surface solar radiation flux and cloud radiative forcing...

VOLUME 54, NO. 10 15 MAY 1997J O U R N A L O F T H E A T M O S P H E R I C S C I E N C E S

q 1997 American Meteorological Society 1289

Surface Solar Radiation Flux and Cloud Radiative Forcing for the AtmosphericRadiation Measurement (ARM) Southern Great Plains (SGP): A Satellite, Surface

Observations, and Radiative Transfer Model Study

CATHERINE GAUTIER

Geography Department and Institute for Computational Earth System Science, University of California, Santa Barbara, Santa Barbara,California

MARTIN LANDSFELD

Institute for Computional Earth System Science, University of California, Santa Barbara, Santa Barbara, California

(Manuscript received 24 October 1995, in final form 24 September 1996)

ABSTRACT

This study presents surface solar radiation flux and cloud radiative forcing results obtained by using a com-bination of satellite and surface observations interpreted by means of a simple plane-parallel radiative transfermodel called 2001. This model, a revised version of a model initially introduced by Gautier et al., relatescalibrated radiance observations from space to incoming surface solar flux. After a description of the model,an evaluation is presented by comparison with a more complex model that the authors have developed, the SantaBarbara DISORT Atmospheric Radiative Transfer model (SBDART) based on the discrete ordinate model ofStamnes et al. This evaluation demonstrates this model’s accuracy for instantaneous surface flux when used toretrieve daily (and monthly) surface solar flux. Limitations related to its lack of treatment of the bidirectionalreflectance properties of clouds are also discussed and quantified by comparison with SBDART for instantaneoussurface solar flux retrievals. The influence of satellite sensor calibration uncertainty is also examined in termsof surface solar flux.

The model has been applied to hourly GOES data collected over the Atmospheric Radiation Measurement(ARM) program’s central cloud and radiation testbed site in Oklahoma during a 14-month period to estimatehourly, daily, and monthly surface solar radiation flux. Comparisons of the model’s results with surface mea-surements made from pyranometers located at the ARM site indicate good overall agreement. The best resultsare obtained for daily integrated clear skies with an rms error less than 10 W m22 (or about 3% of the meanvalue) and a 2.8 W m22 bias. These results indicate that the clear sky model is quite accurate and also that thethreshold-based technique to detect cloudy conditions works well for the resolution of the satellite data used inthis study. For partly cloudy conditions the comparisons show an rms error of about 20 W m22 (or less than7% of the mean) and a 22.5 W m22 bias. The performance of the model degrades with cloud cover conditionswith an rms error of 22 W m22 (or 13% of the mean) and a bias of 13.9 W m22 for overcast conditions. Theresults improve considerably for monthly average values with an rms error of about 11 W m22 (or 4% of themean) and a bias of 2.6 W m22 for all conditions.

The model has also been used to evaluate the cloud radiative forcing at the surface and results indicate largevalues of forcing for the spring and summer reaching daily values over 200 W m22 in May.

1. Introduction

Radiation from the sun drives the energy, water, andbiochemical cycles of the earth surface–atmosphere sys-tem. Over land, the incident solar radiation flux deter-mines, in large part, the surface temperature and the rateof evapotranspiration, with important consequences onatmosphere–surface interactions and the global hydro-logic cycle. The surface solar radiation flux is affected

Corresponding author address: Dr. Catherine Gautier, Institute forComputational Earth System Science, University of California, SantaBarbara, Santa Barbara, CA 93106.E-mail: [email protected]

primarily by clouds, but also by aerosols, atmosphericabsorbers, scatterers, and, to a lesser extent, by surfaceconditions.

Various computational methods, based on visible andnear-infrared observations from meteorological satel-lites, have been proposed to estimate surface solar ra-diation flux (e.g., Tarpley 1979; Gautier et al. 1980;Moser and Rashke 1984; Pinker and Ewing 1985; De-dieu et al. 1987; Darnell et al. 1988). With any of thesemethods it is now possible to map the surface solar fluxaccurately (to better than 10% on a daily timescale) overlarge regions of the globe, but also over the entire globesince global satellite datasets have recently becomeavailable from the International Satellite and Cloud Cli-

1290 VOLUME 54J O U R N A L O F T H E A T M O S P H E R I C S C I E N C E S

matology Project. Global, long-term climatologies ofsurface solar flux generated from satellite observationsare now being produced (Whitlock et al. 1995; Charlocket al. 1994; Bishop and Rossow 1991) that can be usedto investigate the sensitivity of the climate system toclouds and surface processes.

In the present study, we apply the satellite method ofGautier et al. (1980) to GOES-7 Visible and InfraredSpin Scan Radiometer (VISSR) data acquired duringthe first year and a half of the Atmospheric RadiationMeasurement (ARM) experiment over the first experi-mental site located in the U.S. Southern Great PlainsCloud and Atmospheric Radiation Testbed (SGPCART). The objectives of this paper are to 1) quanti-tatively evaluate the method and 2) provide the temporaland spatial average and moments of the variability ofsurface solar radiation flux during the experiment overan area typically covered by the grid cell of a generalcirculation model (GCM). This information is expectedto help general circulation modelers involved in ARMprojects develop improved parameterizations of cloudsand their interaction with radiation in their modelsthrough validation of their results with our satellite de-rived parameters.

In this paper, section 2 describes the main features ofthe radiative transfer model 2001. Section 3 evaluatesthe theoretical limitations of the method for a varietyof atmospheric, surface, and geometrical (illuminationand viewing) conditions by comparing its results withthose from a more complex radiative transfer model.Section 4 discusses the impact of the satellite sensorcalibration on the results of the method. Section 5 pre-sents the methodology applied to compute the surfaceflux. Section 6 compares the satellite estimates madewith our model to the in situ pyranometer measurementsand discusses the results. Section 7 presents the first oneand a half years of surface solar radiation flux over theARM site. Section 8 presents analyses of the resultsconcerning the shortwave radiation forcing at the sur-face. The last section summarizes and discusses the im-plications of the results presented in the previous sec-tions.

2. Method

The 2001 model, used to compute the surface solarradiation flux, was originally developed by Gautier etal. (1980) and is based on simple physical modeling ofthe most important radiative processes occurring withinthe atmosphere, namely scattering and absorption bymolecules, clouds, and aerosols. Since the variability ofsurface solar flux results primarily from changes in solarzenith angle and cloudiness, the method focuses on de-termining the effect of clouds on surface solar flux sincethe solar zenith angle can be computed accurately fromsimple formulas. The method accomplishes this by com-puting cloud albedo, the governing cloud parameter,from GOES VISSR measurements in the visible spec-

trum. The repeat coverage of the GOES VISSR data(one observation every hour in this study) allows anadequate sampling of the diurnal cloud variability.

Images composed of 695 3 415 pixels centered on36.288N and 96.928W and covering a region of over450 000 km2 are acquired every hour from GOES-7 overthe ARM site. The area studied here is commensuratewith the size of a typical GCM grid size. The first stepof the computational procedure is to calibrate the ra-diances measured by the VISSR sensor. This is dis-cussed below. The next step is to estimate the surfacealbedo. For this, we first determine from a time seriesof satellite images (typically 15 days) the minimumbrightness value for each pixel at each observation timeduring the day. This minimum value defines a threshold(taken a few counts higher) that is used to classify eachGOES VISSR pixel as clear or cloudy. This proceduredoes not allow us to determine whether the pixel ispartially covered by clouds or not. But since we utilizefull-resolution data, the error introduced by not mod-eling the resulting effect of partial cloud cover on thesurface solar flux only has an effect for clouds that aresmaller than about 1 km, that is, those clouds that havea minor impact on the surface solar flux.

Once the pixel’s nature (clear or cloudy) has beendetermined, we apply clear and cloudy sky radiativetransfer models accordingly.

In clear sky conditions, surface solar flux is expressedas

2rI 5 S cosuexp(2C /cosu)/(1 2 C A )o o 1 2 s1 2ro

b bo wu uo w3 exp 2a exp 2a , (1)o w1 2 1 2[ ] [ ]cosu cosu

where So is the solar constant; r/ro is the ratio of actualto mean Earth–Sun distance; u is solar zenith angle; uo

and uw are ozone and water vapor amounts, respectively;As is surface albedo; and ao, bo, aw, bw, C1 and C2 arecoefficients (C1 and C2 depend on the type and concen-tration of aerosols and account for Rayleigh scattering).The first two terms, r/ro and cosu, are computed fromthe ephemeris programs of the IPW software (Dozierand Frew 1990). The term 1 2 C2 As accounts for pho-tons that have sustained multiple surface reflections.Equation (1), proposed by Frouin et al. (1989), differsfrom that of the original model in formulation but notin essence. Ozone and water vapor amounts are specifiedfrom climatology, and As is obtained by solving thefollowing equation:

Asat(Bmin) 5 a 1 (1 2 a)(1 2 a1)(1 2 )As,9ao (2)

where Asat is the albedo measured at the satellite (thesurface is assumed to reflect solar radiation isotropi-cally), Bmin is the minimum brightness, and charac-9ao

terizes ozone absorption. The direct and diffuse reflec-tion coefficients a and a1 are respectively determined

15 MAY 1997 1291G A U T I E R A N D L A N D S F E L D

FIG. 1. Comparison of surface solar radiation flux computed by2001 and DISORT.

from the tables of Coulson (1959). Equation (2) simplystates that Asat is the sum of an atmospheric component(photons reflected back to space without surface reflec-tion) and the signal reflected by the surface and diffuselytransmitted to space.

In cloudy sky conditions, the clear sky formulationis modified to account for reflection and absorption byclouds, which are assumed to occur in one layer. Cloudysky surface solar flux is therefore given by

Ic 5 Io (1 2 Ac 2 ac)/(1 2 Ac As), (3)

where Ac is cloud albedo and ac is cloud absorption.The denominator represents the effect of multiple re-flections between the cloud and the surface. This effectis generally small except over snow/ice conditions,which were rarely encountered in this study.

Cloud albedo is obtained by solving the followingquadratic equation:

A 5 a 1 (1 2 a)(1 2 a )(1 2 a9)Asat 1 o c

21 (1 2 A 2 a ) (1 2 a )(1 2 a )A , (4)c c 1 o s

where Asat is the top-of-atmosphere albedo, assumingthat clouds reflect solar radiation isotropically. Thisequation, in fact, gives Ac in the GOES VISSR solarchannel (0.5–0.85 mm). We assume that Ac takes thesame value in the spectral interval of total surface solarflux. Depending on liquid water path, the ratio of nar-rowband to broadband albedo increases or decreaseswith sun zenith angle but, in general, the difference issmall. The cloud absorption parameter ac is adjustableand parameterized as a function of the cloud albedo, aproxy for the cloud optical depth in this case. The pa-rameterization used can be empirically determined (forinstance, in the original version of the model this pa-rameter was expressed as a linear function of cloudalbedo with a slope of 0.07) or adjusted to fit moresophisticated radiative transfer models, as described be-low.

3. Model numerical evaluation

An important issue with this type of simple model isits accuracy. Whereby complex radiative transfer mod-els that handle monochromatic radiation interaction witha surface and a layered atmosphere containing scatteringand absorbing components are available, these modelscannot be efficiently used as the basis of satellite al-gorithms. Simplifications and/or computational artifactsmust be introduced to speed up their execution time andmake it commensurate with the amount of data availablefrom space.

One of the main advantages of the model describedin the previous section is that it is simple and thereforeefficient in its processing time, but also requires a min-imum of ancillary information about the state of theatmosphere. It only requires calibrated radiances andillumination and viewing geometry. Legitimate ques-

tions can then be asked, such as how do the results fromsuch a model compare with those of more sophisticatedmodels and under which conditions does this model failto provide acceptable values?

It is to address this type of questions that we haveundertaken the validation reported here. To evaluate thesimple 2001 model, we compare it with our version ofthe well-known and widely used discrete ordinate model(DISORT) of Stamnes et al. (1988). Our version of DI-SORT, called the Santa Barbara DISORT AtmosphericRadiative Transfer model (SBDART) [briefly describedin appendix A and more fully described in P. Ricchiazziet al. 1997, manuscript submitted to Earth Interactions],allows us to specify micro- and macroscale atmosphericproperties for multiple investigative runs.

In order to evaluate the accuracy of the radiativetransfer computations performed by the 2001 model in-dependently of the input data, we have used SBDARTto compute pairs of upwelling top-of-the-atmospherenarrowband radiance in the wavelength region coveredby VISSR and surface solar (broadband) radiation fluxfor a variety of surface, atmospheric, and cloud con-ditions. These upwelling radiances were input to the2001 model. Then, the derived broadband surface solarradiation was compared with that computed withSBDART. The results from these comparisons are pre-sented in Fig. 1 for a vegetated surface and ozoneamount of 0.3 (cm, NTP). The results are quite similarfor other surfaces. To facilitate the interpretation, theconditions characterized by similar optical depths butdifferent atmospheric conditions and illumination andviewing angle geometry have been plotted using dif-ferent symbols. We can first notice the excellent agree-ment between the two models for clear conditions (t 50), except for a few cases. These cases correspond to avery dry atmosphere containing large amounts of aer-osols. The differences result mostly from the differentways of representing the aerosol effects in both ap-


proaches. In 2001, the aerosol effects have been fittedto those produced by the 5-S model of Tanre et al.(1990). These are based on a horizontal visibility pa-rameter that is related to a vertical profile of aerosolparticle density. The aerosol model included inSBDART follows the parameterization included inLOWTRAN 7 (Kneizys et al. 1988). This aerosol modelcontains a slightly different makeup of aerosol constit-uents and is dependent on the water vapor content ofthe atmosphere. Hence, the differences we see in themodel comparisons are due to the lack of sensitivity ofthe 5 S aerosol parameterization to extremely dry aero-sol conditions, which are rare occurrences in nature,typified for example by dust storms.

In cloudy conditions, for similar optical depths, sev-eral sets of solutions exist for SBDART but only onesolution exists for the 2001 model. This results fromthe fact that for one set of atmosphere/cloud conditions,different radiances correspond to different illuminationand viewing geometries. Since we assume in the 2001model that the radiation is isotropic, this model providesone answer for each radiance. These results thereforeprovide a quantification of the effects of cloud aniso-tropic reflectance on the estimation of the surface solarflux, when isotropy is assumed. Differences as large as200 W m22 (corresponding in some instances to 100%error in the determination of the surface solar flux) canbe found for optical depths varying from 2 to 100 forthe different geometries tested.

To further clarify these anisotropic effects for plane-parallel cloud conditions and evaluate how they varywith different parameters, we have plotted the aniso-tropic factor of shortwave flux at the top of the atmo-sphere for different cloud optical depths and solar andviewing geometries. These are presented in the form ofpolar plots in Fig. 2 for sun zenith angles of 158, 308,and 608, respectively, and optical depths of 2, 10, 30,and 100. The spatial distribution of anisotropic factorsvaries dramatically for the investigated conditionsshowing limb brightening for small optical depths anda rather isotropic field centered around the point of mir-ror reflection for larger optical depths. Limb brighteningdominates for all conditions at large viewing angles(e.g., 608). From these results a number of conclusionscan be reached regarding how these anisotropic effectsof cloud reflectance could be corrected for in the 2001model or any equivalent two-stream model to representthe anisotropy of plane parallel clouds. For large view-ing angles (.608), these results suggest that a limbbrightening correction for all optical depths for homo-geneous clouds would improve the representation of the2001 results by our model. From Fig. 2 it is also ap-parent that an additional anisotropy correction could beincluded to minimize the anisotropy effects of cloudreflectance. In 2001 this correction cannot be based di-rectly on optical depth (since the optical depth is notknown), but could be based on a related parameter suchas cloud albedo. For average surface solar radiation flux

(daily to monthly), these anisotropy effects are com-pensated in part by averaging over different geometries,assuming that the averages simulate a daily or monthlyaverage. This is shown in Fig. 3, which presents a com-parison of 2001 results averaged for all geometries, foreach test condition and the SBDART results. A largereduction in the difference between the two sets of re-sults has occurred with part of the remaining differencedue to the difference between the treatment of aerosols(as discussed above for clear sky conditions). One ofthe difficulties with including such anisotropy correc-tions is that real clouds are not plane parallel, and there-fore even though we might be able to tune 2001 toreproduce SBDART-like results, there is no assurancethat these model results will be closer to reality thanthose presently produced by 2001. As we discuss later,using the standard ERBE Angular Distribution Models(ADMs) does not improve the surface solar irradianceresults obtained from 2001.

As mentioned in the previous section, cloud absorp-tion in 2001 is expressed as a function of cloud albedo.To ensure a similarity with SBDART’s representationof cloud absorption, in 2001 we have adjusted the re-lationship between cloud absorption and cloud albedoto be that obtained from the SBDART model at a con-stant sun zenith angle of zero. This empirical fit is pre-sented in Fig 4. The nonlinearity of the relationship isobvious, especially for large cloud albedo values forwhich cloud absorption is slightly reduced. Admittedly,performing comparisons between 2001 and SBDARTresults using an absorption empirically fitted toSBDART does not represent a completely independentvalidation. Such a validation has yet to be conducted.

Our 2001 model implicitly assumed that the cloudlayer is fixed at 1 km and contains droplets of constanteffective radii equal to 8 mm. To evaluate the impact ofthis assumption on the surface solar irradiance com-putations, we have performed SBDART computationsfor different cloud height and droplet properties. Thedifferences between the standard 2001 model and theresults obtained with modified cloud conditions are pre-sented in Table 1. The values for all of the statistics arevery similar for each case with the exception of meanbias error of the two differing cloud heights. This slight-ly higher bias reflects a high bias at high cloud albedofor all values of Re. The differences are so small, how-ever, that these parameter variations were not incor-porated into our model to maintain its simple nature.Overall, the results presented above show that the 2001model performs very well under most conditions, whenthe issue of anisotropy is, at least partly, removed bytime-averaging processes. Obviously, a model such asSBDART could theoretically be far superior to 2001,but requires such a large number of input parameters(and therefore assumptions) with regards to cloud prop-erties (microphysical and macrophysical) as well as at-mospheric and surface properties that this rapidly out-strips its advantages. Furthermore, a radiance-based


FIG. 2. Cloud anisotropic effects for varying optical depths at illumination angles. Contours are upwelling irradiance from clouds in Wm22sr21.

model such as SBDART is extremely processing timeconsuming and therefore unrealistic for processing largeamounts of data, such as those involved in computationsof global surface solar flux, or even long-term high tem-poral resolution datasets, such as the ARM dataset dis-cussed above.

4. Calibration issues

As mentioned above, the input data to 2001 is cali-brated visible brightness (or radiance). Obviously, sim-ilar radiative transfer-based models have the same cal-

ibration requirement. Operational satellite sensors arecalibrated before launch but have no usable in-flightcalibration system. Therefore, vicarious in-flight cali-bration must be performed to remedy this deficiency.Although not entirely satisfactory, a number of ap-proaches have been proposed and used in the recentpast. While our goal is not to provide an extensive re-view of the presently available calibration methods, itis important to note that despite recent efforts the cal-ibration of the U.S. operational sensors (VISSR on theGOES satellites and AVHRR on the NOAA satellites)is still uncertain to probably at least 6%–7% (Whitlock


FIG. 3. Comparison of angle averaged surface solar radiation fluxcomputed by 2001 and DISORT.

TABLE 1. Results of comparison between the radiative transfer mod-el 2001 and SBDART, varying cloud height and cloud droplet ef-fective radius.

Cloudheight(km)

Re

(mm)

Mean dif-ference

(%)

Std dev ofdifference(W m22)

Mean biaserror

(W m22) R2

rmse(W m22)

111444

48

1648

16

28.2428.4728.3629.1829.2329.03

5.555.385.326.225.905.71

9.510.4

9.917.818.017.1

0.980.970.970.970.970.97

52.352.752.855.855.454.9

FIG. 5. GOES-7 VISSR gain evolution over time estimated fromvicarious calibration (from Whitlock).

FIG. 4. DISORT model comparison between cloud albedo andcloud absorption for shortwave radiation.

et al. 1990; Whitlock et al. 1993). This value representsan estimate obtained by comparing results from all theavailable calibration methods and computing their stan-dard deviation. It therefore does not represent an ab-solute calibration assessment.

A second issue of importance is the drift most visiblesensors experience with time. Both VISSR and AVHRRsensors have been found to deteriorate with time, whichleads to increasing gain with time. While the source ofthis deterioration is not entirely clear (deposition on thesensor, optics deterioration), it can be rather large (e.g.,10% per year for a VISSR sensor). These sensors musttherefore be routinely monitored to adjust for any suchdeterioration. Whitlock et al. (1990) have suggested thatthe gain of the VISSR instrument on GOES-7 has in-creased in the manner described by Fig. 5. As can beseen, the changes since 1991, after Mount Pinatubo’seruption, have not been monitored due to the sensitivity

of the monitoring technique to any long-term change inatmospheric conditions, particularly high aerosol con-centrations. Also, the apparent periodic changes are notcompletely understood and thus difficult to predict.

The third issue with regards to calibration is that ofits impact on the computation of the surface solar flux.To evaluate this impact, Gautier and Frouin (1988) havecomputed the sensitivity of the surface solar flux toradiometric calibration of the visible instrument. Thisevaluation was based on an earlier version of 2001, verysimilar to that presented above. The approach was tolinearize the model equation about a reference state andevaluate the changes in surface solar flux with respectto instrument gain (g) changes (or sensitivity). For that,the sensitivity of the surface solar flux to gain (]/]g)was evaluated from the model’s equations (1), (2), (3),and (4). When applied to these equations, the operator]/]g leads to an equation for ]Is/]g that is a linear com-bination of ]As/]g, ]N/]g, and ]Ac/]g. Further mathe-matical manipulations lead to the following conclusions.1) The sensitivity of surface albedo to calibration gainis a linear function of the surface albedo; the brighterthe surface, the higher the sensitivity. 2) The sensitivityof the cloud albedo to calibration gain is a nonlinear


function of the cloud albedo, the nonlinearity increasingwith surface reflection. 3) The sensitivity of the surfacesolar flux to a 10% uncertainty in calibration gain varieswith surface and cloud conditions with a range of 75W m22 for totally cloudy skies (N 5 1) to 250 W m22

for solar zenith angle of 0 degree with selected condi-tions (Ac 5 0.4, As 5 2, N 5 0.5). Thus, instantaneously,the sensitivity to the calibration can be as large as 275W m22, whereas for monthly average values the errorcan be reduced to 215 W m22 because of error com-pensations. The bias on the solar flux computation isinversely related to the calibration bias; that is, a neg-ative bias in calibration induces a positive bias on thesurface solar flux. This result is very important and notunexpected. As a consequence of the sensor’s deterio-ration with time, we can expect a negative bias in thesensor’s gain and therefore a positive bias in the com-puted surface solar flux, independent of the quality ofthe radiative transfer model used to perform the com-putations. If not properly interpreted in terms of thesensor’s calibration, such a positive bias could be mis-interpreted as a potential cloud effect such as cloudsabsorbing more solar radiation than is represented inany present radiative transfer model. Therefore, greatcare has to be taken to correctly calibrate operationalsensors that are used in conjunction with radiative trans-fer models for estimating the surface solar flux.

5. Surface solar radiation flux computations

We focus on the first 14 months of the ARM programand thus on a dataset that extends from 1 March 1993to 30 April 1994. The experimental site is located atthe ARM Southern Great Plains CART site in centralOklahoma.

a. Satellite data

To compute surface solar radiation flux over the firstCART site (using the satellite calibration method de-scribed above), GOES-7 VISSR visible and near-infra-red (solar channel) data were acquired at full resolution(0.9 km at nadir) every hour during daytime. This res-olution translates into a 1.25-km resolution at the CARTsite. The data, 8-bit coded, navigated, but uncalibratednumerical counts, were made available by the ARMExperimental Data Center. In a preprocessing stage, wecalibrate and check the data for gross navigational errorsand data quality. The navigational errors are only de-tectable in clear sky conditions by visual inspection oflarge surface features. The SGP CART site comparisoncomputations presented are made over a 3 pixel 3 3pixel area to use as high a resolution as possible butstill compensate for locational inaccuracies and angularintegration (2p solid angle) of the surface measurementswith which the satellite estimations are compared.

b. VISSR visible sensor calibration

As mentioned above, the calibration of the satellitedata used is central to the accuracy of the surface solarflux estimated from these datasets. Considering that thelast vicarious calibration made was before Mt. Pinatu-bo’s eruption (June 1991) and that the VISSR calibrationdeteriorates and also varies seasonally, it is rather dif-ficult to confidently extrapolate the gain value from June1991 to 1993 and 1994. It is even more difficult toconfidently extrapolate the amplitude and frequency ofthe apparent oscillation with such a small sample set.We therefore derived a drift only from the data presentedin Fig. 5 to perform our extrapolation. The results fromthis extrapolation provides a gain varying with time inthe following manner:

g(t) 5 g0 1 g1*ty, (5)

where g0 is the instrument gain at t 5 0 and g 5 0.009,g1 is the gain’s rate of change with time (0.00084), andty is the time since launch (years). As for the offset, weassume that it remains equal to zero, as it was in 1991.

Furthermore, although the VISSR sensor is composedof eight detectors that have been found to deteriorate atdifferent rates, we used the gain provided in Eq. (5) forall the detectors. This is probably adequate since NOAA,the operator of the GOES satellite, regularly (and ar-tificially) modifies the gain of some of the sensors toprovide an aesthetic nonstripped appearance to the im-ages that it makes available to the public.

c. Surface measurements

The surface measurements used in this study are thosethat are routinely recorded by the BSRN (Baseline Sur-face Radiation Network) pyranometer at the SGP CARTsite with the correction factor introduced in November1993. The measurements taken at each minute are timeaveraged for the 20-min period encompassing each sat-ellite acquisition time. This is roughly equivalent to theaverage time for clouds to traverse the 3 pixels used inthe computations of the spatially averaged surface solarflux from the satellite data when a typical wind speedis used.

6. Analysis of results

Figure 6 presents comparisons of some daily timeseries of surface solar radiation flux measured by theSGP CART pyranometer described above and hourlyestimations made with 2001 applied to GOES instan-taneous visible radiances every hour. These time seriesclearly show the high-frequency variability of the pyr-anometer’s measurements, which results from the pres-ence of small clouds overhead. The series on the leftpanels present the datasets corresponding to the fourbest comparisons between the surface measurementsand the satellite estimations. The criterion used to select


FIG. 6. Comparison of measured surface solar radiation flux with the RTM (bars). Shown are the four best and fourworst cases based on the daily integrations.

the best or worse cases is based on the comparison ofdaily integrated surface solar flux; therefore, some casesare labeled worst when data are missing during a periodwhen clouds vary greatly. The best cases generally arein rather clear or partly cloudy conditions. In cloudyconditions, the satellite estimations capture the changesin solar flux (due to variations in cloudiness) measuredby the pyranometers. The worst cases correspond to lowsolar flux values (thick clouds), to missing satellite mea-surements for a long period during the day, or to caseswhere there is an oscillatory variation in the solar flux

(cloudiness) and the satellite estimations oscillate outof phase with the pyranometer oscillation; that is, whenthe pyranometer measures a low solar flux value, thesatellite estimation indicates a high solar flux value.

To further quantify the differences between the 2001computations and the surface observations, the hourlyresults have been statistically analyzed. A scatterplot ofthese comparisons is presented in Fig. 7. Also includedare the statistics of the comparison of these two datasets.These statistics are presented in two forms. 1) At thebottom right are the mean differences (or the mean of


FIG. 7. Scatterplot of measured hourly surface solar radiation fluxvs 2001 calculations for all conditions.

FIG. 8. Scatterplot of measured hourly surface solar radiation fluxvs computed flux for clear sky conditions (3 3 3 pixel classifier).

FIG. 9. Difference between measured surface solar radiation andcomputed flux using 2001 as a function of sun angle.

the differences) between the two datasets and the stan-dard deviation of that difference (expressed in W m22)and in percentage of the mean value and 2) at the topleft are the equations for the best linear fit, the meanbias, the squared correlation, and the rms error. Thedatasets compared are very highly correlated (R2 50.869), have a very small negative bias (;211 W m22),and have an rms error of 93 W m22 (or 18 % of themean value of the entire dataset). This dataset includesclear and cloudy conditions. A few outliers are ob-served, however, with a small tendency for the satelliteestimations to be lower than the pyranometer measure-ments for low solar flux values, suggesting that the ab-sorption parameterization used in 2001 may be a littletoo large.

For clear conditions, the 2001 model performs ex-tremely well as shown in Fig. 8. The rms error is 54 Wm22 (or 8.4 % of the mean value of the dataset) and thebias is small (15 W m22) and positive. This figure alsoillustrates the ability of the threshold method to deter-mine clear conditions. Only a few percent of the casesshow that the model clearly mistakes a cloudy scene fora clear one (outliers showing larger values from themodel than what is measured at the surface by the pyr-anometer). In no instance does the model estimate thatit is significantly cloudy when, in fact, it is clear. Thus,if anything, the model will have a small tendency toprovide slightly highly biased results due to the raremisinterpretation of cloudy scenes for clear scenes. Inno way does the surface data analyzed suggest a sig-nificant overestimation by the model.

To investigate whether 2001 results degrade with in-creasing sun angle, we have partitioned our datasets intodifferent sun angle conditions. The results are presentedon Fig. 9 values of the sun zenith angle varying from138 to 708. The plot shows that there is no real tendencyfor a deterioration at large sun zenith angle. This is notsurprising since most radiative transfer model perfor-

mances deteriorate beyond 808 sun angle. That the larg-est differences occur at small solar zenith angles resultsfrom the fact that the surface solar flux is larger at thesesmall angles.

a. Cloudy conditions

To compare the performances of 2001 in differentcloudy conditions, the data have been partitioned intodifferent cloud cover conditions. Comparisons are stillbased on the 3 3 3 pixel array mentioned above, butto ensure an accurate cloud condition classification, fourregimes have been selected based on the number ofcloudy pixels in a 9 3 9 pixel array surrounding theARM-SGP CART site. The classifications used in ourcomputations were clear (no cloudy pixel), mostly clear(less than half of the pixels are cloudy), mostly cloudy


(more than half of the pixels are cloudy), and cloudy(all pixels are cloudy). Results from this partitioning arepresented in Fig. 10 with the statistics for each case.We can see that the statistics deteriorate slightly withthe cloudiness and particularly that there is a tendencyfor the satellite computations to underestimate the val-ues measured at the surface in overcast conditions. Thiscan result from a number of causes. The first one is amisrepresentation of cloud absorption. This is an issuethat is still not resolved and the topic of ongoing dis-cussion within the scientific community (see Ramana-than et al. 1995; Cess et al. 1995; Li et al. 1995; Ste-phens 1996; Arking 1996; Arking et al. 1996). Our pre-liminary work on this issue of cloud absorption indicatesthat absorption in a 3D cloudy atmosphere is complexand varies with sun angle as well as cloud geometry.The simple parameterization used in this model, whichis patterned after a plane-parallel discrete ordinate mod-el (SBDART), is intended to account for some of theprocesses involved. Since a full analysis of this complexissue still remains to be done, we believed that our sim-ple approach is the best possible choice at the moment.Another possibility is calibration. Indeed, if the gain ofthe sensor is improperly characterized and assumed tobe higher than it should be, the brightness correspondingto cloudy pixels would be too high and so would thecloud albedo. The surface solar flux would, in turn, beunderestimated. A departure from linearity of the gainas a function of the count squared could also inducethis type of underestimation for large brightness values.One way to address the origin of the underestimationis to investigate daily comparisons partitioned into clearand cloudy conditions. This is done in the followingsection.

b. Comparisons of daily surface solar flux

The hourly results presented above include some un-certainty due to the cloud anisotropy effects not ac-counted for in 2001, as was discussed earlier. In orderto remove these effects, the hourly data have been in-tegrated over the day using the trapezoidal method withcomputed sunrise and sunset times. We can thereforeexpect an improvement in the statistics of the compar-isons. The overall results obtained from this integrationare presented in Fig. 11 for clear, partly cloudy, overcast,and all conditions.

The cloudy conditions have been partitioned, thistime, into two regimes: overcast and partly cloudy sinceit is rather difficult to come up with stricter criteria thatcan be applied for several hours of the day. Using thispartitioning scheme, we classify daily conditions as ei-ther entirely clear, overcast, or if neither, as partlycloudy.

It is interesting to note that the clear conditions arecomputed with an rms error of less than 10 W m22 (or3% of the mean surface measurements). The overcastconditions are computed with an rms error of about 22

W m22 (or about 15% of the mean surface measure-ments) while the partly cloudy conditions have an rmserror of about 20 W m22 (or about 7% of the meansurface measurements). The rather small slope (0.83)determined for the linear fit between the satellite-basedestimations and the surface measurements in overcastconditions, together with an inspection of the plotteddata suggest that the approach has a tendency to un-derestimate the surface solar radiation flux in the mostcloudy conditions.

c. Sensitivity of results to 2001 assumptions

1) CALIBRATION UNCERTAINTY EFFECTS

Many potential sources for the differences betweenthe 2001 satellite-based computations and the surfaceobservations have been investigated. The first and ob-vious one is calibration error, which is readily under-stood when one considers that the latest available cal-ibration value was obtained in spring 1991, or about 3years before the data analyzed here. Since it is difficultto directly quantify calibration errors, we have per-formed our computations with an alternative availablecalibration source. We have used the GOES-7 VISSRcalibration coefficients provided by the InternationalSatellite Cloud Climatology Project (ISCCP) (Rossowet al. 1992a; Rossow et al. 1992b). We found that thecomputations performed with Whitlock’s calibrationprovided surface solar irradiance values that comparedslightly better with the surface observations than thoseobtained with ISCCP calibration. The standard devia-tions were only slightly different with 12.43% and12.85%, respectively, for the Whitlock and ISCCP cal-ibrations, but the mean biases were significantly differ-ent with 10.81 W m22 and 223.68 W m22, respectively.For daily integrations Whitlock’s calibration resultswere similarly better with the standard deviations of6.95% and 10.26%, respectively, and mean biases of0.65 and 29.49 W m22, respectively.

2) REFLECTANCE ANISOTROPY EFFECTS

Another potential source of error that has been dis-cussed above is the cloud reflectance anisotropy effect.While this can be quantified theoretically for plane-par-allel clouds, its effect for inhomogeneous three-dimen-sional clouds is unknown. Here again, to evaluate theperformances of our model under different sets of as-sumptions that are commonly made, we have performedruns of 2001 using the ERBE (Suttles et al. 1988) an-isotropic cloud reflectance model and compared the re-sults with those obtained with the standard 2001 as-sumptions. For this, we have developed an anisotropiccloud reflectance model based on ERBE climatologicalresults and applied this to the cloud albedo calculationsof the 2001 model. Here again the results obtained withthe 2001 model provide better comparisons with the


FIG

.10

.C

ompa

riso

nof

mea

sure

dan

dca

lcul

ated

surf

ace

sola

rra

diat

ion

for

diff

erin

gcl

oud

cove

rty

pes

(93

9pi

xel

clas

sifi

er).


FIG

.11

.D

aily

inte

grat

edsh

ortw

ave

com

pari

sons

wit

hm

easu

rem

ents

atth

eA

RM

-SG

Pfo

rco

mpl

etel

ycl

ear,

part

lycl

ear,

and

over

cast

days

.


TABLE 2. Comparison between the standard 2001 model and themodel with an ERBE anisotropic cloud reflectance model modifi-cation.

Model Type

Meandif-

ference(%)

Stddev ofdiffer-ence

(W m22)

Meanbias error(W m22) R2

rmse(W m22)

2001Previous2001Previous

HourlyHourlyDailyDaily

27.6416.3

20.4527.35

12.4320.73

6.958.21

11.81223.06

0.6410.39

0.870.710.950.95

92.9147.7

19.724.2

TABLE 3. Comparison between the standard 2001 model and themodel with the previous linear cloud absorption model of ac 5 0.07Ac.

Model Type

Meandif-

ference(%)

Stddev ofdiffer-ence

(W m22)

Meanbias error(W m22) R2

rmse(W m22)

2001ERBE2001ERBE

HourlyHourlyDailyDaily

27.6427.5420.4526.97

12.4312.93

6.9516.07

11.8110.40.649.8

0.870.820.950.76

92.9100.4

19.745.1

FIG. 12. Scatterplot of monthly averaged surface solar radiationflux vs 2001 monthly averaged calculations.

surface observations than those obtained using an ERBEanisotropy correction. These results are summarized inTable 2. They do not suggest that an anisotropy cor-rection is not warranted but only that applying the avail-able one does not lead to improved interpretation ofsatellite radiance in terms of surface shortwave fluxesusing our 2001 model.

3) CLOUD ABSORPTION EFFECTS

Finally, we have evaluated the effects of the cloudabsorption assumption in 2001 by comparing the resultsto those obtained with the previous parameterized ab-sorption (ac 5 0.07 Ac) used in our previous models(Diak and Gautier 1983; Frouin et al. 1989). The resultsobtained again indicate that the present absorption valuebased on SBDART computations provides slightly bet-ter results, which are presented in Table 3.

d. Monthly surface solar flux and its variations

For climate applications, time averaging longer thandaily averaging is often performed to characterize thebehavior of climate variables. To provide an idea of howthe solar flux varies seasonally and annually in the re-gion of the CART site over the scale of a climate model(250–500 km), we have computed monthly averagedvalues over the entire study area. The accuracy of suchmonthly estimates can be assessed in a manner similarto that of the hourly and daily values, that is, by ana-lyzing the scatterplot of the pyranometer and the satellitemonthly mean surface solar flux values. This scatterplotis presented in Fig. 12 with similar statistical infor-mation as for the hourly and daily values. The year andmonth is also indicated next to each plotted point. Thisfigure shows that monthly mean values are obtained withan rms error of 11 W m22 (or 6% of the monthly meanvalue) and a small bias of 2.6 W m22. These resultsshow that the rms error is reduced dramatically by tem-poral averaging, suggesting that a large part of the errorin the shorter time averages is random (noiselike) andlikely due to cloud effects. There still is a bias com-ponent to the rms error (not removed by the averagingprocess), however, which can come from the model orthe data and is on the order of about 10 W m22. This

is consistent with the results discussed earlier concern-ing the calibration effects on the cloudy retrievals, whichwere of the order of 22 W m22 because many days thatmake up the monthly mean are clear or partly cloudyand the monthly rms error can be in large part attributedto the cloudy conditions. The two main outliers thatinfluence the rms error in the monthly mean values arefor springtime.

Monthly mean values for the entire area centered onthe SGP CART site have been computed for the solarradiation flux and the cloud area cover (N). A plot ofthe time series of these parameters is presented in Fig.13. The top panel, which represents the variations ofthe solar radiation flux, shows the strong seasonal vari-ations of the flux induced by both solar zenith anglechanges and cloud effects (cloud cover and type). Thephase of the cycle suggests that the sun effects are dom-inant. The cloud cover variations are presented in thelower panel and also show some seasonality. The min-ima and maxima are, however, out of phase with thoseof the solar radiation flux. In particular, the minima ofcloud cover are in summer and early fall, while latewinter and spring seasons experience the largest cloudcover.

The main result from this analysis is that clouds have


FIG. 13. Time series of monthly average (diamond), standard de-viation (box), and minimum and maximum values (bars) of down-welling shortwave for entire study area.

a tendency to offset the solar zenith angle effects; thatis, if it were not for the clouds, the surface solar radiationflux would have an even larger seasonal cycle.

7. Surface solar flux cloud forcing

Another way of quantifying cloud radiative effects isto investigate the difference in surface solar flux be-tween cloudy and clear conditions. This approach wasintroduced by Charlock and Ramanathan (1985). Sinceclear and cloudy conditions do not exist at the sametime, one has to either rely on models to derive the clearconditions or to compare different time periods, ensur-ing as much as possible that the sun illumination andthe atmospheric conditions (e.g., aerosols and water va-por) between the two datasets compared are similarwithin the accuracy of the measurements used. To com-pute the daily cloud forcing, we need the daily clear skyflux estimated instantaneous (or at least hourly) clearflux. Since there are very few days for which all thesunlight hours are clear, we use our model to computethe clear sky flux for each hour. Cess et al. (1995) hasproposed a method that consists of taking the envelopeof the points on a plot of the downwelling solar radiationflux as a function of the solar zenith angle. If a strictenvelope is taken, the clear sky flux can be overesti-mated because of a few points that correspond to con-ditions when the surface solar irradiance is the sum ofthe total flux plus that reflected on the sides of smallclouds. Some interpretation therefore has to be made.

For this reason and since we have shown that our modelwas accurate for clear sky instantaneous surface solarflux (5.6%, see Fig. 8), we investigate the surface solarflux cloud forcing using a combination of surface mea-surements and model data. Our computations of the sur-face solar flux cloud forcing are made according to

CF 5 2 ,obs modelI Icloudy clear (6)

where is the surface solar flux in clear conditionsmodelIclear

computed with 2001 and is the surface solar fluxobsIcloudy

in cloudy conditions measured with the pyranometers.We will assume a 5% uncertainty in the modeled clearsky conditions, and this is attributable to an imperfectknowledge of the atmospheric conditions, particularlyaerosols. Since typical pyranometer measurements areaccurate to about 5%, the uncertainty on the cloud forc-ing at the surface is expected to be better than 10% forhourly values and 5% for daily values.

We have computed the surface solar flux cloud forc-ing for both hourly and daily values. The hourly resultsare presented in Fig. 14 in the form of a histogram foreach month analyzed. Since our dataset spans over 14months, two months have more data than others. Asexpected, the largest frequencies are for the smallestvalues of hourly surface solar flux cloud forcing. Themonths during which the values are the largest are April,May, and June, for which the surface solar flux cloudforcing reaches values up to 680 W m22. May is themonth during which the largest cloud forcing values areencountered and large values occur most often. For theentire 14-month study period, the mean and standarddeviation for the instantaneous cloud forcings are 147.5W m22 and 152.8 W m22, respectively.

Similar results are found for the daily surface solarflux cloud forcing (Fig. 15). Here again, May is themonth during which the largest number of days havethe largest values. The daily surface solar flux cloudforcing reaches values beyond 200 W m22. So, if wewere interested in performing an experiment whenclouds have the largest effect on the surface solar flux,springtime, and May in particular, would be the bestsuited, if our limited climatology is representative ofthe region. During the entire study period, the mean andstandard deviation for the daily cloud forcings are 53.1W m22 and 50.5 W m22, respectively.

8. Discussion and concluding remarks

The results presented above are very encouragingwith regards to producing high spatial and temporal res-olution climatology of the surface solar flux from sat-ellite observations with our model, 2001. These, onceagain, demonstrate that it is possible to compute time-averaged surface solar flux from satellite data with anaccuracy close to that of surface measurements (madewith operational pyranometers) but with a much higherspatial resolution and coverage. This is particularly im-portant for regions such as oceans or remote regions


FIG. 14. Downwelling hourly shortwave cloud forcing histograms by month.


FIG. 15. Downwelling daily shortwave cloud forcing histograms by month.


where surface measurements are difficult to make on acontinuous basis. While results have not been presentedhere for different types of surfaces, our experience in-dicates that the model performs as well over ocean andother land type surfaces, as over vegetation (as presentedhere). Snow surface conditions have special conditionswith multiple reflections between the surface andclouds, which need to be handled differently than thesimplistic treatment included in 2001.

While rather accurate, the 2001 model also has theadvantage of being particularly simple to use in com-parison with models such as the delta-Eddington or DI-SORT radiative transfer models, which require a muchmore detailed characterization of the physical environ-ment being modeled. The 2001 model can be used withany visible channel satellite radiance data (i.e., AVHRRor any geostationary weather satellites) for which a cal-ibration method is available. Because of the way thecloud discrimination is performed, the 2001 model ac-curacy is expected to deteriorate with the spatial reso-lution of the input data.

Calibration of the data is an important issue that isnot yet solved adequately for the datasets that are madeavailable to the scientific community. Vicarious cali-brations with a surface calibration target are possibleand useful when clear conditions are not affected by thepresence of an unknown amount of aerosols. Other ap-proaches to in-flight calibration must be explored suchas intercalibration with other airborne instruments. Anacceptable solution needs to be developed soon beforelong-timescale satellite-based climatologies are devel-oped with inappropriate calibration coefficients.

The effects of clouds on the surface solar flux, whichhave been computed from a combination of surface mea-surements and radiative transfer model computations,have been found to be large over the SGP CART site,particularly during the spring season. Whereas mostcloud forcing computations have been performed usingnet flux measurements of satellite observations, here wehave computed the cloud effects on the surface down-welling flux solely. This choice was made to avoid errorsintroduced by uncertainties on the surface albedo. Con-sequently, our results represent a different way of com-puting the cloud radiation forcing with data in whichwe have a high degree of confidence. The cloud forcingon the downwelling radiation is larger than the cloudforcing on the net radiation budget by a factor of oneminus the surface albedo. Therefore, for regions of rel-atively low albedo such as the ocean, their differencewill be small, and for regions of high albedo this dif-ference will be large. In the case of the SGP CART, forwhich the surface albedo is close to 20%, we can there-fore expect a reduction by 80% of the net cloud forcingat the surface. This would bring our mean, net instan-taneous and daily values to about 30 W m22 and 10 Wm22, respectively.

Finally, our monthly mean computations have shown2%–11% spatial variability (expressed by the standard

deviation normalized by the mean) in the surface solarflux over the scale analyzed (roughly that of a GCMgrid box), suggesting that over the monthly timescalecloud effects on the surface solar flux can be consideredsmall and random.

The 2001 model is now available to the scientificcommunity. It is currently being used to compute thesurface solar fluxes for the BOREAS experiment (Guand Smith 1995).

Acknowledgments. This research has been funded inpart by the Department of Energy Grant 90ER61062and National Aeronautics and Space AdministrationGrant NAGW-31380. We want to thank Dr. Paul Ric-chiazzi for reviewing an early version of this manuscriptand three anonymous reviewers for their comments,which led to a clearer presentation. Finally, we thankJames Marquez for assistance in its preparation.

APPENDIX

Santa Barbara DISORT Atmospheric RadiativeTransfer (SBDART) Model

The discrete ordinate method (DISORT, Stamnes etal. 1988) provides a numerically stable algorithm tosolve the equations of plane-parallel radiative transferin a vertically inhomogeneous atmosphere. The intensityof both scattered and thermally emitted radiation can becomputed at different heights and directions. Parametersthat define the radiative properties of the atmosphere,such as the single scattering albedo and asymmetry fac-tors of clouds and aerosols, are computed from well-established theories of radiation scattering. The com-putation of gaseous absorption is based on models inthe LOWTRAN 7 (Kneizys et al. 1988) atmospherictransmission code. Our version of DISORT allows gen-eral specification of the model atmosphere and enablesus to treat a very wide variety of atmospheric radiationproblems.

The storage and computing time required by DISORTis considerable, and increasing the spectral resolutions,number of layers, and streams slows the execution timesignificantly. Therefore, we implement the parametersin a modular structure. The gaseous absorption coeffi-cients, aerosol parameters, and solar spectrum in thewavelength region 0.25–33.0 mm are provided at a spec-tral resolution of 20 cm21. For more accurate calculationof narrowband flux and/or intensity in the gas absorptionbands, where higher spectral resolution is required, theK distribution fitting of HITRAN-92 dataset is used.

The distribution of temperature, pressure, and com-position may be selected from a set of standard atmo-spheric profiles or specified from detailed radiosondeprofiles. The single scattering albedo, asymmetry factor,and the extinction efficiency are calculated from a Miescattering code. The current version of the code assumesthe surface is a Lambertian reflector. A BRDF that ap-


plies to different surface types will be implemented forincreased accuracy with intensity calculations.

a. Structure of the model

As a compromise between calculation accuracy, com-puting time, and storage requirements, 40 plane-parallellayers and 16 radiation streams are currently used in thecomputations. The number of streams can be increasedwhen relative narrow band intensity is calculated. Themodel consists of modules to

1) read and validate user input;2) specify solar spectral data and atmospheric profiles;3) specify sensor filter function including predefined

filter types such as AVHRR 1 and 2, Meteosat,GOES, Landsat TM, and SPOT (the code also allowsinput of a user specified filter function);

4) calculate optical thickness of gas and Rayleigh scat-tering;

5) calculate optical thickness, single scattering albedo,and asymmetry factor of aerosols and clouds;

6) calculate surface albedo or BRDF; and7) solve the equation of radiative transfer with DISORT.

b. Optical parameters of the atmosphere

In solving the radiative transfer equations, the param-eters required are the optical thickness, single scatteringalbedo, and asymmetry factor due to gaseous absorp-tion, Rayleigh scattering, aerosols, and clouds. Theydepend on the atmospheric profiles; total amount anddistribution of water vapor, ozone, oxygen, carbon di-oxide, and other gases; type and concentration of aer-osols; and type and liquid water content of clouds.

1) ATMOSPHERIC PROFILES

We have adopted six standard atmospheric profilesfrom the 5 S atmospheric radiation code (Tanre et al.1990), which model the following conditions: tropical,midlatitude summer, midlatitude winter, subarctic sum-mer, subarctic winter, and US62. These models are stan-dard atmospheric profiles recognized by the earth–ra-diation community.

An additional source of information is the TIGR da-taset (Chedin and Scott 1985), which contains 1761 ra-diosonde atmospheric profiles taken between 808S and848N latitude. These data can be used to provide a rangeof typical atmospheric profiles that may be realized atany given latitude.

2) GAS ABSORPTION AND RAYLEIGH SCATTERING

Gaseous absorption is computed using the modelsprovided by Kneizys et. al. (1988). Absorption by H2O(vapor), O3, CO2, N2O, CO, O2, N2, CH4, and trace gasesare computed at a spectral resolution of 20 cm21.

3) AEROSOLS

We use the aerosol dataset from 5 S and LOWTRAN7. In the former case, the single scattering albedo andasymmetry factor of four basic aerosol particles (dust-like, oceanic, water soluble, and soot) are specified, andthe radiative properties of the user-specified aerosol type(continental, maritime, and urban) are calculated bycombining the basic aerosol particle types with differentweights. In the LOWTRAN 7 version, a database thatspecifies radiative properties of rural, urban, oceanic,tropospheric, background, aged and fresh volcanic andmeteor types of aerosols is used directly. In either casethe aerosol optical thickness (which is a function ofwavelength) is set by specifying the meteorological vis-ibility at 550 nm.

4) CLOUDS

Solving the equations of radiative transfer in a cloudyatmosphere requires the knowledge of the cloud singlescattering albedo, asymmetry factor, and extinction ef-ficiency. We have computed these parameters with a Miescattering code for cloud droplets having a lognormalsize distribution for a range of equivalent radii between2 and 64 mm.

For flux calculations we use the Henyey–Greensteinapproximation of the scattering phase function. The Le-gendre expansion of the actual Mie scattering phasefunction can be used to obtain radiation intensity at highangular resolution.

c. Surface conditions

The ground surface cover is one of the most importantfactors determining the radiation environment of theearth–atmosphere system. In our radiation models weuse four basic surface types (clear water, vegetation,snow, and sand) to parameterize the hemispherical spec-tral reflectivity of the surface. The spectral reflectivityof a large variety of ground objects is well approximatedby combinations of these basic types. For example, thefractions of vegetation, water, and sand can be adjustedto generate a new spectral reflectivity representing newor old vegetative growth or deciduous versus evergreenforest. Combining a small fraction of the spectral re-flectivity of water with that of sand yields an overallspectral dependence close to wet soil. We expect thatsix to eight prototypical spectral reflectivities can becombined to get the spectral reflectivity of most groundobjects to an accuracy sufficient for most applications.

REFERENCES

Arking, A., 1996: Absorption of solar energy in the atmosphere—Discrepancy between model and observations. Science, 273,779–782., M. D. Chou, and W. L. Ridgway, 1996: On estimating the effect


of clouds on atmospheric absorption based on flux observationsabove and below cloud level. Geophys. Res. Lett., 23, 829–832.

Bishop, J. K., and W. B. Rossow, 1991: Spatial and temporal vari-ability of global surface solar irradiance. J. Geophys. Res., 96(C9), 16 839–16 858.

Cess, R., and Coauthors, 1995: Absorption of solar radiation byclouds—Observations versus models. Science, 267, 496–499., and V. Ramanathan, 1985: The albedo field and cloud radiativeforcing produced by a general circulation model with internallygenerated cloud optics. J. Atmos. Sci., 42, 1408–1429.

Charlock, T., F. Rose, T. L. Alberta, G. L. Smith, D. Rutan, N. Manalo-Smith, T. D. Bess, and P. Minnis, 1994: Retrievals of the surfaceand atmospheric radiation budget: Tuning parameters with ra-diative transfer top balance pixel-scale ERBE data. Preprints,Eighth Conf. on Atmospheric Radiation, Nashville, TN, Amer.Meteor. Soc., 435–437.

Chedin, A., N. A. Scott, C. Wahiche, and P. Moulinier, 1985: Theimproved initialization inversion method: A high resolutionphysical method for temperature retrievals from satellites of theTIROS-N series. J. Climate Appl. Meteor., 24, 128–143.

Coulson, K. L., 1959: Characteristics of the radiation emerging fromthe top of a Rayleigh atmosphere, 1 and 2. Planet. Space Sci.,1, 256–284.

Darnell, W., W. Staylor, S. Gupta, and F. Denn, 1988: Estimation ofsurface insolation using sun-synchronous satellite data. J. Cli-mate, 1, 820–835.

Dedieu, G., P. Deschamps, and Y. Kerr, 1987: Satellite estimation ofsolar irradiance at the surface of the earth and of surface albedousing a physical model applied to METEOSAT data. J. ClimateAppl. Meteor., 26, 79–87.

Diak, G. R., and C. Gautier, 1983: Improvements to a simple physicalmodel for estimating insolation from GOES data. J. ClimateAppl. Meteor., 22, 505–508.

Dozier, J., and J. Frew, 1990: Rapid calculation of terrain parametersfor radiation modeling from digital elevation data. IEEE Trans.Geosci. Remote Sens., 28 (5), 963–969.

Frouin, R., D. Lingner, and C. Gautier, 1989: A simple analyticalformula to compute clear sky total and photosynthetically avail-able solar irradiance at the ocean surface. J. Geophys. Res., 94(C7), 9731–9742.

Gautier, C., and R. Frouin, 1988: Sensitivity of satellite-derived netshortwave irradiance at the Earth’s surface to radiometric cali-bration. Proc. Fourth Int. Colloquium on Spectral Signatures ofObjects in Remote Sensing, ESA, Aussois, France, 179–184., G. Diak, and S. Masse, 1980: A simple physical model toestimate incident solar radiation at the surface from GOES sat-ellite data. J. Appl. Meteor., 19, 1005–1012.

Gu, J., and E. A. Smith, 1996: A combined solar and infrared surfaceand radiation budget algorithm using GOES imager measure-ments for BOREAS applications. Preprint, Eighth Conf. on Sat-ellite Meteorology and Oceanography, Amer. Meteor. Soc., At-lanta, GA, 541–542.

Kneizys, F., X. E. Shettle, L. Abreu, J. Chetwynd, G. Anderson, W.Gallery, J. Selby, and S. Clough, 1988: Users guide to LOW-TRAN 7. AFGL-TR-88-0177, 137 pp. [NTIS AD 206733.]

Li, Z. Q., H. W. Barker, and L. Moreau, 1995: The variable effect

of clouds on atmospheric absorption of solar radiation. Nature,376, 486–490.

Moser, W., and E. Rashke, 1984: Incident solar radiation over Europeestimated from METEOSAT data. J. Climate Appl. Meteor., 23,166–170.

Nagaragj Rao, C. R., Ed., 1993: Degradation of the visible and near-infrared channels of the Advanced Very High Resolution Ra-diometer on the NOAA-9 spacecraft: Assessment and recom-mendations for correction. NOAA Tech. Rep. NESDIS 70, 1–25.

Pinker, R., and J. Ewing, 1985: Modeling surface solar radiation:Model formulation and validation. J. Climate Appl. Meteor., 24,389–401., and I. Laszlo, 1992: Modeling surface solar irradiance for sat-ellite applications on a global scale. J. Appl. Meteor., 31, 194–211.

Ramanathan, V., R. Cess, E. Harrison, P. Minnis, B. Barkstrom, E.Ahmad, and D. Hartmann, 1989: Cloud-radiative forcing andclimate: Results from the Earth Radiation Budget Experiment.Science, 243, 57–63., and Coauthors, 1995: Warm pool heat budget and shortwavecloud forcing—A missing physics. Science, 267, 499–503.

Ricchiazzi, P., S. Yang, and C. Gautier, 1996: SBDART: A researchand teaching software tool for plane-parallel radiative transferin the Earth’s atmosphere. Earth Interactions, in press.

Rossow, R., C. Brest, and M. Roiter, 1992a: International SatelliteCloud Climatology Project (ISCCP) new radiance calibrations.World Climate Research Programme WMO/TD 736, 71 pp., Y. Desormeaux, C. Brest, and A. Walker, 1992b: InternationalSatellite Cloud Climatology Project (ISCCP) radiance calibrationreport. World Climate Research Programme (ICSU and WMO)WMO/TD 520, 104 pp.

Stamnes, K., S. Tsay, W. Wiscombe, and K. Jayweera, 1988: Nu-merically stable algorithm for discrete-ordinate-method radiativetransfer in multiple scattering and emitting media. Appl. Opt.,27, 2502.

Stephens, G. L., 1996: How much solar radiation do clouds absorb?Science, 273, 1131–1133.

Suttles, J., R. Green, P. Minnis, G. Smith, W. Staylor, B. Wielicki, I.Walker, D. Young, V. Taylor, and L. Stowe, 1988: Angular ra-diation models for earth–atmosphere system. NASA Ref. Publ.1184, 144 pp.

Tanre, D., and Coauthors, 1990: Description of a computer code tosimulate the satellite signal in the solar spectrum: The 5S code.Int. J. Remote Sens., 11, 659–668.

Tarpley, J., 1979: Estimating incident solar radiation at the surfacefrom geostationary satellite data. J. Climate Appl. Meteor., 18,1172–1181.

Whitlock, C. H., W. F. Staylor, J. T. Suttles, G. Smith, R. Levin, R.Frouin, C. Gautier, P. M. Teillet, P. N. Slater, Y. J. Kaufman, B.N. Holben, W. B. Rossow, C. Brest, and S. R. LeCroy, 1990:AVHRR and VISSR satellite instrument calibration results forboth cirrus and marine stratocumulus IFO periods. FIRE Sci.Rep. NASA CP 3083., T. P. Charlock, W. F. Staylor, R. T. Pinker, I. Laszlo, A. Ohmura,H. Gilgen, T. Konzelman, R. C. DiPasquale, C. D. Moats, S. R.LeCroy and N. A. Ritchey, 1995: First global WCRP shortwavesurface radiation budget dataset. Bull. Amer. Meteor. Soc., 76,905–922.

surface solar radiation flux and cloud radiative forcing...

Documents