1. Introduction
Differences in the simulated response of clouds to a warming climate are a significant driver of the wide spread of the equilibrium climate sensitivity (ECS) exhibited by climate models, defined as the global surface temperature response to the doubling of atmospheric CO2 (relative to the preindustrial amount) once the net radiation balance at the top of atmosphere is zero. The most recent phase 6 of the Climate Model Intercomparison Project (CMIP6) ensemble of participating climate models has ECS values ranging from 1.5° to 6°C (Zelinka et al. 2020; Meehl et al. 2020; Dong et al. 2020). The Intergovernmental Panel on Climate Change reports ECS based on multiple peer-reviewed studies using observations, climate models, and paleoclimate evidence. For AR5 (IPCC 2013), the estimates of ECS were 1.5°–4.0°C; the recent AR6 report (IPCC 2023) narrows the range to 2.5°–4.0°C. A recent review paper (Sherwood et al. 2020) reported a 66% likelihood range for ECS of 2.6°–3.9°C based on three lines of evidence: our understanding of feedback processes, the historical climate record, and the paleoclimate record. They concluded that an ECS value lower than 2°C is difficult to reconcile with any of the three lines of evidence.
Simulated cloud trends and feedback in climate models are evaluated using several observational cloud datasets. The longest well-calibrated cloud record (23 years) is from the Clouds and the Earth’s Radiant Energy System (CERES); longer records do not have accurate cloud variability at global scales. The climate research community would benefit from having a reliable observational constraint for cloud–aerosol albedo over a longer time period. Here, we produce a new estimate of shortwave (SW) cloud and aerosol albedo trends from UV satellite observations starting in 1980. This long-term and well-calibrated observational record of cloud albedo can help reduce the intermodel ECS spread by using it to evaluate CMIP6 historical and AMIP simulations.
Recent observation–model comparison studies have identified a number of cloud feedback mechanisms that likely contribute to higher or lower simulated ECS values in climate models: 1) simulation of incorrect spatial patterns of sea surface temperatures (SST) can lead to incorrect ECS values (Zhou et al. 2016), 2) parameterized aerosol–cloud interactions in some CMIP6 models are too sensitive to biomass aerosols and lead to excessively high simulated ECS values (Fasullo et al. 2022), and 3) mixed phase clouds simulated by many GCMs have an unrealistically low fraction of supercooled liquid compared with its ice phase (Tan et al. 2016). This available ice has the potential to transition to the more reflective liquid phase as the simulated climate warms, which leads to a cloud feedback that is too negative and yields simulated ECS values that are too low (Tan et al. 2016).
We have updated our previously reported record of black-sky cloud albedo (BCA) from UV observations (Weaver et al. 2020a, hereafter WEA20). The measured intensities of TOA radiance used here and in prior studies (WEA20; Weaver et al. 2015; Herman et al. 2013) are observed by the Solar Backscatter UV (SBUV) instruments onboard the Nimbus-7, NOAA-9, NOAA-11, NOAA-14, NOAA-16, NOAA-17, NOAA-18, and NOAA-19 spacecraft (Heath et al. 1975; Frederick et al. 1986; Hilsenrath et al. 1995). Nadir radiances from the Suomi NPP Ozone Mapping and Profiler Suite (OMPS) mapper are used for the last 2 years of our record (Seftor et al. 2014). The calibration of these UV radiances, treatment of instrument hysteresis, and the method to adjust the UV derived cloud optical depths (CODs) to a reference local time have all been improved. Because our UV narrowband-derived BCA is not directly comparable to the broadband SW cloud albedo archived on CMIP6 model output, we produce a proxy SW broadband cloud albedo from the BCA and additional independent quantities. We compare this proxy SW cloud albedo with output from 47 CMIP6 models (Eyring et al. 2016), discuss the impact of the cloud feedbacks mentioned above on our results, and finally estimate a value for ECS.
2. Instrument calibration
To produce accurate trends in ozone, the SBUV Version 8.6 ozone processing team has spent decades examining instrument calibration issues (McPeters et al. 2013). In this study, we leverage the efforts and instrument team’s extensive work by using the sun-normalized radiances reported in the SBUV Level-2 datasets (DeLand et al. 2012). These data have been calibrated to address time-dependent changes due to diffuser plate darkening, hysteresis, and other issues. While we adopt the time-dependent diffuser plate darkening rates for each SBUV instrument, we do not include Nimbus-7 observations poleward of 25°S. The Nimbus-7 SBUV radiances are tainted by photomultiplier tube hysteresis as the satellite emerged from polar night over Antarctica (see discussion in appendix A) which could impact our SW albedo trends. Thus, poleward of 25°S, the first month of our trend dataset is February 1985, while elsewhere it is January 1980. The SBUV field of view is approximately 170 km × 170 km at the surface. Since the OMPS nadir mapper has a much smaller field of view (FOV) (50 km at nadir), the observed near nadir intensities are aggregated to the SBUV FOV and processed exactly like the SBUV intensities.
A set of gain coefficients is constructed to intercalibrate the SBUV instruments. These are small multiplicative adjustments to the intensities, one for each instrument. We use a cost minimization approach to intercalibrate the SBUV and the OMPS mapper instruments. The first cost term is the mean of intensity differences when two or more instruments have temporal overlap over the Antarctic and Greenland ice sheets. The second term is the change in the observed ice sheet scene reflectivity over Antarctica over the 29-yr record (1985–2014). Our primary calibration has smaller intensity differences of temporally overlapping instruments balanced by a slight positive change in scene reflectivity over Antarctica (+0.352% over 29 years) but still within the 1σ slope uncertainty. The secondary calibration has larger overlapping intensity differences but a change in scene reflectivity over Antarctica closer to zero (+0.0612%). Note that the gain coefficients do not depend on time, so the interannual variability of a specific SBUV instrument remains intact after the initial calibration.
We first construct an empirical relationship between solar zenith angle and observed radiances over Antarctica and a separate relationship over Greenland using all 12 years of observations from NOAA-16. From this empirical relationship, we determine a reference radiance from the solar zenith angle associated with a given SBUV radiance observation. The fractional difference between the observed and the reference radiance is δΙ. We produce δΙ for all SBUV sensors, always using the same empirical relationship derived from NOAA-16, and then intercompare temporally overlapping sensors (see Fig. 1 and section c of appendix A).
(top) Time series of 100 × δI for SBUV and OMPS instruments over Antarctica and Greenland after our primary calibration. Average differences between temporally overlapping satellite pairs are shown. Solid circles mask intensities with incorrect grating drive positions (they only include observations with correct grating drive positions); open circles also include those with grating drive errors that have been corrected. (bottom) The anomaly of δI after the merged-satellite average is removed. Each dot is an average δI over a 15-day time period. Winter time periods (no observations) have been removed. See section c of appendix A and section 3 of Weaver et al. (2020b) for full description of δI. The dashed black trace is from the Nimbus-7 photo diode.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Figure 1 shows a satellite intercomparison of δΙ (in terms of %) over the Antarctic and Greenland ice sheets after applying our primary calibration. Note that our primary calibration gain coefficients in Table 1 are different than those in Weaver et al. (2020b). While both studies start with the same sun-normalized radiances reported in the SBUV Level-2 dataset, the new calibration averages the SBUV pixels over a finer temporal resolution (15 days) for both ice sheets. The previous calibration used an annual resolution. Also, the previous calibration attempted to correct the Nimbus-7 for hysteresis and included observations over Antarctica in the calibration. The current calibrations only use Nimbus-7 observations over Greenland.
Gain coefficients used for the two different calibrations used in this study. To account for the uncertainty in the calibration, which impacts the final proxy SW cloud albedo trend, we produce two sets of gain coefficients: Our primary set of gain coefficients is designed to minimize differences (offset biases) of instruments when they temporally overlap over Antarctica and Greenland. The second calibration has slightly larger offset biases but a temporal reflectivity change close to zero. The date ranges indicate periods when useful UV observations are available.
3. Calculation of cloud optical depth
Stronger Rayleigh multiple scattering in the UV, compared with visible wavelengths, diffuses the bidirectional reflectance distribution (BRDF) entering the instrument FOV, thus making it appear more Lambertian. Over ocean, rather than simply assuming the reflectance observed in the FOV is Lambertian, we retrieve a COD followed by a cloud albedo—all narrowband at 340 nm. For scenes with clouds over ocean, we use a single BRDF model for the cloud and an empirical BRDF model for the ocean (Cox and Munk 1954a,b). To account for surface wind speed, we use the Goddard Modeling and Assimilation Office MERRA-2 product (Rienecker et al. 2011). Our treatment of scenes of cloud over land is less complex; a single cloud model is assumed over a Lambertian surface.
a. Drifting orbits and cloud diurnal cycle
The diurnal cycle of clouds can easily distort a temporal trend derived from satellites in drifting orbits. A satellite in a sun-synchronous orbit which does not drift crosses the equator at the same local time each orbit and thus observes a given geographic location at the same local time throughout the mission. Satellite orbital drift changes the equator crossing time over time. All of the SBUV instruments used in this study except that on Nimbus-7 flew aboard satellites with drifting orbits. Taking measurements at different points of the daily cloud cycle has posed challenging problems for this and other cloud trend studies which use instruments on drifting satellites.
Instruments on satellites in morning orbits at the start of their missions which drift to afternoon equator-crossing times will yield an apparent decreasing trend in cloudiness over marine stratocumulus clouds and an apparent increasing cloudiness trend over arid locations. To ensure that our trends are not artificially affected by orbit-driven changes in diurnal sampling, we adjust the retrieved COD to a common reference time [1330 local time (LT)] using statistics from NOAA-11 through NOAA-19 (see appendix B). While this diurnal adjustment removes disparities between afternoon observations and the 1330 LT reference time, disparities between morning observations (before 1030 LT) and afternoon still exist. These morning observations are therefore not used in our trend analysis, because except for 1994, afternoon observations are always available.
b. Black-sky cloud albedo
The BCA is calculated from the diurnally adjusted cloud optical depths. For our retrieval, BCA is the ratio (scaled by 100) of hemispherically averaged upwelling TOA flux from clouds and aerosols to the downwelling flux. The flux contribution from the underlying surface and Rayleigh scattering was estimated by the vector linearized discrete ordinate radiative transfer (VLIDORT) package model (Spurr 2006) and was removed (see WEA20 for further details).
4. Estimation of shortwave cloud albedo
Our UV cloud products are all derived from the narrowband 340-nm radiances, but direct comparison with output from CMIP6 models runs requires an estimate of the shortwave broadband cloud albedo. We develop a method to convert the narrowband BCA to a broadband albedo using CERES observations when available and then apply this method over the entire UV data record.
a. CERES
The CERES has provided high-quality estimates of broadband TOA radiation from 2000 to present. The CERES TOA shortwave (0.3–5 μm) clear-sky and all-sky albedos used in this study are the monthly means of the CERES synoptic 1° (SYN1deg) TOA fluxes from Terra, Aqua, and NPP satellites (CERES_SYN1deg-TOA_Ed4.1; Loeb et al. 2018). Since we are interested in the TOA SW radiation from clouds, we use the difference between the clear-sky TOA SW flux and the all-sky flux. The clear-sky product only includes forcing from water vapor and surface albedo, while the all-sky product includes all forcings. The difference of these two products (clear sky minus all sky), defined as the SW cloud albedo radiative forcing (or the SW cloud radiative effect), is largely sensitive not only to cloud properties and amount but also to aerosols. This cloud albedo forcing is normalized by the TOA SW downward flux (i.e., solar insolation) to produce the CERES SW cloud and aerosol albedo.
b. Multivariate model
c. Cloud phase
The real-time CERES cloud phase (Minnis et al. 2021), which is archived with the SYN-1deg dataset and ultimately sourced from MODIS instruments, is used for the evaluation of the Eq. (1) coefficient d1. The CERES cloud phase is the point-spread-function weighted mean of the effective particle phase based on the 3.7-μm channel on the MODIS instruments. Values can account for multiphase clouds and range from 1.0 (pure liquid water droplets) to 2.0 (pure ice crystal). Since CERES cloud phase information is not available before 2000, we decided to use a time-independent cloud phase climatology for ω to estimate pCAalb. For consistency, this is used throughout the entire record. Since the climatology is time-independent, our pCAalb will not accurately account for changes in cloud phase that may have occurred as the atmosphere warmed since 1980.
d. Validation with CERES
We determine the Eq. (1) coefficients using only NOAA-18 observations and validate the approach using observations from other SBUV instruments that temporally overlap with the CERES. Spatial validation results using all NOAA-19 observations show similar patterns for BCA and observed CERES cloud albedo (Fig. 2). At any location, the difference in albedo between the CERES observations and our pCAalb is usually less than 3.5% albedo (Fig. 2d).
Maps for February 2009–13 of (a) UV BCA (%) from NOAA-19, (b) map of SW cloud albedo (%) from CERES (%), (c) UV-derived pCAalb (%), pCAalb, and (d) difference map of pCAalb minus CERES. All units are in terms of % albedo.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Scatterplots of albedo differences on the pixel level (no averaging performed) for both land and ocean (Fig. 3a) show that our pCAalb is often lower than CERES at higher albedos. The difference map (Fig. 2d) shows that this poor agreement (pCAalb is too low) occurs off the west coasts of South America and Africa.
(a) Differences between pCAalb minus CERES SW cloud albedo plotted against CERES during NOAA-19 over ocean (blue) and land (green). (b) Scatterplot of CERES SW cloud albedo vs UV BCA for water clouds (CERES cloud phase: 1.00–1.05, shown in light blue) and ice clouds (CERES cloud_phase: 1.9–2.0, shown in dark blue). Each dot is derived from an SBUV pixel. All units are in terms of % albedo.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Pure water phase clouds have significantly lower 340-nm BCA and CERES cloud albedo compared with pure ice phase clouds (Fig. 3b). In addition, a cloudy pixel scene with a given retrieved BCA value will have a higher (∼5) SW cloud albedo over a pure ice phase cloud compared to a water cloud. Figure 3b only shows the difference between the narrowband 340-nm cloud albedo and the broadband SW albedo; it says nothing about the lower reflectivity (optical depth) of ice crystals compared with liquid droplets for the same cloud water path.
Zonal mean temporal trends of pCAalb and CERES cloud albedo (Fig. 4a) show good agreement in the NH but disparities in the Southern Hemisphere (SH) extratropical latitudes. We suspect that issues with our diurnal cloud adjustment contribute to these differences. The 60°S–60°N average trends for CERES are lower than pCAalb (using daily cloud particle phase information) by 0.107 cloud albedo decade−1. Over the tropics, our pCAalb albedo product captures the positive trends at (5° and 20°N) and decreasing trends at 15°N seen by CERES. Changes in the observed and modeled width and location of the ITCZ are an intense area of research (Zhou et al. 2020; Broccoli et al. 2006) and most likely the driver of this complex trend variability. The narrow 3° latitudinal bands used to determine the trends shown in Fig. 4a result in modest correlations (r ∼0.4, gray trace) between pCAalb and the CERES albedo. Using wider latitudinal bands of 15° improves the correlation (r ∼0.7 not shown). There are geographic locations for which CERES albedo is available, but SBUV observations are missing or not processed, because the scenes include snow or ice. To gauge how much these missing data impact the trends, we show CERES trends that include all available data and trends that only include data where pCAalb is available. Both the red solid and dashed trends (Fig. 4a) show similar latitudinal variability, suggesting that the impact of missing data is small.
Comparison of zonal mean pCAalb with CERES SYN1deg SW cloud albedo (%). (a) Latitudinal variability of trends per decade from 2001 to 2014. Trends of pCAalb (black) along with trends using all CERES data (red dashed) and trends using only CERES data where UV observations are available (red solid) are shown. pCAalb is determined using time-dependent daily cloud particle phase information (solid) or a monthly climatology (dashed). Cosine-latitude weighted trends over 60°S–60°N are shown in the legend. The light gray line is the correlation between the monthly means of pCAalb and the CERES albedo. (b)–(d) Time series of albedos with time-averaged mean removed at select latitudes: 19.5°N, 40.5°S, and 52.5°N.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
We also feature time series at three selected latitudes (Figs. 4b–d). Good agreement is seen at 19.5°N where both time series show similar strong positive trends (Fig. 4b). The trends at 40.5°S highlight the SH disagreement between the pCAalb and CERES (Fig. 4c). At 52.5°N, the pCAalb lacks wintertime observations and overestimates the weaker negative trend seen by CERES which does include wintertime observations.
A similar study derives TOA SW flux from the Aura Ozone Monitoring Instrument (OMI) observations using a neural network approach (Gupta et al. 2016). Like our method, they train using CERES observations. While they input several independent parameters (ozone, cloud optical centroid pressure, and sun-satellite geometry), the equivalent cloud fraction (fc) is the main driver for SW variability in their neural network model. They investigate the spectral dependence of fc using Deirmendjian’s C1 model of a water cloud and using ice crystals (Baum et al. 2014). They show that fc is spectrally invariant and that the ice cloud is ∼18% higher than the water cloud calculation, consistent with our Fig. 3b. Our ability to capture the variability in the CERES broadband SW cloud albedo from a single narrowband UV measurement at 340 nm is consistent with their ability to estimate SW flux from the spectrally invariant fc derived from OMI.
e. Latitudinal comparison time series
A monthly mean time series of the full record (1980–2014) of zonal mean retrieved BCA at 15°-wide latitude bands over ocean and land allows an evaluation of the instrument calibration and the diurnal adjustment (Figs. 5–8). The closed circles on all time series plots denote BCA observed at a common local time close to the diurnal reference time of 1330 LT for equatorial bands, but later in the day in the SH (e.g., 1445 LT for 60°–45°S). The percent differences between each pair of sensors that temporally overlap are listed below the time series (Figs. 5–8).
Monthly mean time series of BCA (%) for latitude band 15°S: equator over ocean. Fraction of latitude band sampled by instrument, local time of observation, and SZA are shown below. The dashed line on the local time panel denotes 1400 LT. Observations within 30 min of the dashed local time are shown by larger closed circles. Differences (%) between temporally overlapping instruments are listed for individual pairs of color-coded instruments. The average percent difference (%diff) and the number of overlapping months (nn) for each pair are listed. The first line of the legend shows the average percent difference (mean %diff) and the average absolute value of the percent difference [mean abs (%diff)] for all pairs.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
As in Fig. 5, but for latitude band 30°S: 15°S over ocean. The dashed line on the local time panel is 1415 LT.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
As in Fig. 5, but for latitude band 45°N: 60°N over ocean and land. The dashed line on the local time panel denotes 1330 LT.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
As in Fig. 5, but for latitude band 45°S: 60°S over ocean and land. The dashed line on the local time panel denotes 1445 LT.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
The latitude band 15°S to the equator over ocean shown in Fig. 5 exhibits small differences in BCA between temporally overlapping instruments. The Nimbus-7, NOAA-9, and NOAA-11 traces show almost identical monthly variability and negligible offset during 1985–91, when they temporally overlap. The three pairs of instruments all have low percent differences (<2.5%) shown below the time series. Toward the end of the NOAA-11 mission, there is a several-month period in late 2000 for which data temporally overlap with our reference instrument NOAA-16. The albedo data exhibit good agreement during this overlap period, with an offset of ∼2%. The slow drift of NOAA-18 and NOAA-19 leads to two overlapping years (2009–11) with observations close to the 1330 LT reference time and a small offset (∼0.2%). The offset of NOAA-18 and NOAA-16 is negligible (∼0.5%). All of these offsets are most likely due to imperfections in our diurnal adjustment since they are larger than the sub one-percent offsets observed over the Antarctica and Greenland ice sheets.
The stratocumulus decks over the latitude band 30°–15°S (Fig. 6) include two large strongly diurnally varying marine stratocumulus decks off the South American and African coasts. Here, there are larger differences in BCA between temporally overlapping instruments compared with other latitude bands. The NOAA-9 and NOAA-11 pair both have full coverage and have small offset (0.6%). Comparisons with Nimbus-7 are not highlighted because it only includes observations between 25° and 15°S. The offsets between sensors observing in the morning (e.g., NOAA-17) and those observing in the afternoon (NOAA-18 and NOAA-19) are often above 10% and point to remaining issues with our diurnal adjustment method. However, except for NOAA-9 during 1994, measurements from sensors observing in the morning are not needed or used in the trend analysis.
At polar latitudes, useful measurements are obtained only during the 6–7-month period of late spring, summer, and early fall. However, even with the polar night gap, the SBUV instruments can detect a decadal oscillation in the 45°–60°N latitude band (Fig. 7). Here, the temporal offsets are small enough so that all instruments contribute to this robust feature. In contrast, the 60°–45°S band (Fig. 8) has such large offsets between overlapping sensors that only a linear trend can be established. This poor performance of our diurnal adjustment may be a result of the strong cloud diurnal amplitude (defined as the difference between morning and afternoon values, ΔCODam_pm) shown in the SH polar latitudes (Fig. B1).
5. Comparison with CMIP6 models
We compare the global mean trends of CMIP6 cloud albedo from AMIP and historical runs with our pCAalb. In addition, we evaluate the spatial variability in the trends of each CMIP6 model with respect to pCAalb, using maps and zonal means. The AMIP simulations are atmosphere-only runs using prescribed SST and sea ice concentrations from 1979 to the present. These comparisons allow for an evaluation of the atmospheric component of the simulated cloud feedback. Historical simulations use a coupled atmosphere–ocean general circulation model (AOGCM). These comparisons also test the accuracy of the changes in the SST distribution simulated by the CMIP6 ocean models as the climate warms (Liu et al. 2022).
Our pCAalb retrieval algorithm is not able to separate the albedo signals from aerosols and clouds, so a CMIP6 product that includes both clouds and aerosols would be the ideal way to compare with our product. Since most modeling groups only archive products that allow construction of a cloud albedo that does not include aerosols, the comparison with our product is not direct. We calculate the CMIP6 cloud albedo from (rsut − rsutcs)/rsdt, where rsut is the TOA outgoing SW flux, rsutcs is the TOA outgoing SW flux assuming clear sky but with aerosols, and rsdt is incoming SW flux. Comparisons for each model are shown in Supplement.historical.pdf and Supplement.AMIP.pdf in the online supplemental material. A few modeling groups archive a TOA outgoing SW flux assuming clear sky and aerosol-free (rsutcsaf) conditions, which allow for a direct comparison (see Supplement.rsutcsAF.historical.pdf).
a. CESM2 model
We first highlight the Community Earth System Model, version 2 (CESM2), from NCAR since it accurately simulates zonal mean and spatial features in our observational record and has a relatively high ECS value of 5.2°C. First, we consider maps of the simulated and observed temporal trends (Figs. 9a–c). The AMIP simulation provided by CESM2 (Fig. 9a) captures many of the observed features over the Pacific and North Atlantic Oceans; the historical simulation (Fig. 9b) shows similar agreement. Both time series of the near global anomaly (60°S–60°N) for the CMIP6 and pCAalb (Figs. 10a,b) show no discernible trend. Time periods impacted by major volcanic eruptions (April 1982–April 1983 and May 1991–December 1993) are shown using dotted lines and are not included in the trend analysis. The 340-nm SBUV measurements do not distinguish between clouds and volcanic sulfate aerosols injected into the stratosphere which explains the positive response of pCAalb during volcanically perturbed time periods. This different treatment of aerosols also explains the model–observation disparity over India and China and is discussed below (section 5e).
Spatial and zonal mean comparison of SW cloud trends from the CESM2 CMIP6 model with our proxy SW cloud+aerosol albedo derived from SBUV sensors (pCAalb). Maps of temporal trends of cloud albedo from (a) CESM2 AMIP runs, (b) CESM2 historical runs, and (c) pCAalb using our primary calibration. The CMIP6 models are contoured from their archived spatial resolution; the pCAalb is spatially gridded to a 3° × 5° resolution. Zonal mean trends in SW cloud albedo for the available (d) CESM2 AMIP and (e) CESM2 historical model realizations compared with pCAalb. For sensitivity analysis, zonal mean trends of pCAalb using the primary (solid) and secondary (dashed) are shown. Area averaged differences in zonal mean trends (60°S–60°N) are shown in title along with correlation r. For subsequent figures, CMIP6 models with high ECS values are colored red, middle ECS values in green, and low values in blue. The ECS values are derived from long-time scale model feedbacks of 1–150 years. Also note that for this CMIP6 model, the TOA outgoing SW assumes clear sky with aerosols (rsutcs), so the comparison with our product is not direct.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Time series of cosine-latitude weighted near global (60°S–60°N) anomaly for various CMIP6 models (colored) and for pCAalb (black). Time periods impacted by volcanoes (April 1982–April 1983 and May 1991–December 1993) are dotted and not included in trend analysis.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Finally, we compare zonal mean trends of pCAalb and the CESM2 model by latitude. Both AMIP and historical runs accurately simulate our observations of increasing cloudiness in the NH tropics and decreasing cloudiness in the NH midlatitudes (Figs. 9d,e). We construct two metrics to evaluate a CMIP6 model. The first compares the latitudinal variability of the CMIP6 model trends with the variability of pCAalb. Specifically, the zonal mean trends of the CMIP6 cloud albedo are correlated with the zonal mean trends of pCAalb. The second metric compares the near-global (60°S–60°N) trend in CMIP6 cloud albedo with our pCAalb. Specifically, we calculate the global average trend of the CMIP6 cloud albedo (rsut − rsutcs)/rsdt and subtract the global average trend of pCAalb; both averages are cosine-latitude weighted. The correlations and global trend differences (% albedo per decade) are both shown in the upper legend in red for the CESM2 model.
b. Equilibrium climate sensitivity
Previous studies demonstrate that differences in simulated cloud feedback from model to model explain a significant spread of the ECS (Zelinka et al. 2020; Cess et al. 1990; Bony and Dufresne 2005). The ECS values used in our study are calculated using the Gregory method; CO2 is instantaneously quadrupled in a fully coupled atmosphere–ocean model and run for 150 years (Gregory et al. 2004). These ECS values quantify long-term feedbacks on time scales > 150 years and are listed in Table 1 from Dong et al. (2020) and Meehl et al. (2020).
c. Pattern effect
A set of coupled climate models with the same global mean surface temperature evolution but with different spatial patterns of the change in the surface temperature can exhibit quite different transient radiative responses. A number of recent studies suggest that if the ocean warming is concentrated in specific geographic locations, the global response may be amplified. These studies identify the eastern Pacific, ascent regions of the western Pacific, and the Southern Ocean as sensitive locations.
Over the eastern Pacific, observationally derived AMIP SSTs show a cooling trend since 1980 but warming trends elsewhere [Fig. 2 of Silvers et al. (2018) and Fig. 3 of Zhou et al. (2016)]. The implication is that the cooling surface waters will strengthen the lower tropospheric inversion leading to more stratocumulus (note that atmospheric temperature advection maintains unchanging warm air above the inversion). Indeed, our Fig. 9c shows increased cloudiness off the coast of South America and the eastern equatorial Pacific. Using carefully designed AMIP-like experiments, Zhou et al. (2016) show that increases in simulated low-level cloud cover over the eastern Pacific are driven by SST spatial patterns as opposed to uniform SST warming. They argue that this pattern effect is responsible for the hiatus in global warming (1998–2013), a negative cloud feedback, and they suggest that climate sensitivities derived from observations during this period might be biased low.
Another study uses a Green’s function approach (Dong et al. 2019) to determine regional locations that control global radiative changes. Here, the collective oceans are divided up into 137 patches. A separate run of the NCAR CAM4 model tests the impact of a local SST perturbation over each patch on the global radiative response. While this study shows results consistent with the above study (SST warming decreases low cloud amount leading to positive TOA radiation change), by far the dominant driver of global radiative change is the ascent regions in the western tropical Pacific. A similar study that also uses a Green’s function approach identifies the Southern Ocean as a key driver of the global climate feedback (Kang et al. 2023).
d. Comparison with all CMIP6 models
We assume that the temporal trend in SW cloud albedo is representative of the SW cloud feedback and that the difference between the trend obtained from a CMIP6 model and our observed trend in pCAalb indicates how that model’s feedback compares with the actual atmosphere. Thus, positive values of this difference indicate that the CMIP6 model has increasing cloudiness compared with the observations and that the cloud feedback within the specific model is too negative, while negative differences indicate that the SW cloud feedback in the model is too strongly positive. The near-global (60°S–60°N) trend differences for all AMIP runs (Fig. 11a) with respect to the ECS of each model are shown. The correlation coefficient (first metric described above) is related to the size of dot and denotes how accurately a model simulates the latitudinal variability of the pCAalb observational trend.
Comparison of global (60°S–60°N) temporal trends in SW cloud albedo from (a) AMIP and (b) historical CMIP6 models with our pCAalb. For each CMIP6 model, its trend minus the pCAalb trend is plotted against the model’s ECS. Positive trend differences have increasing cloudiness compared with observations, while negative trend differences have decreasing cloudiness compared with observations. CMIP6 models within the yellow-shaded zone have global trends that fall within the primary and secondary calibrations. Models that have a correlation coefficient < 0.2 do not simulate the latitudinal variability of the pCAalb observational trend and are not shown. The uncertainty in the albedo trends, shown by the error bars, is based on the variability of the global mean trends from multiple realizations for each CMIP6 model (see text). For the AMIP comparison, the green arrows show the ECS values where linear fits intercept an albedo trend difference equal to zero: 3.4°C for the primary calibration and 5.6°C for the secondary. For the historical model comparison, we focus on the four historical models that simulate albedo trend spatial patterns that qualitatively agree with our observations over the Pacific (Fig. C1). These are highlighted with dark blue circles; a linear fit is shown in dark blue. The green arrows show the ECS values where linear fits intercept an albedo trend difference equal to zero (3.5°C for the primary calibration and 5.1°C for the secondary).
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Since AMIP models use prescribed SST observations trends, they are not affected by the pattern effect, at least over ocean. An AMIP model that accurately simulates atmospheric cloud feedback should show good agreement with our observations. But if this same model does not simulate the observed SST patterns, it will be penalized in the historical comparison. The historical runs of GCMs with high ECS (>5°C) (Fig. 11b) often have stronger cloud feedback than their AMIP counterparts. For example, the AMIP CESM2 has a global mean cloud albedo of 0.01% albedo per decade above our pCAalb observed trends (see RHS of Fig. 11a), but the historical CESM2 has 0.04% albedo per decade below the pCAalb trends. Zhou et al. (2016) also report more negative (dissipative) low cloud cover trends for an ensemble (16 models) of historical CMIP5 models compared with AMIP simulations. They attribute the strong low cloud cover dissipation in the historical models to the ocean GCMs simulating incorrect SST patterns in the Pacific. Interestingly, the CESM2 model highlighted in Figs. 9a and 9b shows similar spatial cloud trends for both the AMIP and historical runs. Most of the other historical CMIP6 models show more spatially uniform trends in cloud albedo than their AMIP kin, although there are other exceptions besides CESM2. Figure C4 shows four historical models that have spatial trends in cloud albedo that are not uniform and show some increase in cloudiness in the eastern Pacific similar to our observations.
Comparison of pCAalb with output of the AMIP atmosphere-only runs allows for an evaluation of the atmospheric component of the modeled cloud feedback (Fig. 11a). The yellow-shaded region shows the CMIP6 models that exhibit global cloudiness trends that fall within our two calibrations. These models are more consistent with our observations and suggest that the simulated cloud feedbacks within these models are more realistic. The ECS value for these models ranges from 2.6°C for the MIROC6 model to the 5.6°C for the CanESM5 model. A linear fit of the albedo trend differences with ECS suggests an ECS of 3.4°C for the primary calibration and 5.6°C for the secondary. To estimate the internal climate variability (ICV) for all AMIP models, we determine the span of the global mean trends for each model. The span is the realization with the maximum trend minus the realization with the minimum trend. Our reported ICV is the AMIP model with the largest span. The average ICV for all AMIP models is 0.05% albedo per decade and 0.09 for all historical CMIP6 models. These are shown by the error bars in Fig. 11. The AMIP models do not have the additional variability from an ocean model so their ICV is less than that for the historical runs.
Figure 12 shows brackets that compare the correlations for the AMIP (red) and historical (blue) simulations. For each CMIP6 model, the AMIP correlations are often higher than the historical ones, but not always. We estimate the range by determining the correlation between pCAalb and the CMIP6 realizations for all AMIP and all historical CMIP6 models. Consistent with the additional ICV in the historical CMIP6 models, the correlation range is 0.36–0.65 for historical runs and 0.47–0.65 for AMIP. The reported low-range value is the average minimum for all realizations, whereas the high value is the average maximum.
Correlation between the zonal mean trends of a CMIP6 model against the zonal mean trends of pCAalb. The size of the dot increases with increasing correlation coefficient. Models that have a correlation coefficient < 0.2 do not simulate the latitudinal variability of the pCAalb observational trend and are not shown. Brackets pair the AMIP:historical correlations for some of the models.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
We highlight two other interesting model comparisons. First, the historical U.K. Earth System Model, version 1, low resolution (UKESM1-0-LL) simulation shows strong cloud dissipation at most tropical and midlatitude locations, clearly in disagreement with our observations (Fig. 13). The cloud albedo time series (Fig. 10c) shows global cloud dissipation over the 34-yr record. Next, both the historical runs of MRI-ESM2-0 (Figs. 14a,d) and the AMIP (not shown) accurately simulate our observations of the latitudinal dependence of the trend in albedo, suggesting that the evolution of simulated SST patterns, at least in the zonal mean, is realistic.
As in Fig. 9, but for UKESM1-0-LL historical CMIP6 model.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Comparison of (a),(d) cloud albedo without aerosols with (b),(e) albedo from clouds and aerosols. Simulated albedos are from historical MRI-ESM2-0 CMIP6 model runs. All other figures in this paper use TOA outgoing SW assuming clear sky with aerosols (rsutcs), but here outgoing SW assumes clear sky free of aerosols (rsutcsAF). Using rsutcsAF affords direct comparison with our pCAalb which includes clouds and aerosols. (c) Map of pCAalb is the same as in Fig. 9c.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
e. Aerosol–cloud interaction
The high ECS exhibited by some of the CMIP6 models may be partly driven by aerosol–cloud interactions in addition to cloud feedback. To learn how the temporal variability in aerosol emissions impacts the climate system, the biomass burning (BB) emissions prescribed for the CMIP6 historical simulations (van Marle et al. 2017) contain separate periods characterized by low and high interannual variability. A climatological BB database with low temporal variability is used prior to 1997. After 1997, satellites began monitoring wildfires at high temporal and spatial resolution. The Global Fire Emissions Database version 4 (GFED; van der Werf et al. 2017) captures this high-resolution information and provides BB emissions for CMIP6 simulations. Targeted sensitivity experiments with CESM2 show that from 40° to 70°N, the high temporal variability of BB emissions drives a thinning of the cloud field, warmer surface temperatures (Fasullo et al. 2022), and amplification of the hydrologic cycle (Heyblom et al. 2022).
A few CMIP6 models have archived output from historical runs that allows a direct comparison with our pCAalb. Analysis of these comparisons may hint whether a CMIP6 model is also sensitive to high temporal BB emissions. The MRI-ESM2-0 modeling group archives a TOA outgoing SW flux assuming clear-sky and aerosol-free (rsutcsaf) conditions, which allows a calculation of a cloud and aerosol albedo and a direct comparison with pCAalb. Comparison of Fig. 14a with Fig. 14b shows that aerosols drive a strong decreasing trend in SW cloud albedo over Europe and North America. A comparison of the zonal means of the albedo trends (Fig. 14d with Fig. 14e) show that the large aerosol-driven negative trends north of 30°N disagree with the more modest trends of pCAalb. Most of the CMIP6 models that archive rsutcsaf also show that the aerosols drive similar strong negative trends that disagree with our determination of pCAalb over these Northern Hemisphere locations (Supplement.rsutcsAF.historical.pdf). This result suggests that other CMIP6 models share the sensitivity of CESM2 to BB emissions, at least to some degree, from 40° to 70°N. Accounting for aerosols also drives an increasing positive trend in albedo over China and India that is more consistent with our observations.
f. Volcanic response
As mentioned above, most CMIP6 models only provide information to calculate albedos that include the effects of clouds. For those few models that provide output that allows for the calculation of an albedo from clouds and aerosols, the volcanic response is positive (see colored dotted lines in Fig. 10e). This result is consistent with the positive response seen in the observations (pCAalb) due to volcanic sulfate aerosols.
6. Discussion
a. Correlation metric
While the evaluation of the CMIP6 cloud albedo trends is dependent on the SBUV instrument calibration approach and its associated uncertainty, the evaluation of the trend in latitudinal variability is not. Changes in the calibration will not impact Fig. 12 or the size of the circles in Fig. 11a or Fig. 11b. About half of the CMIP6 climate models fail to capture features of the latitudinal variability in the observed trend (correlation < 0.2). These models are not used in the evaluations of ECS, nor are they shown in Fig. 11 or Fig. 12. For the remaining CMIP6 models, the AMIP runs (red circles) generally have a more positive SW cloud albedo trend compared with their historical runs (blue circles).
b. Aerosols
As mentioned above, most CMIP6 models only provide information for an indirect comparison between CMIP6 cloud albedo and our determination of pCAalb, which also includes the effect of tropospheric aerosols. Consequently, we have more confidence interpreting model to measurement comparisons for geographic locations far away from aerosol sources, such as the SH and Pacific Oceans. We have less confidence for comparisons over locations with known aerosol trends: for example, over North America and Europe where there were decreases in aerosol optical depths (AODs) from tightening emission controls and over China and India where AOD has risen from human activities (Chin et al. 2014). To address complications posed by aerosols and allow a more robust comparison, we encourage modeling groups to archive the TOA outgoing SW flux assuming clear-sky and aerosol-free conditions (rsutcsaf).
The radiative forcing from aerosols under clear-sky conditions as well as from aerosol–cloud interaction is included in pCAalb. A recent review paper (Bellouin et al. 2020) reports a range from −1.6 to 0.6 W m−2 for total aerosol forcing; estimates of the forcing from cloud–sulfate aerosol interactions from satellite observation constraints (Wall et al. 2022) are −1.1 W m−2. If the atmosphere was free of aerosols, our ECS would likely be significantly higher than our reported estimate of 3.5°C. In fact, this may eventually be the case as the anthropogenic contribution to aerosol forcing diminishes from continued pollution–control policies.
c. Cloud phase and morphology
Our pCAalb product will not accurately capture the expected reduction in albedo due to phase changes from ice crystals to liquid droplets as the climate warms. However, the CERES cloud albedo uses time-varying cloud-phase information from MODIS and should capture this contribution to changes in albedo. We can estimate the phase impact over the CERES time period by comparing the pCAalb using a monthly climatology of phase information (dashed black line in Fig. 4a) with pCAalb using time-dependent phase information (solid black line). The 60°S–60°N average trend accounting for phase changes is less than the climatology by 0.01 cloud albedo per decade. This translates to an underestimate of our ECS by 0.5°C or more.
Another time-varying mechanism that our pCAalb trends will not accurately capture is changes in cloud morphology. While our diurnal adjustments are based on seasonally averaged statistics, the cloud morphology maps (Fig. B1) and the COD daily cycle statistics, used to make the adjustments (Fig. B2), are static over our 34-yr data record. A locational shift, expansion or shrinking of a cloud regime, will not impact the initially retrieved COD but might introduce an error in the diurnally adjusted COD that could result in an incorrect trend.
d. PATMOS-x ISCCP
Even with the issues with aerosols, potential errors from cloud phase and errors from a fixed cloud morphology, the zonal mean trends from our product track the latitudinal variability of two other observations datasets fairly well. Percentage cloud amount trend estimates from the International Satellite Cloud Climatology Project (ISCCP; Rossow and Schiffer 1999) dataset and the Pathfinder Atmospheres–Extended (PATMOS-x; Heidinger et al. 2014) are shown in Fig. 15. Comparison of our pCAalb albedo trend with PATMOS-x cloud trend is good, especially south of 30°N; comparison with ISCCP is less so. Differences in the NH may be related to tropospheric aerosols and need further investigation. Note that the construction procedure for the PATMOS-x and ISCCP datasets removes the global-scale trends (Norris et al. 2016), so we can only compare latitudinal variability.
Our pCAalb (black traces, described in Fig. 9) compared with trends in ISCCP percentage cloud amount (red) and trends in PATMOS-x percentage cloud amount (blue). All trends are from 1983 to 2009. Transcribed from Fig. 2 of Norris et al. (2016) using software by Ankit Rohatgi (WebPlotDigitizer; https://automeris.io/WebPlotDigitizer 4.3). Note that different units are shown.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
e. Calibration
Since the Antarctic albedo is stable over time, it can be used for long-term satellite calibration. The only mechanism that might significantly change the albedo is when sulfate aerosols from the Pinatubo volcanic eruption were advected over the SBUV instrument field of view. Volcanic aerosols were detected by the SAGE II instrument after the eruption (Jaross and Warner 2008). Also, non-sea-salt sulfate has been detected in firn samples from several sites across the Antarctic Plateau after the eruption (Legrand and Wagenbach 1999). Our approach is to force our calibrations to have a negligible change in δΙ over Antarctica during the 29-yr record from 1985 to 2014 (recall that Nimbus-7 observations are not used over Antarctica because of hysteresis issues). The δΙ change is based on a linear fit of δΙ with time excluding the year after the Pinatubo eruption.
In addition, our calibrations minimize the offset biases between instruments when two or more temporally overlap. Our two featured calibrations have a different balance of these two objectives. The intent of the primary calibration is to equally balance both objectives. It has a nonzero (+0.352) change in δΙ over 29 years but still within 1σ of the error in the δΙ change and has minimal instrument bias. The secondary calibration has an almost zero (+0.061) change in δΙ but larger offset biases between overlapping instruments.
f. ESC estimation
Since AOGCMs are used to make future multidecadal climate projections and are used to estimate the ECS value of a CMIP model, comparison with model output from historical simulations at first appears to be the more appropriate choice for estimating the ECS value of the real atmosphere–ocean system. Others have reported that most AOGCMs fail to simulate key observed SST patterns across the Pacific and should be used with caution in this type of analysis. However, there are four AOGCMs that do simulate albedo trend spatial patterns that qualitatively agree with our observations over the Pacific (see Fig. C1 in appendix C). These are shown with dark blue circles in Fig. 11b. The fitted slope of these four historical models (dark blue line) is significantly less than the slope of all the historical models (light blue line) and close to the AMIP fitted slope in Fig. 11a. The ECS value where the fitted linear relationship has a zero difference between simulated and observed albedo trends is our estimate of the true atmosphere–ocean system ECS. Based on our primary calibration, the ensemble of AMIP models and the four selected historical models estimate an ESC between 3.4° and 3.5°C. Our secondary calibration has an almost zero temporal trend in δΙ over Antarctica. We judge this secondary calibration as being almost as likely to be correct as the primary calibration, except that the offset biases of temporally overlapping instruments is larger than the primary calibration. To estimate the lower bound of uncertainty, we use a calibration with a positive δΙ change of (+0.45) over Antarctica, which yields a value for ECS of 2.7°C. While the range of uncertainty of true atmosphere–ocean system ECS likely lies somewhere between 2.7° and 5.1°C, our best estimate is the value of 3.5°C found using the primary calibration.
7. Conclusions
We produce a record of proxy shortwave albedo from clouds and aerosols using SBUV and OMPS mapper radiance observations from 1980 to 2014 and compare this albedo product with archived output from AMIP and historical CMIP6 simulations from 47 climate models. We first consider the latitudinal variability of each CMIP6 model’s zonal mean cloud trend and compare it with zonal mean trends from our observations. A significant number of CMIP6 models were unable to simulate the observed latitudinal variability and were discarded. Also eliminated were historical models that were unable to simulate the observed spatial patterns of cloud albedo trends over the eastern Pacific. The ensemble of the remaining 24 AMIP models and four historical models estimates an ECS of ∼3.5°C with an uncertainty range between 2.7° and 5.1°C. This 2.4°C range of uncertainty is driven by the strong sensitivity of ECS to the temporal trend in Antarctic ice reflectivity (as observed by δΙ) assumed in the calibration. About a dozen CMIP6 models have global mean cloud albedo trends within the range of uncertainty of our UV observations. Future plans are to quantitatively compare the spatial patterns and seasonal variability of the climate model’s cloud albedo trends with our observations. Note that the 34-yr record of our observations is still relatively short compared with the time scale of 150 years used to determine the ECS of the individual CMIP6 models. That said, even with a 34-yr record, we were able to evaluate the fidelity of 47 CMIP6 models and found that a significant number were unable to simulate features seen in our observed albedo trends. We suspect that these poor preforming models will continue to have difficulty simulating our observation as we extend the length of our record in the future.
We have considered nongeophysical factors that potentially could influence our temporal SW albedo trend. For example, Nimbus-7 radiance observations sampled over the SH are subject to instrument hysteresis and were therefore not in this study. Also, more robust cloud albedo trends were obtained, compared to WEA20, by incorporating sampling bias correction for the cloud diurnal cycle and by improving the intersatellite calibration of the SBUV radiances.
Most CMIP6 modeling groups only archive quantities that allow construction of a cloud-only SW albedo, which is different than our product, so our conclusions are more robust for regions that are distant from strong tropospheric aerosol sources. More robust direct comparisons with our record will be possible once modeling groups archive clear-sky and aerosol-free TOA outgoing SW flux as well.
Acknowledgments.
This work was supported by a NASA Making Earth System Data Records for Use in Research Environments (MEaSUREs) Project, the NASA Long-Term Measurement of Ozone Program, and the NASA Earth-Sun Connection Program. We appreciate detailed critique of the calibration section from Matthew DeLand and Liang-Kang Huang along with encouragement from Richard Stolarski. Many thanks are due to Anne Douglass for helpful comments and editing and also to John Fasullo for pointing out aerosol–cloud interaction issues with CMIP6 models.
Data availability statement.
Monthly mean gridded shortwave proxy cloud and aerosol albedo are available from the corresponding author (clark.j.weaver@nasa.gov).
APPENDIX A
Calibration
a. Diffuser plate darkening
The SBUV instrument observes nadir-viewed narrowband backscatter radiances from Earth Io. To ensure that solar variability is not impacting the SBUV observations, a diffuser plate can be deployed to direct the solar radiation into the SBUV monochromator photomultiplier tube (PMT) to measure the solar irradiance F. The directional albedo (I = Io/F), input to our BCA calculation, is independent of solar activity variability. The solar diffuser plate darkens over time and is the main source of SBUV time-dependent changes that must be characterized. NOAA-9 and later SBUV/2 instruments featured an onboard calibration system to track relative changes in diffuser reflectivity using a mercury lamp (Weiss et al. 1991). After the NOAA-9 SBUV/2 instrument experienced problems with lamp stability, an alternative approach was developed that uses the Antarctic Plateau as a stable terrestrial albedo reference (Huang et al. 2003; Jaross and Warner 2008). The long-term SBUV instrument characterization results obtained using the snow/ice radiance technique are generally comparable to results derived from the onboard calibration system at 340 nm. The agreement between these methods is excellent (within ∼0.5%) for NOAA-14 and NOAA-16 (DeLand et al. 2012).
b. Hysteresis
The Nimbus-7, and to a lesser extent the NOAA-9, instrument PMT was not able to respond to the rapid increase in radiance signal when the satellite first emerges from SH polar night darkness on each orbit. Uncorrected radiances can be off by 8%–9% at high SH latitudes, and their impact can reach to the equator during SH winter. The Nimbus-7 radiances were corrected for the V8.6 ozone product using the onboard photodiode instrument (DeLand et al. 2012). To be cautious, we do not use any Nimbus-7 observations poleward of 25°S to ensure that the hysteresis does not impact our trends.
Hysteresis is most apparent at the highest solar zenith angles when the SBUV instrument is seeing its first light after polar night darkness. Even though the PMT intensities have been corrected for hysteresis, at the beginning of the record (1980), the PMT-derived δI is still more than 8% lower than the photodiode-derived δI at the highest solar zenith angle (SZA) (82°; Fig. A1). As expected, the disparity is reduced at lower SZA values. Also note that toward the end of the Nimbus-7 record (1990), there is negligible difference between the PMT and photodiode, suggesting that the Nimbus-7 intensities are accurate.
Time series of δI for SBUV instruments over Antarctica. As in Fig. 1 of the main text, but binned by SZA.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
c. Empirical calibration
Measured intensity at 340 nm from the NOAA-16 SBUV vs SZA over the Antarctic Plateau (blue) and Greenland (green). Each point is a nadir-viewed observation at the native field of view (170 km × 170 km) during the summer (15 days on either side of solstice). All 12 years of NOAA-16 observations are used to generate the polynomial fits. Note that the Greenland intensities are offset from the Antarctic ones. The right panel shows a zoomed-in view (see text for details).
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
Only the highest quality observations are used for the intercalibration. Observations are limited to θο less than 75° because at higher θο, ozone absorption and straylight effects become significant and contaminate results. Furthermore, SBUV observations that have a grating drive error and observations that are likely impacted by PMT hysteresis are not used to intercalibrate.
APPENDIX B
Cloud Optical Depth
The determination of COD from the observed radiances using the vector linearized discrete ordinate radiative transfer package (Spurr 2006) is detailed in WEA20. We assume a C-1 water cloud droplet size distribution (Deirmendjian 1969) with an effective radius of 10 mm over ocean and 22 mm over land field of views. A cloud top and bottom of 700 and 900 hPa, respectively, is assumed.
Diurnal adjustment
Our current diurnal adjustment of COD to a reference local time is an improvement to the approach described in WEA20 and Labow et al. (2011). Adjustments are based on frequency distributions binned by geographic locations and now season (note that there is no binning by solar zenith angle). Locations are determined using a climatology of differences in COD observed in the local time morning (0815–1115 LT) minus afternoon (1245–1545 LT) (ΔCODam_pm). The climatology is constructed from the suite of six SBUV instruments (NOAA-11 through NOAA-19; Fig. B1). Diurnal cycles of the five different geographic areas used in our study (Fig. B2) show the strongest amplitudes in summer/autumn and weakest in the winter consistent with previous observational studies (Wood 2012; Eastman and Warren 2014).
Difference between morning and afternoon observed COD (CODam_pm) from NOAA-11 through NOAA-19 for (a) spring, (b) summer, (c) autumn, and (d) winter. Different months are used for North and South Hemisphere locations. For example, spring uses February, March, and April for NH locations and November, December, and January for SH locations. COD used for this figure has not been diurnally adjusted.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
The COD daily cycle for the five geographic areas used to generate diurnal cloud statistics. The boxed colors are consistent with the legend of the maps in Fig. B1. Morning local time observations are equally distributed throughout the NOAA-11 through NOAA-19 (1989–2014) time period; some are at the beginning and some at the end. The same is true for afternoon observations. However, the only source for near-noon time observations is from Nimbus-7 which is at the start of the record. Since we are concerned that the diurnal adjustment using Nimbus-7 would cause an artificial trend, we opted not to include this sensor when constructing the statistics. Instead, the near-noon statistics are interpolated from late morning and early afternoon local time observations and shown by gray circles. The red circle is the 1330 LT reference time.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
The five geographic areas are shown with different colors in Fig. B1. For example, the geographic area bin “ocean stratocumulus” is defined by locations where ΔCODam_pm > 2.0 and shown by the red/orange contours; low-level marine stratocumulus that peak in early morning frequently exist here. The geographic area bin “cumulus” is defined by locations where ΔCODam_pm is between −2.0 and −0.5 and shown by the light blue contours; higher altitude clouds peaking in afternoon prevail here.
We construct COD frequency distributions for each geographic area, season, and further bin by local time using nine local time bins from 0800 to 1700 LT. Percentile distributions are produced next. Our diurnal adjustment assumes that the percentile value Pobs of a COD observation will not depend on the local time of observation. The term Pobs is the rank of an observation with respect to the others within its bin. So, if an observation is ranked in the 40th percentile based on the statistics for the local time of observations, then the adjusted COD is the value at the 40th percentile determined from the reference (1330 LT) local time statistics.
In section 4, the same seasonal climatology of ΔCODam_pm is used to convert the black-sky cloud albedo to a proxy shortwave cloud albedo. NOAA-9 through NOAA-19 were all launched into an early afternoon orbit around 1330 LT. At the beginning of a mission, the solar diffuser plate only experiences minimal darkening, so we choose this as our reference local time.
APPENDIX C
Spatial Averaging, Merging, and Selected CMIP6 Models
Pixel values of each SBUV instrument are averaged to 15° latitude bands (displayed in Figs. 5–8) or gridded to a 3° × 5° resolution (Fig. 9) using a monthly time window. Merging all the SBUV instruments into a single record involves preferentially using SBUV instruments with a local time close to the reference local time (1330 LT); instruments with local times in the morning or late afternoon are used as a last resort. Specifically, for a given month, we check if there are instruments observing within 1 h of the reference local time. If there are, we merge them and move on to the next month, but if there are no instruments, the local time window is expanded to 2 h and so on. Instruments with local times before 1030 LT are not included in the merged record. Once candidate instruments are chosen for a given month, they are averaged, weighting them by the number of pixel samples for the given month.
a. Treatment of 1994–95
There is no temporal overlap between NOAA-11 and NOAA-14 during 1994. Although observations from NOAA-9 exist, we do not trust them. They are morning observations and there are large disparities between NOAA-9 and the other two when they temporally overlap. For each 15° latitude band, a constant offset is added to the NOAA-9 observations to minimize the temporal overlap.
b. Selected CMIP6 models
Figure C1 shows the four historical models that simulate albedo trend spatial patterns that qualitatively agree with our observations over the Pacific.
Maps of SW cloud albedo trends for four CMIP6 models that show increasing trends in eastern Pacific similar to our pCAalb trends.
Citation: Journal of Climate 37, 11; 10.1175/JCLI-D-23-0170.1
REFERENCES
Baum, B. A., P. Yang, A. J. Heymsfield, A. Bansemer, A. Merrelli, C. Schmitt, and C. Wang, 2014: Ice cloud single-scattering property models with the full phase matrix at wavelengths from 0.2 to 100 μm. J. Quant. Spectrosc. Radiat. Transfer, 146, 123–139, https://doi.org/10.1016/j.jqsrt.2014.02.029.
Bellouin, N., and Coauthors, 2020: Bounding global aerosol radiative forcing of climate change. Rev. Geophys., 58, e2019RG000660, https://doi.org/10.1029/2019RG000660.
Bony, S., and J.-L. Dufresne, 2005: Marine boundary layer clouds at the heart of tropical cloud feedback uncertainties in climate models. Geophys. Res. Lett., 32, L20806, https://doi.org/10.1029/2005GL023851.
Broccoli, A. J., K. A. Dahl, and R. J. Stouffer, 2006: Response of the ITCZ to Northern Hemisphere cooling. Geophys. Res. Lett., 33, L01702, https://doi.org/10.1029/2005GL024546.
Cess, R. D., and Coauthors, 1990: Intercomparison and interpretation of climate feedback processes in 19 atmospheric general circulation models. J. Geophys. Res., 95, 16 601–16 615, https://doi.org/10.1029/JD095iD10p16601.
Chin, M., and Coauthors, 2014: Multi-decadal aerosol variations from 1980 to 2009: A perspective from observations and a global model. Atmos. Chem. Phys., 14, 3657–3690, https://doi.org/10.5194/acp-14-3657-2014.
Cox, C., and W. Munk, 1954a: Statistics of the sea surface derived from sun glitter. J. Mar. Res., 13, 198–227.
Cox, C., and W. Munk, 1954b: Measurement of the roughness of the sea surface from photographs of the Sun’s glitter. J. Opt. Soc. Amer., 44, 838–850, https://doi.org/10.1364/JOSA.44.000838.
Deirmendjian, D., 1969: Electromagnetic Scattering on Spherical Polydispersions. Elsevier Science, 290 pp.
DeLand, M. T., S. L. Taylor, L. K. Huang, and B. L. Fisher, 2012: Calibration of the SBUV version 8.6 ozone data product. Atmos. Meas. Tech., 5, 2951–2967, https://doi.org/10.5194/amt-5-2951-2012.
Dong, Y., C. Proistosescu, K. C. Armour, and D. S. Battisti, 2019: Attributing historical and future evolution of radiative feedbacks to regional warming patterns using a Green’s function approach: The preeminence of the western Pacific. J. Climate, 32, 5471–5491, https://doi.org/10.1175/JCLI-D-18-0843.1.
Dong, Y., K. C. Armour, M. D. Zelinka, C. Proistosescu, D. S. Battisti, C. Zhou, and T. Andrews, 2020: Intermodel spread in the pattern effect and its contribution to climate sensitivity in CMIP5 and CMIP6 models. J. Climate, 33, 7755–7775, https://doi.org/10.1175/JCLI-D-19-1011.1.
Eastman, R., and S. G. Warren, 2014: Diurnal cycles of cumulus, cumulonimbus, stratus, stratocumulus, and fog from surface observations over land and ocean. J. Climate, 27, 2386–2404, https://doi.org/10.1175/JCLI-D-13-00352.1.
Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 1937–1958, https://doi.org/10.5194/gmd-9-1937-2016.
Fasullo, J. T., J.-F. Lamarque, C. Hannay, N. Rosenbloom, S. Tilmes, P. DeRepentigny, A. Jahn, and C. Deser, 2022: Spurious late historical-era warming in CESM2 driven by prescribed biomass burning emissions. Geophys. Res. Lett., 49, e2021GL097420, https://doi.org/10.1029/2021GL097420.
Frederick, J. E., R. P. Cebula, and D. F. Heath, 1986: Instrument characterization for the detection of long-term changes in stratospheric ozone: An analysis of the SBUY/2 radiometer. J. Atmos. Oceanic Technol., 3, 472–480, https://doi.org/10.1175/1520-0426(1986)003<0472:ICFTDO>2.0.CO;2.
Gregory, J. M., and Coauthors, 2004: A new method for diagnosing radiative forcing and climate sensitivity. Geophys. Res. Lett., 31, L03205, https://doi.org/10.1029/2003GL018747.
Gupta, P., J. Joiner, A. Vasilkov, and P. K. Bhartia, 2016: Top-of-the-atmosphere shortwave flux estimation from satellite observations: An empirical neural network approach applied with data from the A-train constellation. Atmos. Meas. Tech., 9, 2813–2826, https://doi.org/10.5194/amt-9-2813-2016.
Heath, D. F., A. J. Krueger, H. A. Roeder, and B. D. Henderson, 1975: The Solar Backscatter Ultraviolet and Total Ozone Mapping Spectrometer (SBUV/TOMS) for Nimbus G. Opt. Eng., 14, 144323, https://doi.org/10.1117/12.7971839.
Heidinger, A. K., M. J. Foster, A. Walther, and Z. Zhao, 2014: The Pathfinder Atmospheres–Extended AVHRR climate dataset. Bull. Amer. Meteor. Soc., 95, 909–922, https://doi.org/10.1175/BAMS-D-12-00246.1.
Herman, J., and Coauthors, 2013: A net decrease in the Earth’s cloud, aerosol, and surface 340 nm reflectivity during the past 33 yr (1979–2011). Atmos. Chem. Phys., 13, 8505–8524, https://doi.org/10.5194/acp-13-8505-2013.
Heyblom, K. B., H. A. Singh, P. J. Rasch, and P. DeRepentigny, 2022: Increased variability of biomass burning emissions in CMIP6 amplifies hydrologic cycle in the CESM2 large ensemble. Geophys. Res. Lett., 49, e2021GL096868, https://doi.org/10.1029/2021GL096868.
Hilsenrath, E., R. P. Cebula, M. T. Deland, K. Laamann, S. Taylor, C. Wellemeyer, and P. K. Bhartia, 1995: Calibration of the NOAA 11 Solar Backscatter Ultraviolet (SBUV/2) ozone data set from 1989 to 1993 using in-flight calibration data and SSBUV. J. Geophys. Res., 100, 1351–1366, https://doi.org/10.1029/94JD02611.
Huang, L.-K., R. P. Cebula, S. L. Taylor, M. T. DeLand, R. D. Mcpeters, and R. S. Stolarski, 2003: Determination of NOAA-11 SBUV/2 radiance sensitivity drift based on measurements of polar ice cap radiance. Int. J. Remote Sens., 24, 305–314, https://doi.org/10.1080/01431160304978.
IPCC, 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, 1535 pp., https://doi.org/10.1017/CBO9781107415324.
IPCC, 2023: Climate Change 2023: Synthesis Report. Cambridge University Press, 184 pp., https://doi.org/10.59327/IPCC/AR6-9789291691647.
Jaross, G., and J. Warner, 2008: Use of Antarctica for validating reflected solar radiation measured by satellite sensors. J. Geophys. Res., 113, D16S34, https://doi.org/10.1029/2007JD008835.
Kang, S. M., P. Ceppi, Y. Yu, and I.-S. Kang, 2023: Recent global climate feedback controlled by Southern Ocean cooling. Nat. Geosci., 16, 775–780, https://doi.org/10.1038/s41561-023-01256-6.
Labow, G., J. R. Herman, L. K. Huang, S. A. Lloyd, M. T. DeLand, W. Qin, J. Mao, and D. E. Larko, 2011: Diurnal variation of 340 nm Lambertian equivalent reflectivity due to clouds and aerosols over land and oceans. J. Geophys. Res., 116, D11202, https://doi.org/10.1029/2010JD014980.
Legrand, M., and D. Wagenbach, 1999: Impact of the Cerro Hudson and Pinatubo volcanic eruptions on the Antarctic air and snow chemistry. J. Geophys. Res., 104, 1581–1596, https://doi.org/10.1029/1998JD100032.
Liu, H., Z. Song, X. Wang, and V. Misra, 2022: An ocean perspective on CMIP6 climate model evaluations. Deep-Sea Res II, 201, 105120, https://doi.org/10.1016/j.dsr2.2022.105120.
Loeb, N. G., and Coauthors, 2018: Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) Top-of-Atmosphere (TOA) edition-4.0 data product. J. Climate, 31, 895–918, https://doi.org/10.1175/JCLI-D-17-0208.1.
McPeters, R. D., P. K. Bhartia, D. Haffner, G. J. Labow, and L. Flynn, 2013: The version 8.6 SBUV ozone data record: An overview. J. Geophys. Res. Atmos., 118, 8032–8039, https://doi.org/10.1002/jgrd.50597.
Meehl, G. A., C. A. Senior, V. Eyring, G. Flato, J.-F. Lamarque, R. J. Stouffer, K. E. Taylor, and M. Schlund, 2020: Context for interpreting equilibrium climate sensitivity and transient climate response from the CMIP6 Earth system models. Sci. Adv., 6, eaba1981, https://doi.org/10.1126/sciadv.aba1981.
Minnis, P., and Coauthors, 2021: CERES MODIS cloud product retrievals for edition 4—Part I: Algorithm changes. IEEE Trans. Geosci. Remote Sens., 59, 2744–2780, https://doi.org/10.1109/TGRS.2020.3008866.
Norris, J. R., R. J. Allen, A. T. Evan, M. D. Zelinka, C. W. O’Dell, and S. A. Klein, 2016: Evidence for climate change in the satellite cloud record. Nature, 536, 72–75, https://doi.org/10.1038/nature18273.
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA’s Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648, https://doi.org/10.1175/JCLI-D-11-00015.1.
Rossow, W. B., and R. A. Schiffer, 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 2261–2288, https://doi.org/10.1175/1520-0477(1999)080<2261:AIUCFI>2.0.CO;2.
Seftor, C. J., G. Jaross, M. Kowitt, M. Haken, J. Li, and L. E. Flynn, 2014: Postlaunch performance of the Suomi National Polarorbiting Partnership Ozone Mapping and Profiler Suite (OMPS) nadir sensors. J. Geophys. Res. Atmos., 119, 4413–4428, https://doi.org/10.1002/2013JD020472.
Sherwood, S. C., and Coauthors, 2020: An assessment of Earth’s climate sensitivity using multiple lines of evidence. Rev. Geophys., 58, e2019RG000678, https://doi.org/10.1029/2019RG000678.
Silvers, L. G., D. Paynter, and M. Zhao, 2018: The diversity of cloud responses to twentieth century sea surface temperatures. Geophys. Res. Lett., 45, 391–400, https://doi.org/10.1002/2017GL075583.
Spurr, R. J. D., 2006: VLIDORT: A linearized pseudo-spherical vector discrete ordinate radiative transfer code for forward model and retrieval studies in multilayer multiple scattering media. J. Quant. Spectrosc. Radiat. Transfer, 102, 316–342, https://doi.org/10.1016/j.jqsrt.2006.05.005.
Tan, I., T. Storelvmo, and M. D. Zelinka, 2016: Observational constraints on mixed‐phase clouds imply higher climate sensitivity. Science, 352, 224–227, https://doi.org/10.1126/science.aad5300.
van der Werf, G. R., and Coauthors, 2017: Global fire emissions estimates during 1997–2016. Earth Syst. Sci. Data, 9, 697–720, https://doi.org/10.5194/essd-9-697-2017.
van Marle, M. J., and Coauthors, 2017: Historic global biomass burning emissions for CMIP6 (BB4CMIP) based on merging satellite observations with proxies and fire models (1750–2015). Geosci. Model Dev., 10, 3329–3357, https://doi.org/10.5194/gmd-10-3329-2017.
Wall, C. J., J. R. Norris, A. Possner, and N. J. Lutsko, 2022: Assessing effective radiative forcing from aerosol–cloud interactions over the global ocean. Proc. Natl. Acad. Sci. USA, 119, e2210481119, https://doi.org/10.1073/pnas.2210481119.
Weaver, C. J., J. Herman, G. Labow, D. Larko, and L.-K. Huang, 2015: Shortwave TOA cloud radiative forcing derived from a long-term (1980–present) record of satellite UV reflectivity and CERES measurements. J. Climate, 28, 9473–9488, https://doi.org/10.1175/JCLI-D-14-00551.1.
Weaver, C. J., D. L. Wu, P. K. Bhartia, G. J. Labow, and D. P. Haffner, 2020a: A long-term cloud albedo data record since 1980 from UV satellite sensors. Remote Sens., 12, 1982, https://doi.org/10.3390/rs12121982.
Weaver, C. J., P. K. Bhartia, D. L. Wu, G. J. Labow, and D. E. Haffner, 2020b: Inter-calibration of nine UV sensing instruments over Antarctica and Greenland since 1980. Atmos. Meas. Tech., 13, 5715–5723, https://doi.org/10.5194/amt-13-5715-2020.
Weiss, H., R. P. Cebula, K. Laamann, and R. D. Hudson, 1991: Evaluation of the NOAA-11 solar backscatter ultraviolet radiometer, mod 2 (SBUV/2): Inflight calibration. Proc. SPIE, 1493, 80–90, https://doi.org/10.1117/12.46686.
Wood, R., 2012: Stratocumulus clouds. Mon. Wea. Rev., 140, 2373–2423, https://doi.org/10.1175/MWR-D-11-00121.1.
Zelinka, M. D., T. A. Myers, D. T. McCoy, S. Po-Chedley, P. M. Caldwell, P. Ceppi, S. A. Klein, and K. E. Taylor, 2020: Causes of higher climate sensitivity in CMIP6 models. Geophys. Res. Lett., 47, e2019GL085782, https://doi.org/10.1029/2019GL085782.
Zelinka, M. D., S. A. Klein, Y. Qin, and T. A. Myers, 2022: Evaluating climate models’ cloud feedbacks against expert judgment. J. Geophys. Res. Atmos., 127, e2021JD035198, https://doi.org/10.1029/2021JD035198.
Zhou, C., M. D. Zelinka, and S. A. Klein, 2016: Impact of decadal cloud variations on the Earth’s energy budget. Nat. Geosci., 9, 871–874, https://doi.org/10.1038/ngeo2828.
Zhou, W., L. R. Leung, J. Lu, D. Yang, and F. Song, 2020: Contrasting recent and future ITCZ changes from distinct tropical warming patterns. Geophys. Res. Lett., 47, e2020GL089846, https://doi.org/10.1029/2020GL089846.
