, the assumptions of linearity and separability of the different feedbacks used in these calculations have been the subject of debate (Aires and Rossow 2003), and the results are sensitive to compensating errors. The changes in different variables are calculated in model runs with some prescribed forcing, usually due to the doubling of CO2. The effect of clouds in the model runs can in principle be inferred from evaluating the NET by calculating the OLR and RSW without clouds, or by backing the effect of the clouds out of the NET sensitivity by subtracting the sum of the model sensitivities calculated with kernels, as done by Soden and Held (2006) and Dessler (2010). Up to now the observational evaluation NET sensitivity has been limited to the analysis of the seasonal and interannual variability of the OLR and RSW from Earth Radiation Budget Experiment (ERBE) and Clouds and the Earth’s Radiant Energy System (CERES) data. Tsushima and Manabe (2001) and Tsushima et al. (2005) used the seasonal variability of Ts. Murphy et al. (2009) analyzed the interannual variability of the NET and found a sensitivity of 21.25 6 0.5 W m22 K21 based on the analysis of ERBE and CERES data between 1985

VOLUME 25

and 2004. Chung et al. (2010a, hereafter CSS2010) used ERBE data from 1985 through 1999, and Dessler (2010) deduced a cloud effect based on the interannual variability of 2000–10 CERES data. Here, we use data from the Atmospheric Infrared Sounder (AIRS; Aumann et al. 2003) to determine RSW and OLR, and their clear counterparts to evaluate the NET sensitivity defined by Eq. (1).

2. Data AIRS is a hyperspectral infrared sounder aboard the Earth Observing System (EOS) Aqua spacecraft, which was launched into a polar sun-synchronous orbit at 705-km altitude in May 2002. The ascending node of the orbit has been maintained within a minute of 1330 UTC. AIRS generates 4 million spectra each day by scanning 6498 cross track with 1.18 (13.5 km at nadir) diameter field of view (FOV). Of these data about 10%, 300 000 spectra each day, are within 38 of nadir. About 2% of the near-nadir spectra, typically 6500, are selected randomly each day. To make the sample ‘‘area representative,’’ the random nadir data from the high-latitude areas are thinned to eliminate the high-latitude spatial overcoverage from the EOS Aqua polar orbit. Typically 50 000 spectra are identified each day as ‘‘clear,’’ based on a number of spatial and spectral tests (Aumann et al. 2006). The daily collection of random nadir spectra, clear spectra, data from deep convective clouds (DCC), and data from a number of special sites used for monitoring the calibration are saved in the AIRS Calibration Data Subset (ACDS). The AIRS data are exceptionally well calibrated and extremely stable (Aumann et al. 2006). Half of the daily samples come from the 0130 LT overpasses, referred to as ‘‘night,’’ and the other half of samples comes from the 1330 overpasses, referred to as ‘‘day.’’ Each spectrum saved in the ACDS is associated with the surface temperature, a land fraction, an infrared clear flag, and reflected light measurements. The surface temperature is obtained from the National Oceanic and Atmospheric Administration (NOAA) Global Forecast System (Iredell and Caplan 1997). The surface temperatures from the 3-, 6-, and 9-h forecast, made daily at 0000, 6000, 1200, 1800 UTC, respectively, on a ½8 grid, are interpolated in space and time to match the AIRS sample times and positions. AIRS also measures the reflected shortwave radiation in three channels in the 8 3 9 pixel array associated with each IR FOV (Gautier et al. 2003): channel 1 (0.40–0.44 mm), channel 2 (0.58– 0.68 mm), and channel 3 (0.75–0.95 mm). The mean of the 8 3 9 pixel array associated with each IR FOV is saved in the ACDS. Since our data are selected from within 38 of

1 OCTOBER 2012

6587

AUMANN ET AL.

nadir, we can ignore slant path effects and glint contamination. Significant surface glinting in the reflected light channels is first apparent at more than 108 off nadir. The random nadir data are used to calculate the ‘‘all sky’’ RSW, OLR, and their ‘‘clear sky’’ counterparts associated with each FOV.

known Sun–Earth distance. We then evaluate the seasonal variation for NET(i) 5 ISR(i) 2 RSW(i) 2 OLR(i), the ClearNET(i) 5 ISR(i) – ClearRSW(i) 2 ClearOLR(i), and the cloud effect CE(i) 5 NET(i) 2 ClearNET(i), as well as Ts(i) by least squares (LSQ) fitting the data to a harmonic time series using

a. RSW and ClearRSW Unlike CERES (Wielicki et al. 1996), which measures the RSW in a broadband 0.1–5-mm channel, AIRS measures the shortwave reflected flux density in the three much narrower channels between 0.4 and 1.1 mm described above. These measurements have to be converted to a spectrally integrated RSW. For this purpose we regressed the AIRS data against the CERES Aqua RSW global 18 3 182 gridded product from September 2002. Referring to the individual measurements of AIRS reflected light channel 2 as vis2, we have to first order RSW 5 Avis2, where A 5 1.04 6 0.02 mm is an effective bandwidth. CERES measures the reflected shortwave radiation under clear conditions using a cloud mask. The AIRS ClearRSW can be inferred from the vis2 associated with footprints identified each day as cloud free. The conversion from vis2 to ClearRSW uses A 5 1.04 6 0.04 mm. Details on the regression training are given in Appendix A.

b. OLR and ClearOLR The OLR and ClearOLR are available from the AIRS level 2 (L2) standard products (Susskind et al. 2003), where a high-quality temperature and water vapor profile, surface emissivity, and surface temperature retrieval are available. The OLR and the ClearOLR for each footprint are derived here directly from AIRS L1b spectral radiances using regression trained on L2 version 6 (V6) OLR and ClearOLR. This provides an OLR and the ClearOLR even under extremely cloudy conditions. Details on the regression training are given in Appendix B. The uncertainty in the derivation of ClearOLR and the OLR are accounted for in the error analysis.

3. Data analysis procedure For our analysis we use the AIRS data from 3240 days between 1 September 2002 and 31 August 2011. The 21 million global random nadir spectra from these days are saved in the ACDS, about 6500 each day. We calculate the time series of the global-mean surface temperature Ts(i), RSW(i), OLR(i), ClearRSW(i), and ClearOLR(i) from these 6500 daily samples, where the i index associated with time t(i) runs from 1 to 3240. The global-mean ISR(i) at the TOA is calculated for each day based on the

4

Y(i) 5 ao 1

å a(k) sin[2kpt(i)] 1 b(k) cos[2kpt(i)],

k51

(2) where t(i) is the sample date. Then ATs 5 Ts(i) 2 Ys(i), ANET(i) 5 NET(i) 2 Ynet(i), and AClearNET(i) 5 ClearNET(i) 2 Yclrnet(i) define the anomaly time series. The lowest frequency fitted in Eq. (2) is the annual (seasonal, k 5 1) mode. This assumes that the seasonal variation seen in 9 yr can be considered as the ‘‘normal’’ seasonal variation, that is, the anomaly represents longer-term (lower frequency) interannual effects.

4. Results Figure 1 shows the daily global-mean Ts(i) (solid) and NET(i) (dashed), smoothed for the graphic presentation only with a 2-month sliding mean, from the 9 yr of data. It is interesting to note the dominance of the Northern Hemisphere: The seasonal variation of Ts peaks in July, while the peak of the NET is in December. The seasonal variation seen in these data is removed in the anomalies. An overlay of the anomalies of the NET and Ts derived from the data presented in Fig. 1 is shown in Fig. 2, also smoothed with a 2-month sliding mean. Note that the smoothing filter was applied only for plotting of the time series. All data processing uses the unfiltered daily data. The effective NET sensitivity is derived from the analysis of the scatter diagram of the daily anomalies, ATs (i) and ANET(i), as shown in Fig. 3. The correlation between the daily values of the ANET and ATs is 20.19. The slope of the linear regression line, 21.43 W m22 K21 with a one-sigma uncertainty of 0.13 W m22 K21, is determined using the linear regression. Using the same method the ClearNET sensitivity, shown in Fig. 4, is 22.0 6 0.1 (1s) W m22 K21. We calculate the cloud forcing sensitivity from the scatter diagram of the (NETClearNET) anomaly and the surface temperature anomaly. The slope of the linear regression line yields a cloud forcing sensitivity of 10.5 6 0.15 (1s) W m22 K21.

5. Discussion The sensitivities and their one-sigma uncertainties quoted above are estimated from linear regression

6588

JOURNAL OF CLIMATE

VOLUME 25

FIG. 1. Global Ts (solid) and global NET (dotted) from 9 yr of AIRS data.

assuming Gaussian noise. There are other factors that need to be considered in assessing the total uncertainty. 1) The uncertainty of the linear regression line of the scatter diagram refers to the standard deviation of the least squares linear fit. To better evaluate the uncertainty of the fit, we apply the bootstrap method (Chernick 1999). In this method the scatterplot is recreated using the original ATs(i) vector, but new ANET(i) vectors are created by randomly permuting the entries in the original ANET(i) array. This procedure destroys the correlation with the ATs(i) vector. As a result, the slope of the scatter diagram varies randomly with zero mean and the standard deviation of the slope values gives an estimate of the slope uncertainty. Using 100 random permutation runs yield a bootstrap slope uncertainty of 0.2 W m22 K21 (1s). For the (NET 2 ClearNET) and cloud sensitivity, the

slope uncertainty is 0.15 W m22 K21 (1s). We accepted the larger uncertainty estimates generated by the bootstrap method. 2) The numerical technique used to create the anomaly time series contributes to the uncertainty of the result. To evaluate this effect, we used empirical mode decomposition (EMD; Huang et al. 1998; Huang and Wu 2008). In this approach the anomaly is defined as the data time series minus the annual EMD mode. We treat the difference between the sensitivity calculated using anomaly calculated from Eq. (2) and the sensitivity calculated using the EMDderived anomaly, 0.05 W m22 K21, as a component of the overall uncertainty. 3) The length of the time series used to define the ‘‘normal’’ seasonal variability introduces an uncertainty in the anomaly. We evaluated the magnitude of this effect by recalculating the anomaly sensitivity

FIG. 2. Overlay of the anomalies ATs (solid) and ANET (dotted).

1 OCTOBER 2012

AUMANN ET AL.

FIG. 3. The anomaly scatter diagram of ANET vs ATs.

using the first 8 and the last 8 of the 9 yr of data. Withholding the last year or the first year changes the NET sensitivity by 60.04 W m22 K21. 4) The uncertainty in the derivations of the RSW and OLR and their clear counterparts contribute to the NET sensitivity uncertainty. Since the anomaly removes the bias, bias has no effect on the NET sensitivity. These uncertainties were modeled as multiplicative errors and as the difference between the results using multivariate regression and a simple single parameter fit as discussed in the appendices. Propagated through the anomaly and sensitivity calculations, the RSW and OLR derivation uncertainties contribute less than 0.05 W m22 K21 to the NET and ClearNET sensitivity slope uncertainty. The RSW and ClearRSW vary seasonally about global means of 100 and 40 W m22, respectively. As a somewhat extreme test of RSW and ClearRSW uncertainty, we assume that both are constant, that is, their anomalies are zero. Then the NET and ClearNET anomaly sensitivities, now calculated from the OLR and the ClearOLR anomalies alone, change by 10.1 W m22 K21. 5) An uncertainty in Ts may contribute to the Ts anomaly. The National Centers for Environmental Prediction (NCEP) Ts agrees globally with the AIRS L2 V6 surface skin temperature with a cold bias of 0.4 K, with a standard deviation of 2.3 K, where a high-quality surface solution is obtained with the AIRS L2 (about 50% of all cases in L2 V6). Assuming that the standard deviation of the difference between

6589

FIG. 4. Scatterplot of the ClearNET vs Ts anomalies.

AIRS L2 and NCEP Ts can be used as a conservative metric of the uncertainty of the NCEP Ts, the Ts uncertainty contributes less than 0.01 W m22 K21 to the sensitivity uncertainty. Based on this analysis, we use in the following discussion leff 5 21.5 6 0.25 (1s) W m22 K21 as the NET sensitivity, leff_clear 5 22.0 6 0.2 (1s) for the ClearNET sensitivity, and 10.5 6 0.2 (1s) for the cloud sensitivity. A simple theoretical evaluation of Eq. (1) can be made using an OLR that assumes that the earth radiates as a blackbody, that is, ignore the influence of its atmosphere and any shortwave reflected component. Then NET 5 ISR 2 OLR. The global-mean OLR of 238 W m22 (appendix B) corresponds to a blackbody at 255 K, and the derivative of the Planck function gives lplanck 5 23.6 W m22 K21. A more careful evaluation in 12 Fourth Assessment Report (AR4) models (Soden and Held 2006, their Table 1) produced an average value of lplanck 5 23.2 W m22 K21. Chung et al. (2010b, hereafter CYS2010), derived a global average value of 12.0 W m22 K21 from the response to a uniform warming and constant relative humidity (RH), but since they refer to the Planck radiative damping 13.6 W m22 K21, we infer that lconstant-RH 5 22.0 W m22 K21. Work on the observational evaluation of Eq. (1) has heretofore been based on ERBE and CERES data. The negative value of Eq. (1), 1D(OLR 1 RSW)/DSST, is used in some papers. Thus, CSS2010 analyzed the NET sensitivity of the tropical oceans using ERBE data from

6590

JOURNAL OF CLIMATE

1985 through 1997 and found lCSS between 22.5 and 23 W m22 K21 [note that their Fig. 1d shows 2D(OLR 1 RSW)/DSST]. However, it is difficult to compare tropical ocean values to global values. Our observed global NET sensitivity leff 5 21.5 6 0.25 (1s) W m22 K21 agrees with the sensitivity of 21.25 6 0.5 W m22 K21 deduced by Murphy et al. (2009) from a totally independent dataset, ERBE and CERES data between 1985 and 2004, but it is closer to the value expected for uniform global warming with constant RH, 22 W m22 K21 (CYS2010). The analysis of regional, seasonal, and interannual variability of 1985–88 ERBE data (CYS2010, their Fig. 2) yielded ClearOLR sensitivities between 12.2 and 12.4 W m22 K21. In terms of ClearNET sensitivity, this corresponds to 22.2 to 22.4 W m22 K21, consistent with our observed global ClearNET sensitivity, leff_clear 5 22.0 6 0.20 (1s). Our leff_cloud 5 10.5 6 0.2 (1s) W m22 K21 agrees within the error bars with 10.54 6 0.36 (1s) W m22 K21 deduced by Dessler (2010), who used the anomaly correlation with a totally independent dataset, 120 monthly means from 2000 to 2010 CERES Terra data, and a different numerical method. Our observations indicate that the cloud component of the NET sensitivity is positive with more than 2-s probability. Our sensitivities were derived from only 9 yr of data, that is, the CO2 forcing and changes in the water vapor were small. To make a comparison of our results with GCM results, we have to distinguish between unforced (control) runs and runs forced by doubling the CO2 abundance followed by model equilibrium. Our observed sensitivities should by comparable to control runs. However, the sensitivities derived from long-term forced GCM results and unforced model runs are similar. This is supported by CSS2010, their Fig. 3c, which shows the OLR1RSW sensitivities of a collection of forced runs and preindustrial control runs. The median forced run sensitivity is about 11.4 W m22 K21, while the median control sensitivity is about 11.0 W m22 K21, but the scatter in the two datasets overlaps. The similarity in the results of forced and unforced runs is also seen in the cloud feedback. Thus, Dessler (2010), who analyzed the cloud sensitivity for 100 yr of control runs, has found a cloud feedback range between 0.34 and 1.11 W m22 K21, which is similar to the cloud sensitivity range of the forced AR4 models (Soden and Held 2006) with an average 10.69 W m22 K21 and a range from 0.15 to 1.18 W m22 K21. The AR4 results for the Planck, the lapse rate, the water vapor, and the surface albedo sensitivities (the sum of which gives the sensitivity without clouds) and the cloud effect were summarized in Table 1 in Soden and Held (2006). The effective sensitivities in the AR4 models with clouds, which were determined with an estimated uncertainty

VOLUME 25

of 0.2 (1s) W m22 K21, range from 20.88 to 21.64 W m22 K21 compared to our leff 5 21.5 6 0.25 (1s) W m22 K21. The clear NET sensitivity in the models ranges from 22.32 to 21.73 W m22 K21, compared to our observed 22.0 6 0.2 (1s) W m22 K21. The cloud sensitivity in the models ranges from 10.14 to 11.18 W m22 K21 compared to our 10.5 6 0.2 (1s) W m22 K21 cloud forcing sensitivity. Hence, the observed effective NET and clear NET sensitivities are statistically consistent with the model sensitivities, considering the observational and model derivation uncertainties. However, the global-mean NET-Ts relationship differs across different time scales and differs between transient and equilibrium conditions. Using the sensitivities derived here from the AIRS data as a metric for the quantitative evaluation of the AR4 models is therefore not unproblematic.

6. Summary and conclusions The sensitivity of the reflected shortwave and outgoing longwave radiation under clear and cloudy conditions to a change in the global-mean surface temperature is evaluated from 9 yr of AIRS data using the anomaly correlation technique. The observed globalmean temperature sensitivity of the NET, including the effects of clouds, is 21.5 6 0.25 (1s) W m22 K21 compared to 22 W m22 K21 expected from the uniform warming with a constant RH. The cloud forcing effect, 10.5 6 0.2 (1s) W m22 K21, is a positive component of the NET sensitivity, consistent with previous work using ERBE and CERES data. The similarity of the NET sensitivities derived for forced and unforced models makes the comparison between our observation and AR4 models illuminating. The effective NET and clear NET sensitivities derived from the AIRS data are statistically consistent with the model sensitivities. However, because the global-mean NET-Ts relationship differs across different time scales and differs between transient and equilibrium conditions, the use of the sensitivities derived here from the AIRS data as a metric for the quantitative evaluation of the AR4 models requires some caution. Acknowledgments. The research described in this paper was carried out at the Jet Propulsion Laboratory at the California Institute of Technology under a contract with the National Aeronautics and Space Administration. We are grateful for the long-term support of Dr. Ramesh Kakar, Aqua Program Scientist at NASA HQ. We acknowledge the discussions with Dr. Joao Teixeira at JPL and the helpful comments by two anonymous reviewers.

1 OCTOBER 2012

6591

AUMANN ET AL.

APPENDIX A Derivation of the All-Sky and Clear-Sky RSW from AIRS Data AIRS measures the reflected shortwave radiation in three channels using a 8 3 9 pixel array associated with each 13.5 IR FOV (Gautier et al. 2003): channel 1 (0.40–0.44 mm), channel 2 (0.58–0.68 mm), and channel 3 (0.75–0.95 mm). The mean of the 72 pixels associated with each AIRS FOV is saved in the ACDS as vis1–vis3, respectively. The three channels exhibit small decreases in signal with time. For the vis2 and vis3 channels, this requires a linear correction of 10.40 6 0.01% yr21 related to differential scan mirror contamination. For presently unknown reasons, the vis1 channel requires a nonlinear correction. These corrections were worked using the observations of DCC (Aumann et al. 2007), with the assumption that DCC act as nearly perfect diffusers of sunlight. We used the CERES_SSF1deg-Month-lite-Aqua_Ed2.5 SW_all to derive an approximation to the all sky RSW (CERES SW) using AIRS data. The CERES product is available as monthly averages on a global 360 3 180 grid. We converted AIRS vis2 measurements from September 2002, March 2003, and September 2009 into 3 monthly 360 3 180 gridded maps, each of which was then collapsed into (90) 18-wide latitude bands. We used monthly-mean CERES data to create the equivalent 18-wide latitude bands for the same month. Figure A1 is an overlay of the AIRS vis2 and CERES SW latitude dependence for the month of September 2002, which was used as the regression training set. Figure A1 shows that the AIRS vis2 and the CERES SW are highly correlated (r 5 0.98). Note the left-to-right latitude asymmetry and the spike at 58N latitude, which is apparently caused by the ITCZ. For the September 2002 training set, RSW 5 1.045vis2 with zero bias relative to CERES by definition. The overlay of CERES and AIRS from the two independent 1-month datasets (March 2003 and September 2009) looks almost identical to Fig. A1, but it shows a bias of 12 and 22 W m22 relative to the training set. The 9-yr global-mean RSW of 100 W m22 is consistent with the global-mean RSW quoted for CERES in CERES (2011). We also evaluated the RSW based on multivariate regression with vis1–vis3 in support of the sensitivity uncertainty analysis. To avoid the uncertainty introduced by a nonlinear vis1 trend correction, we used RSW 5 (1.04 6 0.02)vis2 for the analysis presented in this paper. The 9-yr global-mean RSW is 100 with a 14 W m22 peak-to-peak seasonal variation. During the past 9 yr, the RSW has changed by less than 0.07% yr21 (2-s upper limit).

FIG. A1. Overlay of the AIRS vis2 and CERES SW for September 2002.

We also used the vis2 data to create the time series of clear-sky RSW. The ACDS data include a ‘‘clear’’ flag, which is based on a number of infrared tests. The combination of the clear flag, the land fraction, and Ts allowed us to generate three daily average clear-sky RSW time series, for nonfrozen ocean clear, nonfrozen land clear, and frozen clear. The boundary between frozen and nonfrozen is set at Ts 5 273 K. The three time series are combined by weighting the daily mean values from the three components by their area fractions, 0.64, 0.23 and 0.13, respectively. This approximates an area representative clear-sky RSW dataset. During the past 9 yr, the global-mean clear RSW has changed by less than 0.16% yr21 (2-s upper limit) in the mean of 40 W m22. The parameter used to divide clear frozen and nonfrozen surfaces introduces an uncertainty. If Ts 5 271 K, more appropriate for the transition between sea ice and frozen ocean, then the global-area fractions would be 0.65, 0.23, and 0.12, respectively, and the mean clear RSW would decrease by 3%. This redefinition of the frozen surface changes the ClearNET sensitivity by less than 0.001 W m22 K21. We use ClearRSW 5 (1.04 6 0.04)vis2 for the analysis presented in this paper.

APPENDIX B Derivation of the OLR and ClearOLR from AIRS Data The OLR and the ClearOLR are standard L2 products (Susskind et al. 2003). The OLR, with and without

6592

JOURNAL OF CLIMATE

VOLUME 25

FIG. B1. AIRS L2 V6 OLR as function of bt790 for the training set.

FIG. B2. AIRS L2 V6 ClearOLR version Ts for the training set.

clouds, is calculated from the retrieved Tsurf, «surf, T(p), q(p), and cloud height and fraction. For the AIRS L2 V6, this calculation uses a radiative transfer algorithm provided by Iacono et al. (2008). This calculation is accurate only if the retrieval is of high quality. In the soonto-be-released L2 V6, globally about 50% of retrievals produce a Clear OLR identified as high quality; 84% produce a high-quality OLR. Since we need the OLR and Clear OLR at every footprint, we derive them directly from AIRS L1b data (calibrated radiances), which are available for every footprint independent of the state of cloudiness. The derivation uses regression with global data from 15 days, randomly selected from 9 yr of AIRS data, which were specially processed as part of the L2 V6 OLR and ClearOLR validation. The regression coefficients were derived from 4 days; the uncertainty was derived from the remaining 11 (independent) days.

midtroposphere CO2 sensitive channel at 723 cm21), and bt1419 (an upper-tropospheric water-sensitive channel at 1419 cm21) to fit the OLR. This improves the standard deviation of the regression fit to 5.0 W m22. The fit of the independent set has 5.1 W m22 standard deviation, and the rms bias for the 11 independent days is 0.15 W m22. The 9-yr trend in the daily global-mean OLR calculated in this fashion is less than 0.03 W m22 yr21 (2s) in the mean of 241 W m22. The mean agrees within the nominal uncertainty, with the global mean of 238 W m22 quoted for CERES in CERES (2011). We carried the effect of the OLR uncertainty through the NET sensitivity calculations in two ways: 1) as a multiplicative 1% and 2) by comparing the results using the multivariate regression with a quadratic least squares fit using only bt790.

b. Clear OLR a. OLR Figure B1 shows the scatter diagram of all near-nadir L2 V6 OLR points from the training days as function of bt790, where bt790 is the observed brightness temperature at 790 cm21. This channel is at the longwave end of the 11-mm atmospheric window area, but it has considerable water vapor absorption. The quadratic regression line (shown superimposed in Fig. B1) has a correlation of 0.99 with the AIRS L2 OLR and fits with 7 W m22 standard deviation, corresponding to 3% of the mean. Since a single parameter, bt790, may not capture the full variability of the OLR, we used bt790, bt723 (a

Figure B2 shows the scatter diagram of the all AIRS L2 V6 ClearOLR points in the training set as function of Ts from NCEP. The dependence of ClearOLR on Ts is almost linear, with a correlation of 0.95. The multivariate regression using Ts, bt723, and bt1419 fits the ClearOLR with a standard deviation of 10 W m22 (in the mean of 270 W m22). The rms bias for the 11 independent days was 0.3 W m22. We carried the effect of the ClearOLR uncertainty through the ClearNET sensitivity calculations two ways: 1) as a multiplicative 1% error and 2) by comparing the results using the multivariate regression with a linear least squares fit using only Ts.

1 OCTOBER 2012

AUMANN ET AL. REFERENCES

Aires, F., and W. B. Rossow, 2003: Inferring instantaneous, multivariate and nonlinear sensitivities for the analysis of feedback processes in a dynamical system: Lorenz model case study. Quart. J. Roy. Meteor. Soc., 129, 239–275. Aumann, H. H., and Coauthors, 2003: AIRS/AMSU/HSB on the Aqua mission: Design, science objectives, data products, and processing systems. IEEE Trans. Geosci. Remote Sens., 41, 253–264. ——, S. Broberg, D. Elliott, S. Gaiser, and D. Gregorich, 2006: Three years of Atmospheric Infrared Sounder radiometric calibration validation using sea surface temperatures. J. Geophys. Res., 111, D16S90, doi:10.1029/2005JD006822. ——, T. Pagano, and M. Hofstaedter, 2007: Observations of deep convective clouds as stable reflected light standard for climate research: AIRS evaluation. Atmospheric and Environmental Remote Sensing Data Processing and Utilization III: Readiness for GEOSS, M. D. Goldberg et al., Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 6684), 668410, doi:10.1117/12.734599. Bony, S., and Coauthors, 2006: How well do we understand and evaluate climate change feedback processes? J. Climate, 19, 3445–3482. CERES, 2011: CERES_EBAF_Ed2.6: Data quality summary. CERES, 33 pp. [Available online at http://ceres.larc.nasa.gov/ documents/DQ_summaries/CERES_EBAF_Ed2.6_DQS.pdf.] Cess, R. D., and Coauthors, 1990: Intercomparison and interpretation of climate feedback processes in 19 atmospheric general circulation models. J. Geophys. Res., 95 (D10), 16 601–16 615. Chernick, M. R., 1999: Bootstrap Methods: A Practitioner’s Guide. Wiley Series in Probability and Statistics: Applied Probability and Statistics, Vol. 353, Wiley, 264 pp. Chung, E.-S., B. J. Soden, and B.-J. Sohn, 2010a: Revisiting the determination of climate sensitivity from relationships between surface temperature and radiative fluxes. Geophys. Res. Lett., 37, L10703, doi:10.1029/2010GL043051. ——, D. Yeomans, and B. J. Soden, 2010b: An assessment of climate feedback processes using satellite observations of clear-sky OLR. Geophys. Res. Lett., 37, L02702, doi:10.1029/ 2009GL041889. Dessler, A. E., 2010: A determination of the cloud feedback from climate variations over the past decade. Science, 330, 1523– 1527. ——, 2011: Cloud variations and the earth’s energy budget. Geophys. Res. Lett., 38, L19701, doi:10.1029/2011GL049236. Gautier, C., Y. Shiren, and M. D. Hofstader, 2003: AIRS/Vis near IR instrument. IEEE Trans. Geosci. Remote Sens., 41, 330–341. Hartmann, D. L., and M. L. Michelsen, 2002: No evidence for iris. Bull. Amer. Meteor. Soc., 83, 249–254. Huang, N. E., and Z. Wu, 2008: A review on Hilbert-Huang transform: Method and its applications to geophysical studies. Rev. Geophys., 46, RG2006, doi:10.1029/2007RG000228. ——, and Coauthors, 1998: The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc. Roy. Soc. London, 454, 903–995.

6593

Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/ 2008JD009944. Iredell, M., and P. Caplan, 1997: Four-times-daily runs of the AVN model. National Weather Service, Office of Meteorology Tech. Procedures Bull. 442, 3 pp. [Available online at http:// www.nws.noaa.gov/om/tpb/442.htm.] Kiehl, J. T., and K. E. Trenberth, 1997: Earth’s annual global mean energy budget. Bull. Amer. Meteor. Soc., 78, 197–208. Lindzen, R. S., and Y.-S. Choi, 2009: On the determination of climate feedbacks from ERBE data. Geophys. Res. Lett., 36, L16705, doi:10.1029/2009GL039628. ——, M. D. Chou, and A. Y. Hou, 2001: Does the earth have an adaptive infrared iris? Bull. Amer. Meteor. Soc., 82, 417–432. Murphy, D. M., 2010: Constraining climate sensitivity with linear fits to outgoing radiation. Geophys. Res. Lett., 37, L09704, doi:10.1029/2010GL042911. ——, S. Solomon, R. W. Portmann, K. H. Rosenlof, P. M. Forster, and T. Wong, 2009: An observationally based energy balance for the earth since 1950. J. Geophys. Res., 114, D17107, doi:10.1029/2009JD012105. Ramanathan, V., and W. Collins, 1991: Thermodynamic regulation of ocean warming by cirrus clouds deduced from observations of the 1987 El Nin˜o. Nature, 351, 27–32. Schneider, S. H., 1972: Cloudiness as a global climatic feedback mechanism: The effects on radiation balance and surface temperature of variations in cloudiness. J. Atmos. Sci., 29, 1413–1422. Soden, B. J., and I. M. Held, 2006: An assessment of climate feedbacks in coupled ocean–atmosphere models. J. Climate, 19, 3354–3369. ——, ——, R. Colman, K. M. Shell, J. T. Kiehl, and C. A. Shield, 2008: Quantifying climate feedbacks using radiative kernels. J. Climate, 21, 3504–3520. Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M. B. Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate Change: The Physical Science Basis, Cambridge University Press, 996 pp. Susskind, J., C. D. Barnett, and J. Blaisdell, 2003: Retrieval of atmospheric and surface parameters from AIRS/AMSU/HSB data in the presence of clouds. IEEE Trans. Geosci. Remote Sens., 41, 390–409. Tsushima, Y., and S. Manabe, 2001: Influence of cloud feedback on annual variation of global mean surface temperature. J. Geophys. Res., 106 (D19), 22 635–22 646. ——, A. Abe-Ouchi, and S. Manabe, 2005: Radiative damping of annual variation in global mean surface temperature: Comparison between observed and simulated feedback. Geophys. Res. Lett., 24, 591–597, doi:10.1007/s00382-005-0002y. Waliser, D. E., and N. E. Graham, 1993: Convective cloud systems and warm-pool sea surface temperatures: Coupled interactions and self-regulation. J. Geophys. Res., 98 (D7), 12 881–12 893. Wielicki, B. A., B. R. Barkstrom, E. F. Harrison, R. B. Lee III, G. L. Smith, and J. E. Cooper, 1996: Clouds and the Earth’s Radiant Energy System (CERES): An Earth Observing System experiment. Bull. Amer. Meteor. Soc., 77, 853–868.