Evaluation of the Surface Radiation Budget in the Atmospheric Component of the Hadley Centre Global Environmental Model (Hadgem1)

15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4723

Evaluation of the Surface Radiation Budget in the Atmospheric Component of the Hadley Centre Global Environmental Model (HadGEM1)

A. BODAS-SALCEDO,M.A.RINGER, AND A. JONES Met Office, Hadley Centre, Exeter, United Kingdom

(Manuscript received 13 June 2007, in final form 1 February 2008)

ABSTRACT

The partitioning of the earth radiation budget (ERB) between its atmosphere and surface components is of crucial interest in climate studies as it has a significant role in the oceanic and atmospheric general circulation. An analysis of the present-day climate simulation of the surface radiation budget in the atmospheric component of the new Hadley Centre Global Environmental Model version 1 (HadGEM1) is presented, and the simulations are assessed by comparing the results with fluxes derived from satellite data from the International Satellite Cloud Climatology Project (ISCCP) and ground measurements from the Baseline Surface Radiation Network (BSRN). Comparisons against radiative fluxes from satellite and ground observations show that the model tends

to overestimate the surface incoming solar radiation (Ss,d). The model simulates Ss,d very well over the polar regions. Consistency in the comparisons against BSRN and ISCCP-FD suggests that the ISCCP-FD database is a good test for the performance of the surface downwelling solar radiation in climate model simulations. Overall, the simulation of downward longwave radiation is closer to observations than its shortwave counterpart. The model underestimates the downward longwave radiation with respect to BSRN measurements by 6.0 W mϪ2. Comparisons of land surface albedo from the model and estimates from the Moderate Resolution Im- aging Spectroradiometer (MODIS) show that HadGEM1 overestimates the land surface albedo over deserts and over midlatitude landmasses in the Northern Hemisphere in January. Analysis of the seasonal cycle of the land surface albedo in different regions shows that the amplitude and phase of the seasonal cycle are not well represented in the model, although a more extensive validation needs to be carried out. Two decades of coupled model simulations of the twentieth-century climate are used to look into the model’s simulation of global dimming/brightening. The model results are in line with the conclusions of the studies that suggest that global dimming is far from being a uniform phenomenon across the globe.

1. Introduction tion regarding the top-of-the-atmosphere (TOA) radiation budget, and they have been extensively used as The earth radiation budget (ERB) is the ensemble of validation tools for climate models (Li et al. 1997; radiative fluxes entering and leaving the earth– Chevallier and Morcrette 2000; Bony et al. 2004; Ringer atmosphere system, which drives the earth’s climate. and Allan 2004). However, TOA ERB measurements Therefore, measurements of this balance are needed to do not provide a complete constraint on the atmo- improve our knowledge of the earth’s climate and cli- sphere’s radiative properties. This means that the par- mate change (Ramanathan 1987; Ramanathan et al. titioning of the ERB between the atmosphere (ATM) 1989). Since the 1970s, great efforts have been made to and surface (SFC) components is of crucial interest in measure this budget globally and with sufficient accu- climate studies. Gleckler (2005) showed that this parti- racy by means of broadband sensors aboard satellites. tioning has a significant role in the oceanic and atmo- All these measurements provide invaluable informa- spheric general circulation. In addition, the balance between longwave radiative cooling and latent heating establishes a link between radiative processes and the hydrological cycle. Therefore, any changes in the radia- Corresponding author address: Dr. A. Bodas-Salcedo, Met Of- fice, Hadley Centre, FitzRoy Rd., Exeter EX1 3PB, United King- tive budget of the atmosphere will have an impact on dom. the response of the hydrological cycle (Stephens 2005). E-mail: [email protected] These two reasons highlight the importance of knowing

DOI: 10.1175/2008JCLI2097.1

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

JCLI2097 4724 JOURNAL OF CLIMATE VOLUME 21 both the surface and the atmospheric radiation budgets lation of global dimming/brightening in section 7. (SARB), not only for their direct impact on the general Conclusions are presented in section 8. circulation, but also for their role in climate feedback problems. 2. Model description and experimental design As satellites provide TOA measurements, the SRB has to be modeled from those measurements, together We use present-day climate simulations from the at- with information about the state of the atmosphere and mosphere-only version of the new Hadley Centre cli- the surface. A global perspective of the surface radia- mate model, HadGEM1, referred to as HadGAM1. tion budget, both in the shortwave and longwave parts HadGAM1 uses a horizontal resolution of 1.25° lati- of the spectrum, can only be obtained from satellites. tude by 1.875° longitude, and has 38 vertical levels, the Several studies have used the International Satellite top level being at around 39 km. The simulations used Cloud Climatology Project (ISCCP) C1 data to provide are from a five-member ensemble of model runs of a global perspective of the surface radiation budget HadGAM1, forced with observed sea surface tempera- (Pinker and Laszlo 1992; Darnell et al. 1992; Whitlock tures (SSTs) from the second Atmospheric Model In- et al. 1995; Zhang et al. 1995; Gupta et al. 1999). Li and tercomparison Project (AMIP-II; Gates et al. 1999), Leighton (1993) computed a global climatology of the each member using different initial conditions. The solar radiation budget using data from the Earth Ra- runs start on December 1978, and we use a 20-yr cli- diation Budget Experiment (ERBE), not relying on matology from 1981 to 2000. We use monthly mean ISCCP data. Li (1995) intercompared the net surface diagnostics of the different components of the SARB as well as other diagnostics that help the interpretation of shortwave radiation (NSSR) as derived from ERBE the results (e.g., cloud cover and precipitable water against the NSSR derived from ISCCP by Whitlock et content). Here we give some details of the physical al. (1995). More recently, these datasets have been up- processes relevant for the simulation of the surface ra- graded by using ISCCP-D1 data and improved algo- diation budget, but a more detailed description of Had- rithms (Stackhouse et al. 1999; Zhang et al. 2004). In GAM1, and its performance in terms of global clima- addition, these satellite-based databases, along with tology, variability, and regional climate can be explored surface observations, have been used to evaluate the in Martin et al. (2006) and Ringer et al. (2006). performance of the simulations of the surface radiation The radiation code is that of Edwards and Slingo budget by climate models (Garrat 1994; Li et al. 1997; (1996) used in the third climate configuration of the Wild et al. 1998; Wild 2005). However, interest in the Met Office Unified Model (HadCM3; HadAM3 for the SRB is not confined to global scales. Smith et al. (2002) atmospheric component), with some developments. showed that the regional climate and surface radiation The longwave band from 1200 to 1500 cmϪ1 has been are related, and Bolle et al. (2006) used the surface net split at 1330 cmϪ1 in order to better represent the over- radiation as a possible indicator of changes in the Medi- lap between CH4 and N2O; gaseous absorption is based terranean region. on the updated High-Resolution Transmission (HIT- Note that we usually refer to the atmospheric com- RAN) 2000 database (Rothman et al. 2003); the water ponent of the Hadley Centre Global Environmental vapor continuum is version 2.4 of the Clough–Kneizys– Model version 1 (HadGEM1) as the Hadley Centre Davies (CKD) formulation (Clough et al. 1992) and has Global Atmospheric Model (HadGAM1), and we will been included in the shortwave region; ice crystal use this notation throughout the paper. This paper is sizes are determined using the parameterization by organized as follows: The data used in this study and a Kristjánsson et al. (2000); the sea surface albedo is brief description of the model are presented in section based on the functional form of Barker and Li (1995), 2. Section 3 compares the climatologies of SRB pro- modified in the light of aircraft data; and the land sur- vided by HadGAM1 and ISCCP-FD. The perfomance face albedo is described by Essery et al. (2003). of HadGAM1’s simulation of the surface radiation bud- The simple aerosol climatology used previously (Cu- get is evaluated against surface observations in section sack et al. 1998) has largely been superseded in Had- 4. Section 5 looks at the representation of the interan- GAM1 by schemes to interactively simulate sulfate, nual variability of surface incoming radiation over the fossil-fuel black carbon, biomass-burning and sea-salt tropical Pacific, and section 6 compares HadGAM1’s aerosols, as detailed in Martin et al. (2006). Only the land surface albedo with that from the Moderate Reso- stratospheric sulphuric acid aerosol component of the lution Imaging Spectroradiometer (MODIS). Two de- earlier climatology has been retained. The direct radia- cades of coupled model simulations of the twentieth- tive effect (scattering and absorption of radiation) of all century climate are used to look into the model’s simu- aerosols is included; this means that the “semi-direct”

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4725 effect (the impact on clouds of the warming caused by terval (i.e., S for shortwave, L for longwave, and T for absorbing aerosols; Hansen et al. 1997) is also included. total), the first subscript denotes the level (i.e., t for Parameterizations of both first and second indirect TOA, a for atmosphere, and s for surface), and the aerosol effects (impact on cloud droplet size and on second subscript denotes the direction (i.e., u for up- precipitation efficiency, respectively) are also included, welling, d for downwelling, and n for net). For instance, with sulfate, biomass-burning, and sea-salt aerosols be- St,u is the TOA shortwave upwelling radiation, and Ts,n ing considered cloud condensation nuclei; black carbon is the net total surface radiation. Although we use the aerosols are assumed to be hydrophobic and so do not terms overestimation and underestimation in the com- have indirect effects in HadGAM1. parison between model simulations and ISCCP-FD es- As a global representation of the observed surface timates, one should not forget the inherent limitations radiation budget, we use the ISCCP-FD database in the ISCCP-FD dataset, so these terms have to be (Zhang et al. 2004). The ISCCP-FD dataset contains understood as higher or lower than the ISCCP-FD val- radiative fluxes at the top of the atmosphere, the sur- ues. face, and at three levels in the atmosphere (680, 440, a. Regional means and 100 hPa). They are calculated using ISCCP-D cloud products (Rossow and Schiffer 1999) as the main Figure 1 shows the geographical distributions of the input, along with information from other sources in or- annual means of Ss,d, Ss,n, and total cloud amount for der to complete the radiative description of the atmo- HadGAM1 and ISCCP-FD. HadGAM1 generally sphere and surface. An assessment of the quality of the overestimates Ss,d over landmasses, this overestimation ISCCP-FD fluxes can be consulted in Zhang et al. being more severe in the summer hemisphere (not (2004) and Raschke et al. (2005). Previous versions of shown). A positive bias is also observed in the eastern the methodology are documented in Zhang et al. (1995) Pacific intertropical convergence zone (ITCZ) and and Rossow and Zhang (1995). South Pacific convergence zone (SPCZ). Although the We also use ground measurements at 28 sites belong- representation of cloud has notably improved with re- ing to the Baseline Surface Radiation Network (BSRN; spect to the previous Hadley Centre model (Martin et Ohmura et al. 1998). The BSRN is a project of the al. 2006), HadGAM1 shows a general lack of cloud World Climate Research Program (WCRP) that aims almost everywhere, which is particularly severe over to provide radiation ground measurements to validate the regions where Ss,d is overestimated (Figs. 1c,f,i). satellite-derived products and climate models and to The maps of Ss,n (Figs. 1b,e,h) show an overall picture detect long-term climate variations. Standard BSRN ra- very similar to that of Ss,d, but there are some relevant diation measurements are provided at 1-min temporal differences that highlight differences in the surface al- resolution, although we degrade the time series to a bedo. For instance, in spite of the severe overestimation monthly mean resolution in order to match the tempo- of Ss,d over the Sahara Desert and over most of the ral resolution of the model diagnostics. In addition, ow- landmasses, the difference maps of Ss,n show a mixture ing to the disparity of the spatial resolutions between of positive and negative biases over those regions, the model grid and the ground measurements, we ex- which indicate higher albedos than those used by pect this time-averaging process to reduce the errors ISCCP-FD. These results show relevant differences in introduced by this disparity in resolutions. the surface albedo between HadGAM1 and ISCCP-FD In addition to ISCCP and BSRN data, other data- that deserve more attention. We analyze surface albedo bases have been used as sources of independent infor- in greater detail in section 6. mation on surface and atmospheric properties: 16-day The regional distribution pattern of Ls,d (Figs. 2a,d,g) averages of surface albedo data from the MODIS in- is highly correlated to the distribution of precipitable strument onboard the Terra satellite (Schaaf et al. water content (PWC; Figs. 2b,e,h) and surface tempera- 2002). ture (Figs. 2c,f,i). HadGAM1 shows a negative bias in

Ls,d over most land regions, particularly over the desert 3. Climatology regions of the Northern Hemisphere (NH). These errors in Ls,d are correlated with errors in the distribution In this section results of the comparisons of the 20-yr of PWC over land regions. Although still present, this SRB climatologies between HadGAM1 and ISCCP-FD correlation is not as strong over ocean. It can be ob- are presented. We also examine the representation of served how positive biases in PWC over the Indian surface properties in HadGAM1. We will use the fol- Ocean are not linked to biases in downward LW radia- lowing notation for the radiative fluxes throughout this tion. A possible explanation is the nonlinear response paper: the first (capital) letter denotes the spectral in- of Ls,d with respect to PWC. The actual amount of

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4726 i ie4/C live 1 Fig ORA FCLIMATE OF JOURNAL Unauthenticated |Downloaded 10/10/2104:08PMUTC

Ϫ2 FIG. 1. Annual means of (a), (d), (g) Ss, d and (b), (e), (h) Ss, n. Differences are HadGAM1 minus ISCCP-FD. Units are W m . V OLUME 21 5S 15 i ie4/C live 2 Fig EPTEMBER 2008 OA-ACD TAL. ET BODAS-SALCEDO Unauthenticated |Downloaded 10/10/2104:08PMUTC

Ϫ2 Ϫ2 FIG. 2. Annual means of (a), (d), (g) Ls, d (W m ); (b), (e), (h) precipitable water content (kg m ); and (c), (f), (i) surface temperature (K) for HadGAM1 and ISCCP-FD. Differences are HadGAM1 minus ISCCP-FD. 4727 4728 JOURNAL OF CLIMATE VOLUME 21

PWC over that region is high (greater than 40 kg mϪ2), in all-sky conditions still show a large discrepancies making the lower troposphere already very opaque to among the current generation of GCMs (Wild 2005).

LW radiation. Therefore, the response of Ls,d to a HadGAM1 shows a clear-sky atmospheric absorption change in PWC is almost saturated. Changes in PWC of 69.6 W mϪ2, very close to 70 W mϪ2, quoted by Wild do not seem to explain the differences in Ls,d in the et al. (2006) as the most likely value of solar radiation stratocumulus regions in the eastern basins of the sub- absorbed in the cloud-free atmosphere. The underesti- tropical oceans. Negative biases in PWC are observed mation of cloud amount shown in Fig. 1i is not entirely in these regions, but an impact in Ls,d is observed only responsible for the overestimation of Ss,d. We have in regions close to the coast. In these regions, differ- compared long-term averages of model clear-sky Ss,d ences in low cloud may be compensating for errors in over the 17 BSRN locations used by Wild et al. (2006). PWC. Model SSTs are prescribed, and therefore free of The model shows an overestimation of 8.2 W mϪ2 with errors introduced by model physics. Over land this is respect to the ground observations, which is Ϸ50% of not the case, and errors in PWC are usually associated the all-sky bias (see section 4). This contrasts with the with errors in surface temperature (Figs. 2c,f,i), with negligible bias shown by HadAM3 in clear-sky Ss,d both contributing to errors in Ls,d. Over the ocean, dif- (Wild et al. 2006), aerosols being the main cause for this ferences in PWC and cloud are the main contributors to overestimation. The aerosol species missing in Had- the differences in Ls,d, so it appears that the sensitivity GAM1, and the fact that HadAM3 uses an aerosol cli- to errors in PWC over the ocean is smaller than over matology (Cusack et al. 1998), means HadAM3 has a land. Over the ocean, the greatest differences in surface more accurate (altough less physically based) represen- temperature are observed in midlatitudes, particularly tation of the climatological radiative impact of aerosols. in the winter hemisphere (not shown). Over subtropical Figures 3a–f show the hemispheric annual cycle of deserts HadGAM1 underestimates land surface tem- the surface radiation budget. Values are anomalies with perature with respect to ISCCP-FD. The opposite be- respect to the annual mean values shown in Table 1. havior occurs in tropical rain forests, where HadGAM1 Downward SW flux, surface albedo, and net SW flux overestimates land surface temperature, especially over are shown, as well as downward, upward, and net LW the Amazon. In midlatitude land regions, the differ- fluxes. Figure 3a shows a strong annual cycle in Ss,d, ences are seasonally dependent, and HadGAM1 shows with amplitudes of more than 50 WmϪ2 in both hemi- a positive bias in summer and a negative bias in winter spheres. The Southern Hemisphere (SH) shows a (not shown). greater amplitude, which is consistent with the fact that the sun–Earth distance is a minimum during summer in b. Global means and seasonal cycle the SH, thereby adding the effects of more vertical il- Global and hemispheric values of the SRB for DJF, lumination and less sun–Earth distance (Gupta et al. JJA and annual means are given in Table 1. The largest 1999). In addition, the NH has a stronger annual cycle differences between the model and ISCCP-FD are of precipitable water content than the SH, with high found in the NH, both in absolute and relative values. values in summer, which also helps reduce the ampli-

Differences between members of the ensemble are neg- tude of the annual cycle of Ss,d (Randel et al. 1996; ligible (not shown), and therefore differences between Wittmeyer and Vonder Haar 1994). The model repro- simulations and ISCCP-FD cannot be explained by the duces the annual cycle reasonably well, although it model variability. As seen above, HadGAM1 generally overestimates the amplitude in the NH and underesti- overestimates Ss,d in both seasons, globally and in both mates it in the SH. Annual values are overestimated in hemispheres. HadGAM1, particularly in the NH where the amount

The overestimation of Ss,d has been a consistent bias of SW radiation reaching the surface is 6% more than for many years in previous generations of models, not that obtained by ISCCD-FD. only in all-sky conditions, but also in cloudless atmo- The seasonal cycle of upward SW fluxes (Ss,u; not spheres (Wild et al. 1995; Li et al. 1997; Wild et al. 1998; shown) is one order of magnitude smaller than that of Chevallier and Morcrette 2000). Possible origins of this the downward fluxes (typical hemispheric albedos of error are the underestimated water vapor absorption in Ϸ0.1). Surface albedos are shown in Fig. 3c. HadGAM1 the near infrared and the crude representation of aero- shows small differences in amplitude and phase with sols. Improvements in the radiative transfer codes of respect to ISCCP-FD in the NH. However, the SH several models have led to a better representation of shows big differences both in amplitude and phase. the global mean energy distribution between the atmo- HadGAM1 shows a well-defined seasonal cycle, with a sphere and the surface under cloud-free conditions maximum during winter consistent with greater solar

(Wild et al. 2006). However, global mean values of Ss,d zenith angles, whereas ISCCP-FD shows a very weak

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4729

FIG. 3. Hemispheric, seasonal cycle of surface radiation budget. Anomalies with respect to annual mean values are represented. Units are W mϪ2, except for the albedo, expressed in %. (a) Surface downward SW radiation, (b) surface downward LW radiation, (c) surface albedo, (d) surface upward LW radiation, (e) surface SW net radiation, and (f) surface LW net radiation. Gray shades in HadGAM1 curves show the range of variability from the five-member ensemble of model runs.

seasonal cycle with maximum in September–October. and Ss,u. The overestimation in the surface albedo

Also note the difference in the annual mean values, partly compensates for the overestimation in Ss,d, mak- with HadGAM1 showing surface albedos greater than ing the relative differences in annual means in Ss,n ISCCP-FD in both hemispheres (Table 1). Figure 3e smaller than those in Ss,d. shows the results for the SW net flux, which are similar The downward LW flux (Ls, d) is presented in Fig. 3b. to those of the downward flux owing to the big differ- Because of the greater amplitude of the annual cycle of ences in the amplitudes of the seasonal cycles of Ss,d surface temperature over landmasses and the coupling

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4730 JOURNAL OF CLIMATE VOLUME 21

TABLE 1. Global and hemispherical means of the different com- balance of the atmosphere occurs as a balance between ponents of the SRB in HadGAM1. Values in parentheses are for the radiative cooling of the atmosphere and latent heat- Ϫ2 the ISCCP-FD climatology. Units are W m . ing associated with precipitation, and therefore cloud NH SH Globe feedbacks that affect the radiative heating of the atmosphere will influence the response of the hydrological Ss,d DJF 140.0 (130.7) 263.3 (260.4) 201.1 (195.6) cycle (Stephens 2005). JJA 252.7 (239.8) 129.3 (121.6) 190.4 (180.7) Figure 4 shows the comparisons for the zonal means Annual 200.2 (188.1) 196.0 (190.4) 197.5 (189.3) of the radiation budget (left-hand column) and cloud Ss,u radiative forcing (right-hand column) at TOA, ATM, DJF 21.4 (16.3) 36.0 (35.2) 28.8 (25.7) and SFC. The scales are not intended to provide accu- JJA 37.3 (29.9) 12.7 (9.4) 25.1 (19.7) Annual 31.1 (24.8) 23.9 (22.0) 27.6 (23.4) rate detail, but to provide a global picture of the energy

Ss,n distribution. At TOA, St,n, Lt,n, and Tt,n are well repre- DJF 118.6 (114.4) 227.3 (225.3) 172.3 (169.8) sented by the model, although with some compensation JJA 215.4 (209.9) 116.6 (112.2) 165.4 (161.0) between errors in the SW and LW in the tropics. This Annual 169.2 (163.4) 172.0 (168.5) 169.9 (165.9) shows the well-known distribution of Tt,n, with positive Ls,d DJF 309.5 (318.5) 352.6 (354.6) 330.2 (336.6) values between 40°S and 40°N, and negative values JJA 373.3 (378.8) 325.1 (325.4) 348.6 (352.1) poleward. The atmosphere shows a consistent radiative Annual 339.9 (347.3) 338.6 (340.1) 338.6 (343.7) cooling at all latitudes (ϳ100 W mϪ2), caused by Ls,u greater LW cooling than SW heating. Both SW heating DJF 370.6 (372.0) 407.6 (397.6) 388.3 (384.8) and LW cooling are greater in the tropics than in mid- JJA 435.5 (426.1) 384.9 (375.7) 409.7 (400.9) Annual 402.9 (399.2) 396.1 (387.1) 398.8 (393.1) latitudes. The surface shows net heating at all latitudes,

Ls,n dominated by SW heating with a strong latitudinal de- DJF Ϫ61.1 (Ϫ53.5) Ϫ55.0 (Ϫ43.0) Ϫ58.1 (Ϫ48.3) pendence. The LW cooling has less latitudinal depen- JJA Ϫ62.3 (Ϫ47.3) Ϫ59.8 (Ϫ50.3) Ϫ61.1 (Ϫ48.8) dence, with maxima in the subtropics (deserts). It is Annual Ϫ62.9 (Ϫ52.0) Ϫ57.5 (Ϫ46.9) Ϫ60.3 (Ϫ49.4) observed how the global balance at TOA is achieved mainly by SW heating at the surface and LW cooling in the atmosphere. These quantities define the basic en- between the surface and the lower troposphere, the NH ergy distribution driving the global circulation, and so shows an amplitude twice as large as the SH (Gupta et we expect the models to simulate them correctly. al. 1999). The model compares resonably well with The zonal pattern of cloud radiative forcing (Figs. ISCCP-FD, although overestimating slightly the ampli- 4d–f) is also well captured by HadGAM1 at TOA, tude of the cycle in the NH. A similar behavior is ex- showing a well-known result from TOA measurements L hibited by s,u (Fig. 3d), although with bigger differ- of SW and LW radiation (Ramanathan et al. 1989; Har- ences in the NH, where HadGAM1 shows a greater rison et al. 1990): clouds produce net cooling of the amplitude of the seasonal cycle. These differences in earth–atmosphere system as a result of two opposite L s,u cause the amplitude of the NH seasonal cycle of effects: SW cooling and LW heating. Differences be- L s,n (Fig. 3f) to be significantly smaller in HadGAM1 tween HadGAM1 and ISCCP-FD are mainly caused by than in ISCCP-FD, although with the correct phase. In an underestimation of the impact of clouds in the SW in the SH, the amplitude and phase are reasonably well HadGAM1, although there are nonnegligible differ- captured by HadGAM1. It should be noted that, in the ences in LW in the tropics and Southern Hemisphere. LW, there is no compensation of errors between down- In the atmosphere, the SW cloud forcing C is small, ward and upward fluxes in the annual mean values. SW showing a small SW heating of the atmosphere due to HadGAM1 underestimates the downward LW radia- clouds, although the magnitude of this has been dis- tion and overestimates the upward LW radiation, which cussed in the literature (Cess et al. 1995; Ramanathan makes the surface LW cooling to be overestimated by et al. 1995; Li et al. 1995; Stephens 1996; Arking 1996). Ϸ18%. Therefore, the net cloud forcing (CNET) is controlled by the LW component (C ). ISCCP-FD data show that c. Energy distribution LW clouds produce LW effective heating (reduced cooling) The derivation of the surface radiation budget from in the tropics and LW cooling in the subtropics and satellite data also allows us to obtain the atmospheric midlatitudes. In the tropics, the LW effective heating is radiation budget as a residual from the TOA and SFC due to the increase of infrared absorption and emission fluxes. Both the atmosphere and the surface budgets at colder temperatures than for clear skies. In the ex- are critical in the cloud feedback problem: the energy tratropics, the mechanism seems to be more subtle: be-

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4731

FIG. 4. Zonal means of the (left) radiation budget and (right) cloud radiative forcing at TOA, within the

atmosphere (ATM) and at the surface (SFC). CSW, CLW, and CTOT are the SW, LW, and total cloud forcing, respectively. Solid lines with symbols are ISCCP-FD results, and nonsolid lines are HadGAM1 simulations. cause of lower PWC than in the tropics, clouds increase edge of the vertical distribution of clouds provided by atmospheric emissivity (particularly in the water vapor current satellite missions carrying active instruments “window” region) and therefore increase the outgoing (Stephens et al. 2002). Finally, as can be observed in the LW radiation. This increase in emissivity seems to off- lower plot, the surface is dominated by a net cooling set the effect of the decrease in the effective emission effect, produced by SW cooling. This cooling is partially temperature owing to the presence of clouds (Rossow offset by LW heating, mainly in the extratropics. This and Zhang 1995). HadGAM1 captures this behavior, seems to confirm the previous argument regarding CLW although the region with net effective heating effect is the atmosphere. In the tropics, the atmosphere is gen- larger, comprising all latitudes from 50°Sto50°N. erally very opaque to LW radiation because of the high Therefore, although clouds have almost no net effect PWC, and therefore clouds have little impact on the globally in the atmosphere (2.4 W mϪ2 from 60°Sto LW flux reaching the surface. 60°N), they may enhance the latitudinal gradient in LW cooling in the tropics, thereby reinforcing the meridio- 4. Comparison against ground measurements nal gradient of the forcing of the atmospheric circulation (Rossow and Zhang 1995; Stephens 2005). The un- For this study, we use all-sky downwelling SW and derstanding of the impact of clouds in the atmospheric LW fluxes at the surface measured at the BSRN sta- radiation budget will be improved with a better knowl- tions listed in Table 2, where for each station, its geo-

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4732 JOURNAL OF CLIMATE VOLUME 21

TABLE 2. List of BSRN stations used in this study. They have been classified in five climate groups according to the climate classification by Trewartha and Horn (1980). BSRN label, longitude, latitude, and period used in the comparison are shown.

Station Label Lon Lat Period Tropical humid Florianopolis (Brazil) Flo Ϫ48.52 Ϫ27.53 Jul 1994–Dec 1999 Ilorin (Nigeria) Ilo 4.57 8.53 Sep 1992–May 1995 Kwajalein (Marshall Islands) Kwa 167.63 8.72 Apr 1992–Jan 2001 Momote (Papua New Guinea) Man 147.43 Ϫ2.05 Oct 1996–Jan 2001 Nauru Island (Pacific Ocean) Nau 166.92 Ϫ0.52 Nov 1998–Jan 2001 Subtropical Bermuda (United Kingdom) Ber Ϫ64.77 32.30 Jan 1992–Jan 2001 Billings, OK (United States) Bil Ϫ97.52 36.60 Jul 1993–Jan 2001 Carpentras (France) Car 5.03 44.05 Sep 1996–May 2000 Goodwin Creek, MS (United States) Gcr Ϫ89.87 34.25 Jan 1995–Jan 2001 Tateno (Japan) Tat 140.13 36.05 Feb 1996–Jan 2001 Temperate Bondville, IL (United States) Bon Ϫ88.37 40.07 Jan 1995–Jan 2001 Lauder (New Zealand) Lau 169.68 Ϫ45.00 Aug 1999–Jan 2001 Lindenberg (Germany) Lin 14.12 52.22 Oct 1994–Jan 2001 Payerne (Switzerland) Pay 6.95 46.82 Oct 1992–Jan 2001 Rock Springs, MD (United States) Psu Ϫ73.93 40.72 Jun 1998–Jan 2001 Polar Barrow, AK (United States) Bar Ϫ156.60 71.32 Jan 1992–Jan 2001 Georg von Neumayer (Antartica) Gvn Ϫ8.25 Ϫ70.65 Apr 1992–Jan 2001 Ny Ålesund, Spitsbergen (Norway) Nya 11.95 78.93 Aug 1992–Jan 2001 South Pole (Antarctica) Spo 0.00 Ϫ90.00 Jan 1992–Jan 2001 Syowa (Antarctica) Syo 39.58 Ϫ69.00 Jan 1994–Jan 2001 Dry Alice Springs (Australia) Asp 133.88 Ϫ23.80 Jan 1995–Jan 2001 Boulder, CO (United States) Bou Ϫ105.00 40.05 Jan 1992–Jan 2001 De Aar (South Africa) Daa 24.00 Ϫ30.67 Jun 2000–Dec 2000 Desert Rock (United States) Dra Ϫ116.02 36.65 Mar 1998–Jan 2001 Fort Peck, MT (United States) Fpe Ϫ105.10 48.32 Jan 1995–Jan 2001 Regina (Canada) Reg Ϫ104.72 50.20 Jan 1995–Nov 1999 Solar Village, Riyadh (Saudi Arabia) Sov 46.42 24.92 Sep 1998–Jan 2001 Tamanrasset (Algeria) Tam 5.52 22.78 Mar 2000–Jan 2001 graphical position and period used in the comparison (e.g., gvn, nya, spo, syo). They are the only ones that are shown. Although the BSRN fluxes are provided at show negative biases (HadGAM1 underestimating a very high temporal resolution (1 min), we use Ss,d). The bias (standard deviation) for all the monthly monthly means as the standard model diagnostics have means included in the comparison (1810 samples) is that time resolution. The BSRN monthly means have 16.2 (28.0) W mϪ2. been computed by first obtaining a mean monthly di- In the case of the Antarctic sites, the agreement is urnal cycle to avoid any bias due to missing data. remarkable, with a negligible bias, and a relatively low Figure 5 shows the comparisons of observed and standard deviation despite the high values of surface computed SW downward fluxes at the selected sites. In insolation. This is not caused by persistent cloudless addition to the scatterplot, each plot shows the bias conditions in this site, as the seasonal cycle of simulated (model minus observed) and standard deviation of the cloud amount over the South Pole shows values be- differences in watts per meters squared, and number of tween 20% and 60% over the months with sunlight. points used in the comparison. Generally, HadGAM1 Therefore, the agreement is achieved despite the non- shows a tendency to overestimate Ss,d, which is consis- negligible cloud amount. This is consistent with the sea- tent with our previous comparisons against ISCCP sonal cycle of clear-sky insolation for Georg von Neu- fluxes, implying that the ISCCP-FD database is a good mayer, Antartica (Wild et al. 2006), which shows maxi- test for the performance of the incoming solar radiation mum values around 50 W mϪ2 higher than the ones for in climate model simulations. The overestimation in Ss,d all-sky presented here. does not seem to be regionally dependent. An interest- Figure 6 is similar to Fig. 5, but for Ls,d. The overall ing exception are those stations located in polar regions bias (standard deviation) of all the monthly means

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4733

FIG. 5. Scatterplots of monthly mean SW downward radiation at BSRN sites against HadGAM1. See Table 2 for details on the sites.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4734 JOURNAL OF CLIMATE VOLUME 21

FIG. 6. As in Fig. 5, but for downwelling LW radiation.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4735

plot of long-term annual means of Ss,d against Ls,d as observed at BSRN sites and calculated by HadGAM1.

The segments show the distance in the (Ls,d,Ss,d) space between simulations and observations. It is clearly observed that there is a privileged direction, with most of the segments pointing toward the top of the plot. This

means that the positive bias in Ss,d dominates over the low or positive biases in Ls,d. In Fig. 8 we compare the mean annual cycles of Ss,d at the selected locations as observed by BSRN (solid), simulated by HadGAM1 (dashed), and derived by ISCCP-FD (dot–dashed). Note that the scales are different, being adapted to the amplitude of the annual cycle. The annual cycle of surface insolation is mainly controlled by orbital geometry, and is modulated by cloud, aerosols, and water vapor variations. Stations located in the NH midlatitudes show a maximum insolation around June, and a minimum around December, with a typical amplitude of ϳ200WmϪ2. Obviously, a similar cycle is observed in the SH midlatitudes (e.g., FIG. 7. Scatterplot of annual means of SW and LW downward Lauder, New Zealand), but with opposite phase. At radiation as observed at BSRN sites and calculated by Had- polar latitudes this cycle is exaggerated, with ampli- GAM1. Observational and model points are connected by solid tudes of more than 200 W mϪ2, and a few months with line segments. The end of the segment with the number corre- no insolation during the polar winter. At tropical sites sponds to the observed values; the end with the diamond represents the values computed by HadGAM1. The legend shows the (e.g., Ilorin, Nigeria; Kwajalein, Marshall Islands; Mo- labels of BSRN stations, as listed in Table 2. mote, Papua New Guinea; and Nauru Island), the annual cycle has an amplitude smaller than 100 W mϪ2, and contains a semiannual harmonic, as over these sites (1783 samples) included in the comparison is Ϫ6.0 the solar declination coincides with their latitude (over- (19.6) W mϪ2. This is again consistent with the negative head sun) twice a year. The modulation of the mean bias with respect to ISCCP-FD shown in Table 1 for the annual cycle of Ss,d by clouds at tropical sites can be globe. Although HadGAM1 shows a negative bias for very strong, and indeed can dominate the orbital effect. most of the sites, this behavior is not as general as the This is clearly observed in Ilorin, where a minimum of positive bias shown in Ss,d. There are 12 sites where Ss,d is observed in August, when the insolation at TOA HadGAM1 shows a positive bias, although for 6 of is maximum. This is caused by the strong annual cycle them the bias is less than 2 W mϪ2. The comparisons at in cloud cover, which is positively correlated with the polar sites do not show any distinct behavior, being variation of TOA insolation. Overall, HadGAM1 cap- similar to other sites at lower latitudes. This contrasts tures the shape of the annual cycle of Ss,d quite well at with comparisons of Ls,d against surface measurements all the sites, although it shows a general tendency to in previous generations of models, which showed an overestimate Ss,d, mainly due to lack of cloud. At mid- underestimation of downwelling LW radiation at high latitude sites, the general overestimation is usually latitudes, and no biases or even slight overestimations present throughout the year, being more prominent at low latitudes (Wild et al. 2001). This behavior was during summer months. As was highlighted in Fig. 5, also found for clear-sky fluxes (Allan 2000). The global polar sites show a remarkably good agreement, captur- mean value from HadGAM1 is 339 W mϪ2, greater ing the mean annual cycle with great accuracy. The than the value of 333 W mϪ2 from HadAM3, and closer annual cycles from ISCCP-FD are close to the obser- to the best estimate of 344 W mϪ2 given by Wild et al. vations by BSRN, and generally are in better agree-

(2001). This increase in global mean Ls,d with respect to ment with the observations than the simulations. This HadAM3 is mainly due to improvements in the simu- gives confidence on the use of the ISCCP-FD short- lation of low cloud over the oceans (Martin et al. 2006). wave fluxes for model evaluation.

From these general comparisons we conclude that, Similarly, Fig. 9 shows the mean annual cycles of Ls,d overall, the simulation of Ls,d is better than that of Ss,d. from BSRN (solid), HadGAM1 (dashed), and ISCCP- This can also be seen in Fig. 7, which shows a scatter- FD (dot–dashed). Midlatitude sites show a mean an-

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4736 JOURNAL OF CLIMATE VOLUME 21

FIG. 8. Comparison of mean annual cycle of SW downward flux as observed at BSRN sites (solid) and computed by HadGAM1 (dashed), and ISCCP-FD (dot–dashed). See Table 2 for details about the sites.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4737

FIG. 9. As in Fig. 8, but for downwelling LW radiation.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4738 JOURNAL OF CLIMATE VOLUME 21 nual cycle with amplitude around 100 W mϪ2, a maxi- GAM1 capturing very well the negative anomalies in mum in the summer months and a minimum in winter, the tropical central Pacific in Los Niños of 1986/87, lagging by approximately 1 month the mean annual 1991/92, and 1997/98. The opposite effect of La Niña cycle in Ss,d. The polar regions follow a similar pattern, can be observed over (roughly) the same areas in 1988/ but with a smaller amplitude, whereas the tropical sites 89 and 1999/2000, with a high anomaly in Ss,d. The im- show no annual cycle in Ls,d. The general characteristics pact over the tropical eastern Pacific is less strong, and of the annual cycle are well captured by HadGAM1, only the strong 1997/98 event shows a high anomaly in although it shows a tendency to produce smaller fluxes. that region. This anomaly extends farther west in Had- Particularly severe are the cases of sites in desert re- GAM1 than in ISCCP. The impact of El Niño/La Niña gions (e.g., Alice Springs, Australia; Solar Village, Ri- events is also clearly visible in Ls,d (Figs. 10c,d), show- yadh, Saudi Arabia; and Tamanrasset, Algeria). The ing noticeable differences. The downward longwave ra- longwave fluxes from ISCCP do not show a better diation at the surface is dominated by the emission agreement with BSRN fluxes than the model simula- coming from the lower levels of the troposphere, and tions. HadGAM1 tends to underestimate Ls,d, whereas indeed this is correlated with surface temperature ISCCP-FD shows an overestimation of the same mag- (Gupta et al. 1992). Therefore, Ls,d anomalies are cor- nitude. The bias (standard deviation) from all the related with the SST anomalies (not shown), with cloud monthly means analyzed is 6.3 (28.8) W mϪ2 for variations playing a secondary role. During Los Niños ISCCP-FD, as compared to Ϫ6.0 (19.6) W mϪ2 for of 1986/87, 1991/92, and 1997/98, the central and east-

HadGAM1. This means that the usability of ISCCP-FD ern Pacific shows a positive anomaly in Ls,d due to the longwave fluxes for model evaluation is less than that of warm anomaly in SST over the same regions. During shortwave fluxes. La Niña events, the cold anomalies are correlated with

smaller Ls,d. 5. Interannual variability Figure 11 shows the spatial patterns of the February climatology (1984–2000) of surface radiative fluxes over In this section we focus our attention on the interan- the tropical Pacific (left-hand side column), as well as nual variability of surface incoming radiation over the the anomalies for the strong 1998 El Niño (right-hand tropical Pacific (TP; 10°S–10°N), as El Niño is the main side column). The shortwave incoming radiation is mode of variability of that region and provides a useful shown in Figs. 1a–d, and it is observed that HadGAM1 means to test cloud–climate interactions in climate reproduces reasonably well the climatological pattern, models (Lu et al. 2004). Here we analyze the impact of with minima over deep convective regions and in the El Niño on interannual variability from a surface radia- stratocumulus area off the coast of Chile. The main tion perspective, and show HadGAM1 performance to differences are the excess of radiation reaching the capture the impact of El Niño events on the surface ground in several regions, such as the western part of radiation budget. Because SSTs are prescribed in atmo- the Maritime Continent and Australia. A good simula- sphere-only experiments, we look at the response of the tion of the mean state does not necessarily imply a good atmosphere to changes in SSTs under El Niño condi- simulation of the variability. Figures 1b,d show that the tions. The weakening of the gradient in SST between simulation of the anomaly pattern of Ss,d is also very the tropical eastern and western Pacific under El Niño well captured by HadGAM1, with a dipolar anomaly in conditions causes the Walker circulation to weaken, or the equatorial Pacific. A high anomaly is observed over even collapse in strong events (Cess et al. 2001). This the warm pool, where a reduction in the convective collapse of the Walker circulation is well captured by intensity causes cloud radiative impact to decrease and

HadGAM1 (Johns et al. 2006). Clouds respond to these hence increase in Ss,d. By contrast, a low anomaly is changes in the tropical circulation, with higher-than- observed in the central equatorial Pacific, where con- average cloud amount in the tropical eastern Pacific vection intensifies, produces more cloud and decreases and lower-than-average amounts in the tropical west- the amount of solar radiation reaching the ground. The ern Pacific (Cess et al. 2001). longwave counterpart is shown in Figs. 11e–h. To first

Figure 10 shows Hovmoeller plots of Ss,d and Ls,d for order, the climatological pattern of Ls,d shows a picture the tropical Pacific as derived from satellite data and similar to that of Ss,d, but with maxima where we had simulated by HadGAM1. It shows the time series of the minima before and vice versa. The Ls,d anomalies also anomalies with respect to the mean annual cycle of the show the correlation with changes in cloud and water overlapping months of the ISCCP database and Had- vapor, with a low anomaly in Ls,d over the warm pool, GAM 1 run. The anomaly patterns of Ss,d (Figs. 10a,b) and a high anomaly over the equatorial central Pacific, in ISCCP and HadGAM1 are very similar, with Had- extending eastward to the coast of South America.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4739

FIG. 10. Hovmoeller plots of the tropical Pacific (10°S, 10°N) as derived from satellite data and simulated

by HadGAM1. (a) Ss, d from ISCCP-FD, (b) Ss, d from HadGAM1, (c) Ls, d from ISCCP-FD, and (d) Ls,d from HadGAM1.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

Fig 10 live 4/C 4740 JOURNAL OF CLIMATE VOLUME 21

FIG. 11. Surface radiative fluxes climatology for February 1984–2000 and anomalies for February 1998 over the tropical Pacific as derived from satellite data and simulated by HadGAM1. (a) Ss, d climatology from HadGAM1, (b) February 1998 Ss, d anomaly from HadGAM1, (c) Ss, d climatology from ISCCP-FD, (d) February 1998 Ss, d anomaly from ISCCP-FD, (e) Ls, d climatology from Had- GAM1, (f) February 1998 Ls, d anomaly from HadGAM1, (g) Ls, d climatology from ISCCP-FD, and (h) February 1998 Ls, d anomaly from ISCCP-FD.

6. Land surface albedo tation of the surface albedo with an independent source of information, we have chosen version 4 of the As shown in section 3, comparisons of surface albedo MODIS 16-day 0.05° global albedo product (MOD43C1; between HadGAM1 and ISCCP-FD highlighted some Schaaf et al. 2002). The MOD43C1 is a level 3 climate differences. However, as ISCCP-FD surface albedo is modeling grid product that provides measures of based on the GISS GCM surface albedo, modified with earth’s “white-sky” (only diffuse radiation) and “black- a procedure similar to that explained in Zhang et al. sky” (only direct radiation) albedo in 16-day averages (1995), to some extent we are comparing the albedos of with a 0.05° spatial resolution. It provides spectral al- two GCMs. To obtain a better picture of the represen- bedo in seven MODIS bands, and in three broad bands:

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

Fig 11 live 4/C 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4741

FIG. 12. Regional means of land surface albedo for January and July for HadGAM1 and MODIS. Maps for HadGAM1 correspond to 20-yr averages, whereas MODIS are 2-yr averages. White areas are water surfaces or missing data points. visible, near infrared, and total solar spectrum. As Had- from observations of 14 land cover classes at 1-km reso- GAM1 does not have any spectral dependence in the lution derived from the Advanced Very High Resolu- specification of the land surface albedo, we only use the tion Radiometer (AVHRR) data (Hansen et al. 2000), albedo from MODIS in the whole solar spectrum. The and mapped onto the MOSES 2 surface types (Essery MODIS white-sky albedo is generally a good approxi- et al. 2003). A leaf area index for each vegetation tile is mation to the monthly-average albedo (Wang et al. read from maps based on those used by the second 2004), and therefore we use this product to compare Simple Biosphere Model (Sellers et al. 1996). A con- directly with the model albedo. stant land surface albedo for all spectral bands is then HadGAM1 uses the Met Office Surface Exchange computed for each grid box using the surface and leaf Scheme version 2 (MOSES 2; Essery et al. 2001, 2003), area index distributions, and the simulated snow which includes a tiled representation of heterogeneous amount (Essery et al. 2001). surfaces. It represents each grid box as a mixture of five Figure 12 presents comparisons of the regional vegetation types (i.e., broadleaf trees, needleleaf trees, means of surface albedo from HadGAM1 with white- temperate grass, tropical grass, and shrubs) and four sky albedo from MODIS and shows maps for January nonvegetated surface types (i.e., urban, inland water, and July of the 20-yr averages for HadGAM1 and 2-yr soil, and ice). Vegetation distributions are obtained averages for MODIS. Similar comparisons have been

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

Fig 12 live 4/C 4742 JOURNAL OF CLIMATE VOLUME 21 made with each of the 2 yr used in the MODIS average MODIS. Although the snow cover is responsible for (not shown), and significant differences with the 2-yr the differences in those regions, the presence of snow is mean have only been found in a few regions. The ac- negligible in many regions of France, the United King- curacy of the white-sky albedo MODIS retrieval is re- dom, and the Iberian Peninsula, which indicates an ported to be within 0.02 in magnitude (Wang et al. overestimation in the model snow-free surface albedo 2004); the color scale in the maps showing the differ- over those regions. ences has been selected to show in light green those Figure 14a shows the seasonal cycle of the surface regions that lie within that interval. It is observed that albedo averaged over the Iberian Peninsula [(38°N, HadGAM1 overestimates the surface albedo over 8°W) to (43°N, 1°W)]. The fact that the snow-cover deserts in southern Africa, Australia, and South impact on the surface albedo over this area is negligible America. This overestimation is also evident in the is shown by the fact that the value of MOD43C1 is very deserts of the NH, although with some differences. The close to the spatially complete product. It is obvious Sahara Desert and the Arabian Peninsula show a mix- that seasonal cycle of the surface albedo is not well ture of positive and negative biases, caused by a higher represented in HadGAM1 over that region. The mean spatial variability in the albedo as obtained by MODIS annual value is slightly too high, whereas the amplitude than in HadGAM1. HadGAM1 shows a very small spa- of the oscillation is too small. In addition, the phases of tial variability, with albedos greater that 0.35 every- both oscillations do not coincide: HadGAM1 reaches where in North Africa and the Arabian Peninsula, the maximum value in April/May, and the minimum in which is not the case in MODIS. Another important September, whereas MODIS peaks in August, and difference in Africa is the Sahel region, where Had- reaches its minimum value in December/January. The GAM1 underestimates the surface albedo. This is representation of the seasonal cycle of ISCCP-FD is caused by a lack of a transition region between the also very poor, with opposite behavior to MODIS. Sahara Desert and central Africa in HadGAM1. While Figure 14b shows a similar plot for the Southeast MODIS shows values of 0.25 to 0.35 that represent the United States [(31°N, 95°W) to (36°N, 85°W)], with a transition between the higher values over the Sahara negligible influence of snow during the winter months. (typically Ͼ0.35) and the lower values in central Africa This plot shares some features with the Iberian Penin- (typically Ͻ0.20), HadGAM1 does not show that tran- sula, in the sense that HadGAM1 shows a consistent sition, going quite abruptly from the higher values to negative bias, and a poor representation of the seasonal the lower ones. Finally, there is also a very high bias cycle. The errors are particularly relevant during the over midlatitude landmasses in the NH in January. This winter months, when MODIS measures the lowest al- overestimation is not present in July, which indicates bedos, and HadGAM1 shows a quasi-constant albedo the existence of seasonal-dependent errors. throughout all the year, close to the value measured by To explore these errors in more detail, Fig. 13 shows MODIS in the summer months. Again, the representa- the regional surface albedo over Europe in January. It tion of the seasonal cycle of ISCCP-FD is also very shows the model monthly means averaged over 20 yr, poor as compared to MODIS. as well as two MODIS products: the MOD43C1 prod- Figure 14c shows the same information as Fig. 14a, uct already used in Fig. 12 and the spatially complete but for eastern Europe [(50°N, 25°E) to (55°N, 30°E)]. albedo. The spatially complete albedo is a “value- From May to October, HadGAM1 follows quite well added” product, which uses an ecosystem-dependent the seasonal cycle of surface albedo shown by MODIS, temporal interpolation technique to fill missing or sea- although showing greater values. This result is also ob- sonally snow-covered data in the official MOD43B3 tained in other regions in the NH midlatitudes, such as product (from which MOD43C1 is obtained; Moody central Asia (not shown). The impact of the snow cover et al. 2005). This figure also shows the HadGAM1 in the surface albedo is evident between November and monthly snow amount (kilograms per meter squared) March. This can be seen in the MOD43C1 albedo in averaged over 20 yr (Fig. 13b). As can be seen, the January, which shows a value of ϳ0.28 as compared to presence of snow in HadGAM1 increases the albedo ϳ0.11 given by MODIS-SC. The differences from Had- over central and northern Europe, and explains the dif- GAM1 with respect to MOD43C1 in January are large, ferences over the areas that are not actually covered by with HadGAM1 showing a surface albedo that is too snow in the MODIS dataset. By comparing Figs. high. This may indicate an excess in the amount of snow 13b,c,e, it can be deduced that the area covered by snow simulated, but also deficiencies in the snow albedo pa- is much wider in HadGAM1 than in MODIS, although rameterization. However, MODIS data are less reliable the difference might not be significant as we are com- for snow-covered regions, particularly for nonforested paring a 20-yr climatology against a 2-yr average from regions (Jin et al. 2002). This, and the fact that ISCCP

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4743

FIG. 13. Surface albedo over Europe in January. (a) HadGAM1 monthly albedo aver aged over 20 yr, (b) HadGAM1 monthly snow amount (kg mϪ2) averaged over 20 yr, (c) white-sky albedo from MODIS MOD43C1 for the first 16 days of 2001 and 2002, (d) HadGAM1 minus MODIS MOD43C1, (e) white-sky albedo from MODIS spatially complete product for the first 16 days of 2000–04, and (f) HadGAM1 minus MODIS spatially complete. White areas are water surfaces or missing data points. shows similar values to HadGAM1 in January, may 7. Global dimming indicate that MODIS is underestimating the surface albedo of snow. A comprehensive intercomparison of the Numerous studies have noted an observed long-term different surface albedo databases is needed in order to trend in incoming surface shortwave radiation over the understand their quality and differences. late twentieth century, a phenomenon known as “glob-

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

Fig 13 live 4/C 4744 JOURNAL OF CLIMATE VOLUME 21

2002) and those from the Global Energy Balance Ar- chive (GEBA) of ETH Zurich (Gilgen et al. 1998). The trend originally observed was negative (hence, the name given to the phenomenon), and an increase in the concentrations of anthropogenic aerosols, acting via both direct and indirect effects on clouds, was ascribed as the most probable cause (Stanhill and Cohen 2001). The amount of reduction in global-mean surface solar radiation varies between different studies and the time periods chosen. For example, Stanhill and Cohen (2001) suggest a reduction of 20 W mϪ2 over the period 1958–92, whereas Liepert (2002) suggests a decrease of 7WmϪ2 for the period 1961–90. More recent observations (Wild et al. 2005) have since shown a reversal of the trend in some areas of the world, beginning sometime in the late 1980s. This reversal is attributed to a decrease in anthropogenic aerosol emissions caused by the implementation of air quality legislation in areas such as Europe and the United States and by the collapse of the former Soviet Union, and to a recovery from the emissions caused by major volcanic eruptions such as El Chichón and Pinatubo. This is corroborated by the study of Streets et al. (2006), which analyzes the changes in anthropogenic aerosol emissions over the period from 1980 to 2000, and concludes that these changes are likely to be the cause of the transition from dimming to brightening. The HadGAM1 simulations analyzed in the previous sections of this paper are not ideal for evaluating the model’s simulation of global dimming/brightening as they only cover the 20-yr period up to 2000. We therefore use decadal-mean data from simulations using the full atmosphere–ocean coupled model HadGEM1 (Johns et al. 2006). These simulations cover the period 1860–2010 and include the effects of anthropogenic changes in greenhouse gases, aerosols and land use, as well as natural changes from volcanic eruptions and solar variations. Time-varying historical emissions of anthropogenic aerosols (or aerosol precursors) were used: the emission data of Smith et al. (2004) were used FIG. 14. Seasonal cycle of the surface albedo over three differ- for SO , and those compiled by T. Nozawa (National ent regions: (a) Iberian Peninsula from (38°N, 8°W) to (43°N, 2 1°W); (b) Southeast United States from (31°N, 95°W) to (36°N, Institute for Environmental Studies, Japan, 2003, per- 85°W); (c) eastern Europe from (50°N, 25°E) to (55°N, 30°E). sonal communication) for biomass-burning and fossil- Solid line shows the model results, whereas dashed–dotted line fuel black carbon. Figure 15a shows the difference in shows the observations from the MODIS spatially complete incoming surface shortwave radiation between the dataset. The asterisks are the values for January and July from 1950s and the 1980s. This shows dimming of up to Ϫ10 MOD43C1. The dashed line shows ISCCP-FD. WmϪ2 over Europe and the United States, in reason- able agreement with Liepert (2002). There are also ar- al dimming” (e.g., Stanhill and Cohen 2001; Liepert eas with much larger decreases, with values of Ϫ15 to 2002; Wild et al. 2004). These studies have used a va- Ϫ20WmϪ2 over parts of Africa, India, and China. riety of surface-based radiometer data, such as those Figure 15b shows the subsequent change between the derived from the world radiation network set up for the 1980s and the 2000s. This shows that, in agreement with International Geophysical Year (Stanhill and Moreshet Wild et al. (2005), areas such as Europe and the United

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4745

FIG. 15. Decadal means of difference in incoming surface shortwave radiation as simulated by HadGEM1. Difference between (a) the 1950s and the 1980s and (b) the 1980s and the 2000s. Units are W mϪ2.

States show an increase in surface shortwave radiation. emissions in Africa and industrial emissions over However, both the dimming between the 1950s and the Southeast Asia. 1980s (Fig. 15a) and the brightening between the 1980s It is apparent, however, that in global-mean terms and the 2000s (Fig. 15b) are by no means global, with the amount of dimming shown in Fig. 15a is markedly strong spatial heterogeneity evident in both cases. In- less, at Ϫ2WmϪ2, than the values given by the obser- deed the brightening over parts of Europe and the vational studies of Liepert (2002) and Stanhill and United States shown in Fig. 15b is more than offset by Cohen (2001). This result is similar to that from some the dimming from increases in biomass-burning aerosol other modeling studies of global dimming, such as

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC

Fig 15 live 4/C 4746 JOURNAL OF CLIMATE VOLUME 21

Liepert et al. (2004) with the ECHAM4 model, who Also important is the positive bias over midlatitude obtained a reduction of Ϫ3.8WmϪ2 over the century landmasses in the NH in January. A more detailed from the 1880s to the 1980s. While there are undoubt- analysis of the land surface albedo for different re- edly shortcomings in the models and their treatment of gions has shown that the amplitude and phase of the aerosols and their effects, these model results are more seasonal cycle is not well represented in HadGAM1, in line with the conclusions of Alpert et al. (2005). Their although a more extensive validation needs to be car- examination of GEBA data over the period 1964–89 ried out. Our results also suggest that a comprehen- suggests that global dimming is a highly variable phe- sive intercomparison of the different surface albedo nomenon, dominated by the pollution generated by in- databases is needed in order to understand their creasing urbanization. A lower value for the global- quality and differences. mean change in surface shortwave radiation is also sup- • Global dimming: in global-mean terms the amount of ported by the study of Pinker et al. (2005), who use dimming simulated by HadGEM1 is less than the val- ISCCP data to analyze the change in surface solar ra- ues given by the studies of Liepert (2002) and Stanhill diation over the whole globe for the period 1983–2001. and Cohen (2001). The model results are more in line They conclude that the tendencies in surface solar ra- with the conclusions of Alpert et al. (2005), suggest- diation observed at a global scale are much smaller than ing that global dimming is far from being a uniform those obtained from ground-based observations. phenomenon across the globe.

Acknowledgments. This work was supported by the 8. Conclusions UK Department for Environment, Food, and Rural Af- fairs under Contract PECD 7/12/37. The ISCCP-FD The surface radiation budget from HadGAM1 in a data were obtained from the International Satellite 20-yr present-day climate simulation has been com- Cloud Climatology Project Web site (http:// pared with the surface radiation budget derived from isccp.giss.nasa.gov) maintained by the ISCCP research satellite data and measured with ground stations, and group at the NASA Goddard Institute for Space Stud- we summarize the main conclusions. ies. The World Radiation Monitoring Center is ac-

• HadGAM1 generally overestimates Ss, d over land- knowledged for the release of BSRN data. The MODIS masses. The bias (standard deviation) with respect to MOD43C1 data are distributed by the Land Processes ground measurements is 17.2 (28.6) W mϪ2. Had- Distributed Active Archive Center (LP DAAC), lo-

GAM1 simulates Ss, d very well over the polar re- cated at the U.S. Geological Survey (USGS) Center for gions. Although regional differences of Ss, d are cor- Earth Resources Observation and Science (EROS) related with errors in cloud amount or cloud optical Web site (http://LPDAAC.usgs.gov). We also thank the thickness, this overestimation is also present under comments of two anonymous reviewers. clear skies, consistent with a low aerosol optical thickness compared with observations. This contrasts REFERENCES with the negligible bias shown by HadAM3 in clear- Allan, R. P., 2000: Evaluation of simulated clear-sky longwave sky Ss, d (Wild et al. 2006). Certain missing aerosol radiation using ground-based observations. J. Climate, 13, species in HadGAM1, and the fact that HadAM3 1951–1964. uses an aerosol climatology (Cusack et al. 1998), Alpert, P., P. Kishcha, Y. J. Kaufman, and R. Schwarzbard, 2005: means HadAM3 provides a more accurate (although Global dimming or local dimming?: Effect of urbanization on sunlight availability. Geophys. Res. Lett., 32, L17802, less physically based) representation of the climato- doi:10.1029/2005GL023320. logical radiative impact of aerosols. Arking, A., 1996: Absorption of solar energy in the atmosphere: • HadGAM1 tends to underestimate Ls, d. The bias Discrepancy between model and observations. Science, 273, (standard deviation) compared with ground measure- 779–782. ments is Ϫ6.0 (19.6) W mϪ2. The global mean L is Barker, H. W., and Z. Li, 1995: Improved simulation of clear-sky s, d shortwave radiative transfer in the CCC-GCM. J. Climate, 8, closer to observations than HadAM3. This is mainly 2213–2223. due to improvements in the simulation of low cloud. Bolle, H.-J., and Coauthors, 2006: Mediterranean Land-Surface • HadGAM1 overestimates the surface albedo over Processes Assessed from Space. Springer-Verlag, 760 pp. deserts in South Africa, Australia, and South Bony, S., J.-L. Dufresne, H. Le Treut, J.-J. Morcrette, and C. A. America. This overestimation is also evident in the Senior, 2004: On dynamic and thermodynamic components of cloud changes. Climate Dyn., 22, 71–86, doi:10.1007/ deserts of the NH, although mixed with underestima- s00382-003-0369-6. tions over some regions. HadGAM1 severely under- Cess, R. D., and Coauthors, 1995: Absorption of solar-radiation estimates the surface albedo over the Sahel region. by clouds: Observations versus models. Science, 267, 496–499.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 15 SEPTEMBER 2008 BODAS-SALCEDO ET AL. 4747

——, M. Zhang, P.-H. Wang, and B. A. Wielicki, 2001: Cloud mate model HadGEM1: Evaluation of coupled simulations. structure anomalies over the tropical Pacific during the 1997/ J. Climate, 19, 1327–1353. 98 El Niño. Geophys. Res. Lett., 28, 4547–4550. Kristjánsson, J. E., J. M. Edwards, and D. L. Mitchell, 2000: Im- Chevallier, F., and J.-J. Morcrette, 2000: Comparison of model pact of a new scheme for optical properties of ice crystals on fluxes with surface and top-of-the-atmosphere observations. climates of two GCM’s. J. Geophys. Res., 105 (D8), 10 063– Mon. Wea. Rev., 128, 3839–3852. 10 079. Clough, S. A., M. J. Iacono, and J.-L. Moncet, 1992: Line-by-line Li, Z., 1995: Intercomparison between two satellite-based prod- calculations of atmospheric fluxes and cooling rates: Appli- ucts of net surface shortwave radiation. J. Geophys. Res., 100, cation to water vapor. J. Geophys. Res., 97 (D14), 15 761– 3221–3232. 15 785. ——, and H. G. Leighton, 1993: Global climatologies of solar ra- Cusack, S., A. Slingo, J. M. Edwards, and M. Wild, 1998: The diation budgets at the surface and in the atmosphere from 5 radiative impact of a simple aerosol climatology on the years of ERBE data. J. Geophys. Res., 98, 4919–4930. Hadley Centre Atmospheric GCM. Quart. J. Roy. Meteor. ——, H. W. Barker, and L. Moreau, 1995: The variable effect of Soc., 124, 2517–2526, doi:10.1002/qj.49712455117. clouds on atmospheric absorption of solar-radiation. Nature, Darnell, W. L., W. F. Staylor, S. K. Gupta, N. A. Ritchey, and 376, 486–490. A. C. Wilber, 1992: Seasonal variation of surface radiation ——, L. Moreau, and A. Arking, 1997: On solar energy disposi- budget derived from International Satellite Cloud Climatol- tion: A perspective from observation and modeling. Bull. ogy Project C1 data. J. Geophys. Res., 97, 15 741–15 760. Amer. Meteor. Soc., 78, 53–70. Edwards, J. M., and A. Slingo, 1996: Studies with a flexible new Liepert, B. G., 2002: Observed reductions of surface solar radia- radiation code. I: Choosing a configuration for a large-scale tion at sites in the United States and worldwide from 1961 model. Quart. J. Roy. Meteor. Soc., 122, 689–720. to 1990. Geophys. Res. Lett., 29, 1421, doi:10.1029/ Essery, R. L. H., M. Best, and P. Cox, 2001: MOSES 2.2 technical 2002GL014910. documentation. Hadley Centre Tech. Note 30, Hadley Cen- ——, J. Feichter, U. Lohmann, and E. Roeckner, 2004: Can aero- tre, Met Office, Fitzroy Road, Exeter, United Kingdom, sols spin down the water cycle in a warmer and moister 30 pp. world? Geophys. Res. Lett., 31, L06207, doi:10.1029/ ——, M. J. Best, R. A. Betts, P. M. Cox, and C. M. Taylor, 2003: 2003GL019060. Explicit representation of subgrid heterogeneity in a GCM Lu, R. Y., B. W. Dong, R. D. Cess, and G. L. Potter, 2004: The land-surface scheme. J. Hydrometeor., 4, 530–543. 1997/1998 El Niño: A test for climate models. Geophys. Res. Lett., L12216, doi:10.1029/2004GL019956. Garrat, J. R., 1994: Incoming shortwave fluxes at the surface—A 31, comparison of GCM results with observations. J. Climate, 7, Martin, G. M., M. A. Ringer, V. D. Pope, A. Jones, C. Dearden, 72–80. and T. J. Hinton, 2006: The physical properties of the atmosphere in the new Hadley Centre Global Environmental Gates, W. L., and Coauthors, 1999: An overview of the results of Model (HadGEM1). Part I: Model description and global the Atmospheric Model Intercomparison Project (AMIP I). climatology. J. Climate, 19, 1274–1301. Bull. Amer. Meteor. Soc., 80, 29–55. Moody, E. G., M. D. King, S. Platnick, C. B. Schaaf, and F. Gao, Gilgen, H., M. Wild, and A. Ohmura, 1998: Means and trends of 2005: Spatially complete global spectral surface albedos: shortwave irradiance data at the surface estimated from glob- Value-added datasets derived from Terra MODIS land prod- al energy balance archive data. J. Climate, 11, 2042–2061. ucts. IEEE Trans. Geosci. Remote Sens., 43, 144–158, Gleckler, P. J., 2005: Surface energy balance errors in AGCMs: doi:10.1109/TGRS.2004.838359. Implications for ocean-atmosphere model coupling. Geo- Ohmura, A., and Coauthors, 1998: Baseline Surface Radiation phys. Res. Lett., 32, L15708, doi:10.1029/2005GL023061. Network (BSRN/WCRP): New precision radiometry for cli- Gupta, S. K., W. L. Darnell, and A. C. Wilber, 1992: A param- mate research. Bull. Amer. Meteor. Soc., 79, 2115–2136. eterization for longwave surface radiation from satellite data: Pinker, R. T., and I. Laszlo, 1992: Modeling surface solar irradi- J. Appl. Meteor., Recent improvements. 31, 1361–1367. ance for satellite applications on a global scale. J. Appl. Me- ——, N. A. Ritchey, A. C. Wilber, C. H. Whitlock, G. G. Gibson, teor., 31, 194–211. and P. W. Stackhouse, 1999: A climatology of surface radia- ——, B. Zhang, and E. G. Dutton, 2005: Do satellites detect tion budget derived from satellite data. J. Climate, 12, 2691– trends in surface solar radiation? Science, 308, 850–854. 2710. Ramanathan, V., 1987: The role of Earth Radiation Budget stud- Hansen, J., M. Sato, and R. Reudy, 1997: Radiative forcing and ies in climate and general circulation research. J. Geophys. climate response. J. Geophys. Res., 102 (D6), 6831–6864. Res., 92, 4075–4095. Hansen, M. C., R. S. Defries, J. R. G. Townshend, and R. Sohl- ——, R. D. Cess, E. F. Harrison, P. Minnis, B. R. Barkstrom, E. berg, 2000: Global land cover classification at 1 km spatial Ahmad, and D. Hartmann, 1989: Cloud radiative forcing and resolution using a classification tree approach. Int. J. Remote climate: Results from the Earth Radiation Budget Experi- Sens., 21, 1331–1364, doi:10.1080/014311600210209. ment. Science, 243, 57–63. Harrison, E. P., P. Minnis, B. R. Barkstrom, V. Ramanathan, ——, B. Subasilar, G. J. Zhang, W. Conant, R. D. Cess, J. T. R. D. Cess, and G. G. Gibson, 1990: Seasonal variation of Kiehl, H. Grassl, and L. Shi, 1995: Warm pool heat-budget cloud radiative forcing derived from the Earth Radiation and shortwave cloud forcing: A missing physics. Science, 267, Budget Experiment. J. Geophys. Res., 95, 18 687–18 703. 499–503. Jin, Y., C. B. Schaaf, F. Gao, X. Li, A. H. Strahler, X. Zeng, and Randel, D. L., T. H. Vonder Haar, M. A. Ringerud, G. L. R. E. Dickinson, 2002: How does snow impact the albedo of Stephens, T. J. Greenwald, and C. L. Combs, 1996: A new vegetated land surfaces as analyzed with MODIS data? Geo- global water vapor dataset. Bull. Amer. Meteor. Soc., 77, phys. Res. Lett., 29, 1374, doi:10.1029/2001GL014132. 1233–1246. Johns, T. C., and Coauthors, 2006: The new Hadley Centre cli- Raschke, E., A. Ohmura, W. B. Rossow, B. E. Carlson, Y.-C.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC 4748 JOURNAL OF CLIMATE VOLUME 21

Zhang, C. Stubenrauch, M. Kottek, and M. Wild, 2005: Cloud ——, and Coauthors, 2002: The CloudSat mission and the A- effects on the radiation budget based on ISCCP data (1991 to Train. Bull. Amer. Meteor. Soc., 83, 1771–1790. 1995). Int. J. Climatol., 25, 1103–1125, doi:10.1002/joc.1157. Streets, D. G., Y. Wu, and M. Chin, 2006: Two-decadal aerosol Ringer, M. A., and R. P. Allan, 2004: Evaluating climate model trends as a likely explanation of the global dimming/ simulations of tropical cloud. Tellus, 56, 308–327. brightening transition. Geophys. Res. Lett., 33, L15806, ——, and Coauthors, 2006: The physical properties of the atmo- doi:10.1029/2006GL026471. sphere in the new Hadley Centre Global Environmental Trewartha, G. T., and L. H. Horn, 1980: An Introduction to Cli- Model (HadGEM1). Part II: Aspects of variability and re- mate. 5th ed. McGraw-Hill, 416 pp. gional climate. J. Climate, 19, 1302–1326. Wang, Z., X. Zeng, M. Barlage, R. E. Dickinson, F. Gao, and Rossow, W. B., and Y. C. Zhang, 1995: Calculation of surface and C. B. Schaaf, 2004: Using MODIS BRDF and albedo data to top of atmosphere radiative fluxes from physical quantities evaluate global model land surface albedo. J. Hydrometeor., based on ISCCP data sets. 2: Validation and first results. J. 5, 3–14. Geophys. Res., 100 (D1), 1167–1197. Whitlock, C. H., and Coauthors, 1995: First global WCRP short- ——, and R. A. Schiffer, 1999: Advances in understanding clouds wave surface radiation budget data set. Bull. Amer. Meteor. from ISCCP. Bull. Amer. Meteor. Soc., 80, 2261–2287. Soc., 76, 905–922. Rothman, L. S., and Coauthors, 2003: The hitran molecular spec- Wild, M., 2005: Solar radiation budgets in atmospheric model troscopic database: Edition of 2000 including updates intercomparisons from a surface perspective. Geophys. Res. through 2001. J. Quant. Spectrosc. Radiat. Transfer, 82, 5–44, Lett., 32, L07704, doi:10.1029/2005GL022421. doi:10.1016/S0022-4073(03)00146-8. ——, A. Ohmura, H. Gilgen, and E. Roeckner, 1995: Validation Schaaf, C. B., and Coauthors, 2002: First operational BRDF, al- of general circulation model radiative fluxes using surface J. Climate, bedo nadir reflectance products from MODIS. Remote Sens. observations. 8, 1309–1324. ——, ——, ——, ——, M. Giorgetta, and J.-J. Morcrette, 1998: Environ., 83, 135–148. The disposition of radiative energy in the global climate sys- Sellers, P. J., S. O. Los, C. J. Tucker, C. O. Justice, D. A. Dazlich, tem: GCM-calculated versus observational estimates. Climate G. J. Collatz, and D. A. Randall, 1996: A revised land surface Dyn., 14, 853–869, doi:10.1007/s003820050260. parameterization (SiB2) for atmospheric GCMS. Part II: The ——, ——, ——, J. J. Morcrette, and A. Slingo, 2001: Evaluation generation of global fields of terrestrial biophysical param- of downward longwave radiation in general circulation mod- eters from satellite data. J. Climate, 9, 706–737. els. J. Climate, 14, 3227–3239. Smith, G. L., A. C. Wilber, S. K. Gupta, and P. W. Stackhouse Jr., ——, ——, ——, and D. Rosenfeld, 2004: On the consistency of 2002: Surface radiation budget and climate classification. J. trends in radiation and temperature records and implications Climate, 15, 1175–1188. for the global hydrological cycle. Geophys. Res. Lett., 31, Smith, S. J., R. Andres, E. Conception, and J. Lurz, 2004: Histori- L11201, doi:10.1029/2003GL019188. cal sulfur dioxide emissions 1850–2000: Methods and results. ——, and Coauthors, 2005: From dimming to brightening: Decad- PNNL Rep. 14537, Joint Global Change Research Institute, al changes in solar radiation at the earth’s surface. Science, College Park, MD, 16 pp. 308, 847–850. Stackhouse, P. W., Jr., S. J. Cox, S. K. Gupta, R. C. DiPasquale, ——, C. N. Long, and A. Ohmura, 2006: Evaluation of clear-sky and D. E. Brown, 1999: The WCRP/GEWEX Surface Radia- solar fluxes in GCMs participating in AMIP and IPCC-AR4 tion Budget Project Release 2: First results at 1 degree reso- from a surface perspective. J. Geophys. Res., 111, D01104, lution. Preprints, 10th Conf. on Atmospheric Radiation: A doi:10.1029/2005JD006118. Symp. with Tributes to the Works of Verner E. Suomi, Madi- Wittmeyer, I. L., and T. H. Vonder Haar, 1994: Analysis of the son, WI, Amer. Meteor. Soc., 520–523. global ISCCP TOVS water vapor climatology. J. Climate, 7, Stanhill, G., and S. Cohen, 2001: Global dimming: A review of the 325–333. evidence for a widespread and significant reduction in global Zhang, Y., W. B. Rossow, and A. A. Lacis, 1995: Calculation of radiation with discussion of its probable causes and possible surface and top of atmosphere radiative fluxes from physical agricultural consequences. Agric. For. Meteor., 107, 255–278. quantities based on ISCCP datasets 1. Method and sensitivity ——, and S. Moreshet, 2002: Global radiation climate changes: to input data uncertainties. J. Geophys. Res., 100, 1149–1165. The world network. Climatic Change, 21, 57–75. ——, ——, ——, V. Oinas, and M. I. Mishchenko, 2004: Calcula- Stephens, G. L., 1996: How much solar radiation do clouds ab- tion of radiative fluxes from the surface to top of atmosphere sorb? Science, 271, 1131–1133. based on ISCCP and other global data sets: Refinements of ——, 2005: Cloud feedbacks in the climate system: A critical re- the radiative transfer model and input data. J. Geophys. Res., view. J. Climate, 18, 237–273. 109, D19105, doi:10.1029/2003JD004457.

Unauthenticated | Downloaded 10/10/21 04:08 PM UTC