Solar Energy 135 (2016) 435–445

Contents lists available at ScienceDirect

Solar Energy
journal homepage: www.elsevier.com/locate/solener

A Fast All-sky Radiation Model for Solar applications (FARMS): Algorithm

and performance evaluation
Yu Xie a,⇑, Manajit Sengupta a, Jimy Dudhia b
a
Power System Engineering Center, National Renewable Energy Laboratory, Golden, CO 80401, United States
b
National Center for Atmospheric Research, Boulder, CO 80301, United States

a r t i c l e i n f o a b s t r a c t

Article history: Radiative transfer (RT) models simulating broadband solar radiation have been widely used by
Received 15 September 2015 atmospheric scientists to model solar resources for various energy applications such as operational fore-
Received in revised form 31 May 2016 casting. Due to the complexity of solving the RT equation, the computation under cloudy conditions can
Accepted 1 June 2016
be extremely time consuming though many approximations (e.g. two-stream approach and delta-M
Available online 15 June 2016
truncation scheme) have been utilized. Thus, a more efficient RT model is crucial for model developers
as a new option for approximating solar radiation at the land surface with minimal loss of accuracy. In
Keywords:
this study, we developed a fast all-sky radiation model for solar applications (FARMS) using the simplified
Solar radiation
Radiative transfer model
clear-sky RT model, REST2, and simulated cloud transmittances and reflectances from Rapid Radiation
Cloud Transfer Model (RRTM) with a sixteen-stream Discrete Ordinates Radiative Transfer (DISORT).
Simulated lookup tables (LUTs) of cloud transmittances and reflectances are created by varying cloud
optical thicknesses, cloud particle sizes, and solar zenith angles. Equations with optimized parameters
are fitted to the cloud transmittances and reflectances to develop the model. The all-sky solar irradiance
at the land surface can then be computed rapidly by combining REST2 with the cloud transmittances and
reflectances. This new RT model is more than 1000 times faster than those currently utilized in solar
resource assessment and forecasting since it does not explicitly solve the RT equation for each individual
cloud condition. Our results indicate the accuracy of the fast radiative transfer model is comparable to or
better than two-stream approximation in term of computing cloud transmittance and solar radiation.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction satellite- or surface-based remote sensing measurements (Hsu

et al., 2006; Long et al., 2006; Min and Harrison, 1996; Minnis
Atmospheric radiative transfer (RT) models, e.g., Discrete Ordi- et al., 2011; Nakajima and King, 1990; Sengupta and Ackerman,
nates Radiative Transfer (DISORT) (Stamnes et al., 1988), Rapid 2003; Strow et al., 2003; Wei et al., 2004; Xie, 2010; Xie and Liu,
Radiative Transfer Model (RRTM) (Mlawer et al., 1997), MODerate 2013; Xie et al., 2014, 2009, 2012b). Solar radiation computed by
resolution atmospheric TRANsmission (MODTRAN) (Berk et al., broadband RT models provides quantitative information on solar
1998), 6S (Kotchenova et al., 2006), and Community Radiative resource at specific locations and is needed for the development
Transfer Model (CRTM) (Chen et al., 2008), simulate light scattering of solar energy applications (Sengupta et al., 2010, 2014; Sun
and absorption by air molecules, aerosols and clouds, numerically et al., 2012, 2014; Yang et al., 2006). For the efficient use and man-
solve the RT equation for the Earth’s atmosphere and consequently agement of solar energy on the electricity grid, RT models with
have broad applications in many fields of science and industry. cloud properties from satellites and cloud motion from NWP mod-
Narrow- and broadband RT models for visible through microwave els are used to produce short-term solar forecasts (Lorenz et al.,
wavelengths are often used to simulate upwelling and 2009; Perez et al., 2010).
downwelling radiances and irradiances and thereby retrieve Broadband RT models particularly designed for clear skies, i.e.
aerosol, cloud, and other atmospheric properties from cloudless conditions, are of great interest for solar applications
since they are usually associated with the maximum available
solar resource during a certain period of time. Although high-
⇑ Corresponding author at: Power System Engineering Center, National spectral-resolution RT models, e.g. line-by-line models (Clough
Renewable Energy Laboratory, 15013 Denver West Parkway, Golden, CO 80401, et al., 1992), provide rigorous solutions of clear-sky solar irradiance
United States.
by accounting for the integration of spectral absorption in the
E-mail address: [email protected] (Y. Xie).

http://dx.doi.org/10.1016/j.solener.2016.06.003
0038-092X/Ó 2016 Elsevier Ltd. All rights reserved.
436 Y. Xie et al. / Solar Energy 135 (2016) 435–445

atmosphere, a number of clear-sky RT models have been developed The rest of this paper is organized as follows. The analytical for-
using parameterizations of broadband solar transmittance from mulation of FARMS and the parameterization of the LUTs for cloud
surface observations or model simulations in spectral bands reflectances and transmittances are presented in Section 2. The
(Atwater and Ball, 1978; Bird and Hulstrom, 1981; Gueymard, accuracy and computational efficiency of FARMs are compared to
1989, 2003, 2008; Hoyt, 1978; Ineichen, 2008; Lacis and Hansen, the state-of-the-art RT models and discussed in Section 3.
1974; Rigollier et al., 2000). Those clear-sky RT models are efficient Conclusions can be found in Section 4.
and heavily used in solar applications despite their limited capabil-
ities to compute spectral or broadband radiation at various levels 2. Methodology
of the atmosphere beyond the land surface.
Compared to clear-sky RT models, RT computations under Fig. 1 is a flowchart demonstrating the computation of solar
cloudy skies are considerably more computationally expensive radiation using FARMS. FARMS calculates broadband solar
due to the complexity of solving the RT equation (Chandrasekhar, radiation under all-sky conditions, i.e. clear, cloud overcast, and
1950; Liou, 2002; Wendisch and Yang, 2012). Two-stream approx- partially cloudy skies. The clear-sky solar radiation in the direct
imation and its numerous variants, e.g. Eddington approximation, and diffuse directions are computed by REST2 (Gueymard, 2008)
modified Eddington, quadrature, and delta function (Irvine, 1965, because it has been proved to be computationally efficient yet
1968; Meador and Weaver, 1979; Sagan and Pollack, 1967), are highly accurate. The formulation of REST2 model is detailed in
well-known as efficient solutions of the RT equation under cloudy Gueymard (2008). The rest of this section introduces the formula-
conditions. While two-stream approximation has been applied to tion of overcast cloud conditions since solar radiation under a
solar resource assessment and forecasting (Pinker and Laszlo, partially cloudy sky can be derived from the measurements of
1992; Ruiz-Arias et al., 2012; Skamarock and Klemp, 2008; cloud fraction and a weighted average of the radiation from the
Skamarock et al., 2005), a number of challenges exist for solar clear and cloud overcast conditions. For instance, global horizontal
energy applications. For example, reducing the computational irradiance (GHI) and direct normal irradiance (DNI) for a partially
burden due to the increasing spatial and temporal resolutions of cloudy sky can be given as
satellite data and GCM and NWP experiments demands RT models
GHI ¼ fGHIcld þ ð1  f ÞGHIclr ð1aÞ
with a significantly reduced computational burden than the
two-stream approximation can provide. Moreover, the expansion DNI ¼ fDNIcld þ ð1  f ÞDNIclr ð1bÞ
of discrete ordinate streams (Chandrasekhar, 1950) can lead to where f denotes cloud fraction, and ‘‘cld” and ‘‘clr” represent cloud
improved accuracy in simulating radiation compared to overcast and clear-sky conditions, respectively.
two-stream approximation, especially at visible wavelengths
(Ding et al., 2009; Meador and Weaver, 1979; Qiu, 2001; Xie 2.1. Analytical formulation
et al., 2006; Yan and Stamnes, 2003).
Rapid development of satellite remote sensing has accelerated To derive solar radiation under a cloudy sky, the commonly
the studies of fast cloudy-sky RT models aimed at simulating the used single-layer cloud model (Coakley et al., 2005; Kattawar
upwelling radiances (also known as forward RT models) for speci- et al., 2016; Lawless et al., 2006; Ramanathan, 1987) is assumed
fic satellite channels (Minnis et al., 2011; Niu et al., 2007; Wang as a first-order approximation in the analytical formulation below.
et al., 2011, 2013; Wei et al., 2004; Xie et al., 2012b). Compared For simplification of the formulation, the light scattering and
to rigorous RT models, the fast RT models can substantially reduce absorption by clouds are assumed as occurring at the top of the
the computational time at the expense of accuracy since they often atmosphere (TOA), which may result in 1 W/m2 uncertainties
carry pre-computed lookup tables (LUTs) of cloud bi-directional on average in simulating solar radiation at the land surface. The
reflectance distribution function (BRDF) and bi-directional trans- uncertainties in solar radiation associated with this assumption
mittance distribution function (BTDF). The existing fast RT models are tested and discussed in Appendix A. Also note that only the
for cloudy-skies are not favorable candidates for solar applications scattering and absorption by cloud particles are considered within
because (1) they are developed for narrow satellite channels that the cloud layer while the absorption by the atmospheric gases is
are sensitive/insensitive to the amount of aerosol, cloud or specific computed in other parts of the atmosphere. As the atmospheric
trace gases while solar applications desire broadband radiation in gases are optically thin compared to the total atmospheric column
the complete solar region; (2) they need significantly more the diffuse absorption by the atmospheric gases can be considered
computational time to compute angular-dependent radiances negligible.
and convert them to downwelling irradiance for solar applications; Under the above assumptions, the direct solar flux in the
(3) absorption by trace gases under clouds is more important in the downwelling direction at the surface can be given by
simulation of downwelling radiation. However, the application of
LUTs for cloud BRDF and BTDF can be used as guide in the develop- F d ¼ l0 F 0 T cld clr
dd T dd ð2aÞ
ment of fast RT models for solar applications.
where l0 is the cosine of the solar zenith angle, F0 is the radiative
In this study, we developed a fast all-sky radiation model for
solar applications (FARMS) to provide users a new option of flux at the TOA, T cld
dd is the transmittance of the cloud for direct

efficiently simulating solar irradiance at the land surface with less incident radiation and direct outgoing radiation after scattering by
than 5% overall uncertainty compared to surface observations. This clouds, and T clr
dd is the transmittance of the clear-sky atmosphere
new model takes the advantage of the existing RT models by pre- relative to direct incident and output fluxes. Thus, DNI is given by
computing cloud transmittance and reflectance. Instead of directly Fd
using LUTs that require substantial computer memory and data DNI ¼ ¼ F 0 T cld clr
dd T dd ð2bÞ
l0
storage, the cloud transmittances and reflectances are parameter-
ized to increase computational efficiency. The computational time In the derivation below, F represents the radiative flux in the
for the new model can be significantly reduced compared to two- downwelling direction. The first character of the subscript
stream approximation since the complex solution of solving the RT indicates whether the incident radiation we consider is direct
equation is not involved. Moreover, FARMS uses the computation (‘‘d”) or diffuse (‘‘u”). The second character of the subscript repre-
from more discrete ordinate streams than two-stream approxima- sents the state of the outgoing radiation after atmospheric scatter-
tion, which can potentionally improve its accuracy. ing. For example, ‘‘dd” represents the incident and outgoing fluxes
 Y. Xie et al. / Solar Energy 135 (2016) 435–445 437

Fig. 1. Flowchart of the computation of GHI, DNI and DHI using FARMS.

that are both direct. Similarly subscript, ‘‘uu”, represents the inci- where Rs is surface reflectance. Rclr cld
uu and Ruu are aerosol and cloud
dent and outgoing fluxes that are both diffuse. The subscript, reflectances for diffuse fluxes, respectively. It can be seen that the
‘‘du”, represents direct incidence and diffuse outgoing. second term in the right hand side of Eq. (6) represents the first
The first order total downward flux is order downward flux reflected by land surface and reflected back
to surface by the aerosol in the atmosphere. The third term on the
F 1 ¼ l0 F 0 T cld
dd T dt þ l0 F 0 T du T uu
clr cld clr
ð3Þ
right hand side of Eq. (6) denotes the first order downward flux
where T clr
is the transmittance of the clear-sky atmosphere for
dt
reflected by both land surface and cloud, that comes back to the
direct incidence and total outgoing flux. Thus, land surface.
Then the diffuse horizontal irradiance (DHI) can be given by
T clr clr clr
dt ¼ T dd þ T du ð4Þ
DHI ¼ F total  F d ð7Þ
In the computation of T clruu , the transmitted diffuse radiation
through the cloud layer is assumed isotropic though cloud In Eqs. (2)–(7), T clr
dd , T clr
anddu , T clr
uu , Rclr
can be computed by REST2
uu

transmittance is exactly computed. Thus, the isotropic assumption model for a non-reflecting land surface.
only slightly affects the computation of transmittance of the atmo- The cloud transmittance for direct flux can be given on the basis
of the Beer–Bouguer–Lambert law for extinction (Liou, 2002):
sphere. Since radiance, I, is independent of zenith angle, T clr
uu can be
s=l0
approximated by T cld
dd ¼ e ð8Þ
R1
l
IT clr l l
dd ð Þd where s is the cloud optical thickness in the solar region. To numer-
T clr
uu 
0
R1
0
l l
Id
ð5Þ
ically solve for solar radiation in the direct and diffuse directions,
Z 1 i.e., Eqs. (2) and (7), T cld cld
du and Ruu are computed and parameterized
¼2 l
T clr l l
dd ð Þd in Sections 2.2 and 2.3, respectively.
0

where l is the cosine value of the zenith angle for an individual 2.2. Parameterization of diffuse transmittances of clouds
solar beam.
The total downwelling flux (or GHI) for all surface conditions is To circumvent the complex solutions of the RT equation, a LUT
given by considering the multiple reflectance between land surface of cloud transmittance for diffuse flux, T cld
du , is simulated using
and cloud: RRTM (Mlawer et al., 1997; Oreopoulos and Barker, 1999) with
F total ¼ F 1 þ F 1 Rs Rclr clr cld clr sixteen-stream DISORT (Stamnes et al., 1988) for all possible cloud
uu þ F 1 Rs T uu Ruu T uu þ . . .
conditions. RRTM utilizes the parameterizations of cloud optical
¼ F 1 þ F 1 Rs ðRclr cld clr2
uu þ Ruu T uu Þ þ . . . ð6Þ properties over fourteen contiguous bands and simulates radiation
clr2 1 in the individual bands as well as the integrated shortwave region.
¼ F 1 ½1  Rs ðRclr
uu þ Rcld
uu T uu Þ
T cld
du can be simply computed by placing a cloud layer at the top of
438 Y. Xie et al. / Solar Energy 135 (2016) 435–445

the TOA and considering the downwelling irradiance above and

under the cloud layer. In this study, T cld du is computed for cloud
optical thickness s = 0.0001, 0.01, 0.1, 0.2, . . .,0.5, 1.0, . . ., 3.0, 4.0,
. . ., 8.0, 10.0, . . ., 20.0, 25.0, . . ., 50.0, 60.0, . . ., 100.0, 140.0, . . ., 300.0,
cloud effective particle size De = 5.0, 6.0, . . ., 10.0, 12.0, . . .,
30.0, 35.0, . . ., 60.0, 70.0, . . ., 120.0 lm and l0 = 0.02, 0.04, . . ., 1.0.
The LUT is computed as a function of cloud optical thickness as it
directly represents the extinction of solar radiation and is often
retrieved by satellite-based measurements of clouds. To adapt
FARMS to models that provide cloud microphysical variables
rather than optical thickness, cloud optical thickness can be
obtained from cloud liquid water path (LWP) or ice water path
(IWP) and effective particle size information from formulas such as
s ¼ LWPða þ b=re Þ ð9Þ
where re is effective radius of water clouds, a and b are parameters
given by Hu and Stamnes (1993). Details on the determination of
cloud optical thickness from LWP or IWP are not restated here as
they have been extensively introduced by Hu and Stamnes (1993),
Fu (1996), Matrosov et al. (2003), and Xie et al. (2012a). Cloud opti-
cal thickness is calculated externally and used as an input to FARMS.
Therefore, cloud optical thickness may or may not be influenced by
effective particle size depending on how it is computed. The cloud
scattering properties, including single scattering albedo and asym-
metry factor, are determined by cloud effective particle size and
parameterized by Hu and Stamnes (1993) and Fu (1996). For clouds
composed of water droplets, the cloud effective particle size is
defined as
De ¼ 2  re ð10Þ
For ice clouds, the definition of cloud effective particle size follows
the description of Fu (1996).
Fig. 2 illustrates the simulated T cld
du for water and ice clouds as
functions of cloud optical thickness and solar zenith angle. It is
seen that the diffuse transmittances increase with cloud optical
thickness due to the increasing scattering by clouds when they
are optically thin. The diffuse transmittances increase with optical
thickness for values of s from 0.2–5 and decrease after that. The
cloud diffuse transmittances are also sensitive to solar zenith
angle. When clouds are very thin, larger solar zenith angles lead
to greater T cld
du because of the longer path of the radiation in the
du (dots) for (a) water clouds with De = 16 lm and (b) ice clouds
Fig. 2. Simulated T cld
cloud. Also, T cld
du decreases with increasing solar zenith angle for with De = 70 lm as functions of cloud optical thickness and solar zenith angle and
thicker clouds. The effective particle sizes for the water and ice their parameterization (solid lines).
clouds are assumed to be 16 and 70 lm, respectively. Compared
to cloud optical thickness and solar zenith angle, the cloud
particle size has lower impact on the radiation at visible
For water clouds, the variables of Eq. (11) are parameterized as
wavelengths as demonstrated by previous studies (King et al.,
1996; Xie, 2010). sP ðDe ; l0 Þ ¼ ½2:885 þ 0:002ðDe  60Þl0  0:007347 ð12aÞ

du;P ðDe ; l0 Þ ¼ 0:7846½1 þ 0:0002ðDe  60Þl0

To reduce the computing burden of FARMS, T cld du should be T cld 0:1605
ð12bÞ
parameterized as a function of solar zenith angle, cloud optical
Dðl0 Þ ¼ 0:644531l0 þ 1:20117 þ 0:129807l1 0
thickness and particle size since checking values from the LUT con-
sumes considerable computational resource. As Fig. 2 clearly  0:00121096l2 0 þ 1:52587  10 l0
7 3
ð12cÞ
shows, the distribution of T cld
du can be approximated by exponential
f ðs; De ; l0 Þ ¼ 1 þ sinh½0:012l0 ðs  sP ðDe ; l0 ÞÞ ð12dÞ
functions. We parameterize T cld
du in the form of where sinh denotes the hyperbolic sine function.
2  2 3 For ice clouds,
6 log10 sP ðDse ;l Þ 7 
2:8487l0  0:0029 for De 6 26 lm
du ¼ f ðDe ; s; l0 ÞT du;P ðDe ; l0 Þ exp 4
T cld cld
ð11Þ
0
5 sP ðDe ; l0 Þ ¼
Dðl0 Þ
½2:8355 þ 0:006ð100  De Þl0  0:00612 for De > 26 lm
ð13aÞ
where sP is the cloud optical thickness as T cld
du reaches its maximum,
du;P ðl0 Þ ¼ 0:756l0
T cld ð13bÞ
0:0883

du;P ðDe ; l0 Þ is the maximum of T du for each given solar zenith angle,
T cld cld
Dðl0 Þ ¼ 0:0549531l0 þ 0:617632 þ 0:17876l1
0  0:002174l0
2

D(l0) is a function that is related to the width of T cld

du , and f(s, De, l0) ð13cÞ
is a scaling factor that adjusts T cld
du especially when cloud optical f ðs; De ; l0 Þ ¼ 1 þ sinh½0:01l0 ðs  sP ðDe ; l0 ÞÞ ð13dÞ
thickness is large.
 Y. Xie et al. / Solar Energy 135 (2016) 435–445 439

Details on parameterization of the functions in Eq. (11) can be For water cloud, the simulated Rcld
uu can be fitted using
found in Appendix B. (
The solid lines of Fig. 2 represent the parameterization using 0:107359s h for s < 1i
Rcld ¼ 10 sÞ
2 ð14aÞ
Eqs. (11)–(13). A generally good agreement is shown though larger uu 1:03  exp  ð0:5þlog
3:105
for sP1
differences can be found for extremely thin or thick clouds. Note
that cloud transmittance has minor impacts on solar radiation For ice cloud, Rcld
uu can be given by
when the cloud is extremely thin or thick. Thus, FARMS should (
have reasonable accuracy when it is used to simulate solar irradi- 0:094039s h for s < 1i
Rcld ¼ 10 sÞ
2 ð14bÞ
ance. Further evaluation on the performance of FARMS is discussed uu 1:02  exp  ð0:5þlog
3:25
for sP1
in Section 3.
It is obvious from the solid lines of Fig. 3 that the parameteriza-
tion accurately represents the average of Rcld
uu over all the Des.
2.3. Parameterization of diffuse reflectances by clouds

Similar to T cld cld 3. Performance evaluation

du , Ruu can also be simulated using RRTM with
sixteen-stream DISORT when the incident radiation is assumed
Two-stream approximation (Irvine, 1965, 1968; Meador and
isotropic. Fig. 3 shows the simulated Rclduu as functions of cloud
Weaver, 1979; Sagan and Pollack, 1967) adopted by RRTM for
optical thickness and particle size. For both water and ice clouds,
GCM applications (also known as RRTMG) and the RRTM with an
Rcld
uu is slightly sensitive to De, which can then be parameterized expansion of discrete ordinate streams provide time-efficient
using simple functions of cloud optical thickness. solutions of the RT equation and have been widely used in the sim-
ulation of solar radiation (Pinker and Laszlo, 1992; Ruiz-Arias et al.,
2012; Skamarock and Klemp, 2008; Skamarock et al., 2005; Xie and
Liu, 2013; Xie et al., 2012a). To understand the performance of
FARMS, we compare the accuracy and efficiency of the two-stream
approximation, RRTM and FARMS for the same computations.

3.1. Accuracy in the computation of cloud transmittance

RRTM can compute solar radiation with a selection of discrete

ordinate streams ranging from four to sixteen. Although the use
of more discrete ordinate streams can provide improved accuracy
of solar radiation, Ding et al. (2009) discovered that four and
thirty-two terms of Legendre polynomials in the cloud scattering
phase function are normally sufficient for DISORT to compute radi-
ance at infrared and visible wavelengths, respectively. However,
the uncertainty related to the number of discrete ordinate streams
can be averaged down when irradiance is developed from the
integration of radiance over space (Li et al., 2015).
We use RRTM with sixteen-stream DISORT as a benchmark to
simulate cloud total transmittances, i.e. T clddt , for all possible cloud
conditions as described in Section 2.2. The same computations
are performed using RRTMG which uses the two-stream
approximation and FARMS and evaluated using those from the
sixteen-stream approximation. Fig. 4 shows the relative errors of
water cloud transmittance from two-stream approximation and
FARMS as compared with the sixteen-stream approximation. The
red and blue colors in the figure denote under- and
over-estimation of the transmittance from the sixteen-stream
approximation, respectively. Compared to the sixteen-stream
approximation, two-stream approach significantly overestimates
transmittances for large solar zenith angles, unlike FARMS which
is significantly more accurate. For small solar zenith angles, the
transmittances are slightly underestimated by two-stream approx-
imation when the clouds are thick. This underestimation becomes
much more obvious when using FARMS as can be also seen in
Fig. 2a. The parameterization of FARMS slightly underestimates
the transmittances though the relative differences are large due
to the low transmittances.
Fig. 5 is the same as Fig. 4 except for ice clouds with De = 70 lm.
Similar to water cloud transmittances, those of ice clouds are
significantly overestimated by two-stream approximation for large
solar zenith angles. For small solar zenith angles, two-stream
approximation also increasingly overestimates ice cloud
Fig. 3. Simulated Ruu (dots) for (a) water clouds and (b) ice clouds as functions of transmittances with increasing cloud optical thickness. In Fig. 5b,
cloud optical thickness and particle size and their parameterization (solid lines). uncertainties of ice cloud transmittance occur for large solar zenith
440 Y. Xie et al. / Solar Energy 135 (2016) 435–445

dt (De = 16 lm) simulated by (a) two-stream

Fig. 4. Relative error of water cloud T cld dt (De = 70 lm) simulated by (a) two-stream
Fig. 5. Relative error of ice cloud T cld
approximation and (b) FARMS. approximation and (b) FARMS.

angles or when cloud is thick. Since those solar and cloud conditions
are associated with low solar radiation, they are normally less where x represents cloud transmittance simulated by RRTM with
important compared to regions associated with higher solar energy. sixteen-stream DISORT, xm denotes the simulations using two-
To numerically evaluate the overall performance of two-stream stream approximation or FARMS. Table 1 shows those statistics
approximation and FARMS, we consider various statistical mea- associated with the results of Figs. 4 and 5. It can be found that both
sures including percent error (PE), absolute percent error (APE), two-stream approximation and FARMS have reasonable perfor-
mean bias error (MBE), mean absolute bias error (MAE), and mance in computing cloud transmittance under frequent solar
root-mean-square error (RMSE) of simulated cloud transmittance and cloud conditions. The PEs of FARMS are 0.0425% and 3.0987%
and GHI defined as follows, for typical water and ice clouds, respectively. Compared to two-
Pn stream approximation, FARMS significantly reduces PE and MBE
i¼1 ðxm  xÞ while its APE and MAE are slightly smaller.
PE ¼ Pn ð15aÞ
i¼1 x
Pn
i¼1 jxm  xj
APE ¼ Pn ð15bÞ 3.2. Accuracy in the computation of solar radiation
i¼1 x
1 Xn The accuracy of FARMS can also be evaluated by comparing
MBE ¼ ðxm  xÞ ð15cÞ
n i¼1
satellite-based retrievals of atmospheric properties and surface
1 X n
MAE ¼ jxm  xj ð15dÞ measurements of solar radiation. In FARMS, we use data from the
nsffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i¼1
AVHRR Pathfinder Atmospheres-Extended (PATMOS-x) algorithms
Pn 2 (Heidinger et al., 2013; Sengupta et al., 2014) applied to the
i¼1 ðxm  xÞ
RMSE ¼ ð15eÞ Geostationary Operational Environmental Satellite (GOES) data
n
where the atmospheric properties for clear-sky RT models, e.g.
 Y. Xie et al. / Solar Energy 135 (2016) 435–445 441

Table 1
Statistics of PE, APE, MBE, MAE, and RMSE of simulated cloud transmittance and GHI using two-stream approximation and FARMS.

RT model PE (%) APE (%) MBE MAE RMSE

Transmittance for water cloud
Two stream 3.7674 4.7164 0.0151 0.0189 0.0308
FARMS 0.0425 2.8716 0.0002 0.0115 0.0148
Transmittance for ice cloud
Two stream 7.7293 7.7293 0.03 0.03 0.0325
FARMS 3.0987 5.6971 0.012 0.0221 0.027

RT model PE (%) APE (%) MBE (W/m2) MAE (W/m2) RMSE (W/m2)
GHI over ARM SGP site
Two stream 7.109 31.7935 20.2983 90.7798 132.749
FARMS 1.8936 31.886 5.4068 91.044 130.286

30-min precipitable water vapor (PWV), were produced 3.3. Comparison of computing time
by the National Oceanic and Atmospheric Administration
(NOAA)/National Centers for Environmental Prediction (NCEP). Fig. 8a compares the computing time of FARMS, RRTMG with
Since ozone has limited impact on solar radiation when clouds two-stream approximation and RRTM with the four- to sixteen-
are present the ozone amount in the atmosphere is assumed to stream approximations for the cloud transmittances discussed in
be 300 DU in the absence of ozone from the PATMOS-x data. In Section 2.2. The computing time is recorded based on the
two-stream approximation, the clear-sky atmospheric profile fol- computations using National Renewable Energy Laboratory’s
lows the 1976 US Standard Atmosphere where the PWV values (NREL’s) High Performance Computing (HPC) system which has
are scaled by the PATMOS-x data.
For both two-stream approximation and FARMS, the cloud
microphysical and optical properties of the GOES data are available
every 30 min for 4-km resolution pixels. Daily averaged aerosol
optical depth (AOD) is derived from the Moderate Resolution
Imaging Spectroradiometer (MODIS) and Multi-angle Imaging
SpectroRadiometer (MISR) AOD product that is scaled using the
surface-based Aerosol Robotic Network (AERONET) data
(Sengupta et al., 2014). The total AOD is used by FARMS to compute
the clear-sky transmittance and reflectance as described by
Gueymard (2008). The broadband shortwave surface albedo is
given by a climatology using NASA’s Modern-ERA Retrospective
Analysis for Research and Applications (MERRA) data (Rienecker
et al., 2011).
The surface-based solar radiation, with a temporal resolution of
1 min, is observed at the Atmospheric Radiation Measurement
(ARM) Southern Great Plain (SGP) site in Oklahoma (Stokes and
Schwartz, 1994). The solar radiation is then averaged over
30 min and collocated with the satellite products which are used
as inputs to two-stream approximation and FARMS to simulate
GHIs. Fig. 6 demonstrates the comparison of the GHIs between
the model simulations and ARM measurements where the colors
represent the density of the data. A total number of 9669 scenarios
associated with overcast cloudy conditions are selected during
2009–2012. It is obvious that the simulations from both
two-stream approximation and FARMS have good agreement to
the surface measurements. Based on Table 1, the PE and MBE for
FARMS computation of GHI are 1.8936% and 5.4068 W/m2,
respectively, much better than two-stream approximation.
Two-stream approximation and FARMS perform similarly when
evaluated by APE, MAE, and RMSE. Note that the uncertainties in
Fig. 6 come from the satellite-based cloud property retrievals,
other input uncertainties as well as the RT models.
Fig. 7 demonstrates the monthly and daily mean GHIs from
Fig. 6. Compared to ARM measurements, the model simulations
generally underestimate GHI especially in winter. As two-stream
approximation and FARMS slightly overestimate cloud transmit-
tance, the bias in GHI is probably because the MERRA climatology
data underestimates surface albedo when there is snow. Compared
to two-stream approximation, the solar radiation from FARMS has
better agreement with ARM surface measurements, which indi-
cates FARMS has a better simulation of multiple reflectance Fig. 6. Comparison of GHIs between surface measurements over ARM SGP site and
between land surface and cloud. those simulated using satellite retrieval of cloud properties and (a) two-stream
approximation and (b) FARMS.
442 Y. Xie et al. / Solar Energy 135 (2016) 435–445

Fig. 8. Computing time of (a) cloud transmittance and (b) surface radiation over
Fig. 7. Comparison of (a) monthly mean and (b) daily mean GHIs between surface ARM SGP site using FARMS, RRTMG with two-stream approximation and RRTM
measurements over ARM SGP site and those simulated using satellite retrieval of with the four- to sixteen-stream approximations.
cloud properties and two-stream approximation (blue) and FARMS (red). (For
interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this article.)
4. Concluding remarks

By utilizing a clear-sky radiative transfer model, REST2, and

6912 Intel Xeon E5-2670 ‘‘SandyBridge” processors, 24192 Intel LUTs of cloud transmittances and reflectances pre-computed by
Xeon E5-2695v2 ‘‘IvyBridge” processors and 576 Intel Phi many- RRTM, a fast all-sky radiative transfer model, FARMS, is developed
core co-processors providing a total capability of 1190 Trillion to simulate surface radiation for solar applications. For simplicity
floating point calculations per second. During the computations in application and reducing the computing time, the LUTs are
of cloud transmittance, the efficiency for water and ice cloud do parameterized by exponential functions of solar zenith angles
not show much difference. For all the water cloud conditions, and cloud microphysical and optical properties. REST2 is used to
two-stream approximation uses 31 min and 31 s on a single pro- compute the light scattering and absorption by aerosols and trace
cessor while sixteen-stream consumes 3 h and 30 min. For the gases in the atmosphere. The all-sky surface radiation is calculated
same computation, FARMS dramatically reduces the computing by accounting for the influences of the clear-sky atmosphere,
time to 0.018 s. clouds, and the multiple reflections between cloud and land
Fig. 8b illustrates the computing time related to the surface surface.
radiation over the ARM SGP site. In addition to computing the The performance of FARMS is evaluated using RRTMG with two-
radiative transfer within clouds, these computations also account stream approximation and RRTM with four- to sixteen-stream
for the light scattering and absorption by the clear sky and the approximations where the computation by the sixteen-stream
multiple reflectance between cloud and land surface. FARMS, approximation serves as a benchmark. It is found that two-
RRTMG with two-stream approximation and RRTM with sixteen- stream approximation overestimates cloud transmittance in large
stream approximation use 2 s, 12 min and 32 s and 2 h and solar zenith angles unlike FARMS. Compared to the two- and
2 min, respectively. FARMS is not as efficient in computing solar sixteen-stream approximations, FARMS underestimates cloud
radiation compared to cloud transmittance calculations as shown transmittance for thick clouds. However, both FARMS and two-
in Fig. 8a but is still over 300 times faster than two-stream stream approximation show similar accuracy for the most cloud
approximation. This is because multiple computations of Eq. (5) conditions of relevance to solar energy applications. The perfor-
is used to compute clear-sky transmittance for diffuse radiation. mance of FARMS has also been evaluated using measurements at
To further reduce the computing time of FARMS, a parameteriza- the ARM SGP site. The solar radiation is simulated using the
tion of Eq. (5) is required. two-stream approximation and FARMS, with the cloud and other
 Y. Xie et al. / Solar Energy 135 (2016) 435–445 443

atmospheric properties being determined from satellite retrievals

and reanalysis models. Compared to ARM measurements during
2009–2012, two-stream approximation and FARMS have similar
performance in terms of APE, MAE, and RMSE. However, FARMS
performs better than two-stream approximation in terms of PE
and MBE. The comparison of the computing time indicates FARMS
is much more efficient than other RT models used in weather, cli-
mate and solar energy applications. Thus, it can be used widely for
various applications because of its accuracy and efficiency.
Several points are noteworthy. The recent emergence of solar
energy as an alternative to the more conventional domestic energy
sources has resulted in a demand for deriving global solar resource
from satellite data. NREL has utilized GOES data and developed the
National Solar Radiation Data Base (NSRDB) covering the North and
South America at a resolution of 4 km for every 30 min of
1998–2014 (Sengupta et al., 2014). The cloudy-sky solar radiation
from the NSRDB originally used a product of the Satellite Algorithm
for Surface Radiation Budget (SASRAB) model (Pinker and Laszlo,
1992) where the relationship between solar reflectance and trans- Fig. A1. Comparison of GHIs over ARM SGP site simulated by two-stream
mittance is given by a computation of two-stream technique approximation where cloud heights are given by PATMOS-X and a fixed value,
(Joseph et al., 1976). As discussed in Section 3, FARMS demonstrates 20 km.
its potential to better accommodate the increasing spatial and tem-
poral resolution from the next generation of satellites. Due to the
simplicity of the parameterization, FARMS also accommodates
parallel computation, which further reduces the computing time.
It is therefore currently being used as the RT model in the NSRDB.
As two-stream approximation has been widely used in NWP
models, e.g. the Weather Research and Forecasting (WRF) model
(Skamarock et al., 2005), other applications of FARMS might be
those in NWP models to provide higher temporal resolution solar
forecasting; efforts on updating the NWP models are underway on
this. Typically, for efficiency, NWP models have a radiation time step
that is an order of magnitude longer than its dynamical/microphys-
ical time step, so the cloud-radiation interaction is time-sampling
the clouds and not making full use of the resolved-cloud time vari-
ation. FARMS allows the surface radiation to respond to cloud
changes at the full model time resolution when applied at the mod-
el’s dynamical/microphysical time step. Finally, FARMS follows the
wavelength range from RRTM, i.e. 0.2–12.2 lm, to consider the con-
tribution from the broadband solar radiation. Since Photovoltaic
(PV) technology is spectral sensitive (Argazzi et al., 1994), future
studies on the RT models for narrow spectral bands can serve as a
theoretical framework for assessing the PV plant performance.

Acknowledgements

This work was supported by the U.S. Department of Energy

under Contract No. DE-AC36-08GO28308 with the National
Renewable Energy Laboratory. The U.S. Government retains and
the publisher, by accepting the article for publication, acknowl-
edges that the U.S. Government retains a nonexclusive, paid-up,
irrevocable, worldwide license to publish or reproduce the pub-
lished form of this work, or allow others to do so, for U.S. Govern-
ment purposes.
We would also like to acknowledge the DOE Solar Program for
supporting this research. Specifically we would like to thank Minh
Le, Dr. Ranga Pitchumani and Dr. Subhashree Mishra for their sup-
port and encouragement. We also thank the four anonymous
reviewers for their constructive comments.

Appendix A. The impact of cloud height on the simulation of

solar radiation

For simplification of the formulation, the light scattering and du;P for (a) water and (b) ice clouds as functions of De and l0 (dots) and
Fig. B1. T cld
absorption by clouds are assumed as occurring at the TOA. Thus, parameterization (solid lines) using Eqs. (12) and (13).
444 Y. Xie et al. / Solar Energy 135 (2016) 435–445

surface. Thus, compared to the other uncertainties, e.g., those in

the retrieval of cloud microphysical and optical properties, the
assumption of a cloud layer at the TOA has minor impact on the
performance of FARMS.

Appendix B. Parameterization of the functions in Eq. (11)

To parameterize T cld
du using exponential functions, it is important
to accurately represent the functions in Eq. (11). Fig. B1 demon-
du;P as a function of De and l0. It can be seen that T du;P
strates T cld cld

has near-linear relationships with l0. De slightly impacts the

du;P for water clouds when l0 is large. For ice clouds
magnitude of T cld
(see Fig. B1(b)), T cld
du;P is almost independent on De. Considering a
median De (60 lm), we test multiple combinations of coefficients
(see Eqs. (12b) and (13b)) to best fit the numerical simulations of
T cld
du;P . Then we slightly adjust the coefficients for water cloud cases
based on the dependence on De.
Fig. B2 shows sP as a function of De and l0. It is seen that sP
increases with l0 while it is also sensitive to the variation of De.
du;P , sP can be also represented
Similar to the parameterization of T cld
by functions of De and l0 (see Eqs. (12a) and (13a)). Compared to
du;P , unavoidable uncertainty exists in the parameterization of sP
T cld
as can be seen in Fig. B2. The impact of this uncertainty will be
understood in the performance evaluation of Eq. (11) and the com-
putation of solar radiation.
du;P and sP are parameterized, we further test multiple coef-
As T cld
ficients of D(l0) to create best fit of the simulation of T cld
du . As can be
seen in Fig. 2, the simulated T cld
du is not perfectly asymmetrical
around sP , especially for extremely small and great cloud optical
thickness. To further reduce the uncertainty in Eq. (11), we intro-
duce a scaling factor, f(s, De, l0), to adjust the magnitude of T cld
du
when the cloud is thick.

References

Argazzi, R., Bignozzi, C., Heimer, T., Castellano, F., Meyer, G., 1994. Enhanced
spectral sensitivity from ruthenium(II) polypyridyl based photovoltaic devices.
Inorg. Chem. 33 (25), 5741–5749.
Atwater, M., Ball, J., 1978. A numerical solar radiation model based on standard
Fig. B2. sP for (a) water and (b) ice clouds as functions of De and l0 and
meteorological observations. Sol. Energy 21 (3), 163–170.
parameterization (straight lines) using Eqs. (12) and (13).
Berk, A., Bernstein, L., Anderson, G., Acharya, P., Robertson, D., Chetwynd, J., Adler-
Golden, S., 1998. MODTRAN cloud and multiple scattering upgrades with
application to AVIRIS. Remote Sens. Environ. 65 (3), 367–375.
Bird, R., Hulstrom, R., 1981. A Simplified Clear Sky Model for Direct and Diffuse
the absorption of trace gases within the cloud layer is considered Insolation on Horizontal surfacesRep. Solar Energy Research Institute, Golden,
under the cloud with the other part of the atmosphere. To numer- CO.
ically understand the uncertainty associated with this assumption, Chandrasekhar, S., 1950. Radiative Transfer. Oxford Univ. Press, Oxford, p. 393.
Chen, Y., Weng, F., Han, Y., Liu, Q., 2008. Validation of the community radiative
we simulated solar radiation over the ARM SGP site using two- transfer model by using CloudSat data. J. Geophys. Res. 113 (D8), 2156–2202.
stream approximation and cloud product introduced in Section 3.2 Clough, S., Iacono, M., Moncet, J., 1992. Line-by-line calculations of atmospheric
where cloud heights and other properties are retrieved by the fluxes and cooling rates: application to water vapor. J. Geophys. Res. 97 (D14),
15761–15785.
PATMOS-X algorithm. The whole atmosphere is divided into 90 Coakley, J.A., Friedman, M.A., Tahnk, W.R., 2005. Retrieval of cloud properties for
layers in the two-stream approximation. The cloud whose physical partly cloudy imager pixels. J. Atmos. Oceanic Technol. 22 (1), 3–17.
thickness is assumed as 0.5 km is inserted into one of the layers Ding, S.G., Xie, Y., Yang, P., Weng, F.Z., Liu, Q.H., Baum, B., Hu, Y.X., 2009. Estimates of
radiation over clouds and dust aerosols: optimized number of terms in phase
based on cloud height. The solar radiation is also simulated using function expansion. J. Quant. Specrtrosc. Radiat. Transfer 110 (13), 1190–1198.
a cloud height of 20 km while all the other cloud properties, Fu, Q., 1996. An accurate parameterization of the solar radiative properties of cirrus
including cloud physical thickness, are kept the same. Fig. A1 com- clouds for climate models. J. Climate 9 (9), 2058–2082.
Gueymard, C., 1989. A two-band model for the calculation of clear sky solar
pares the simulated GHIs using different cloud heights where very
irradiance, illuminance, and photosynthetically active radiation at the earth’s
tiny differences are shown for all cloud conditions. Over the 9669 surface. Sol. Energy 43 (5), 253–265.
scenarios, an averaged underestimation of 1.005 W/m2 associated Gueymard, C., 2003. Direct solar transmittance and irradiance predictions with
broadband models. Part I: detailed theoretical performance assessment. Sol.
with the cloud height of 20 km is discovered. This is related to
Energy 74 (5), 355–379.
the overestimation of atmospheric absorption by the trace gases Gueymard, C., 2008. REST2: high-performance solar radiation model for cloudless-
above realistic cloud layers. The impact of cloud height on solar sky irradiance, illuminance, and photosynthetically active radiation – validation
radiation simulation is extremely small as the concentration of with a benchmark dataset. Sol. Energy 82 (3), 272–285.
Heidinger, A., Foster, M., Walther, A., Zhao, X., 2013. The pathfinder atmospheres
trace gases above clouds is significantly low and cloud physical extended (PATMOS-x) AVHRR climate data set. Bull. Amer. Meteor. Soc. 95,
positions only affect multiple reflections between cloud and land 909–922.
 Y. Xie et al. / Solar Energy 135 (2016) 435–445 445

Hoyt, D., 1978. A model for the calculation of solar global insolation. Sol. Energy 21 Woollen, 2011. MERRA: NASA’s modern-era retrospective analysis for research
(1), 27–35. and applications. J. Climate 24 (14), 3624–3648.
Hsu, N., Tsay, S., King, M., Herman, J., 2006. Deep blue retrievals of Asian aerosol Rigollier, C., Bauer, O., Wald, L., 2000. On the clear sky model of the ESRA—European
properties during ACE-Asia. IEEE Trans. Geosci. Remote 44 (11), 3180–3195. solar radiation atlas—with respect to the Heliosat method. Sol. Energy 68 (1),
Hu, Y.X., Stamnes, K., 1993. An accurate parameterization of the radiative properties 33–48.
of water clouds suitable for use in climate models. J. Climate 6 (4), 728–742. Ruiz-Arias, J., Gueymard, C., Dudhia, J., Pozo-Vazquez, D., 2012. Improvement of the
Ineichen, P., 2008. A broadband simplified version of the solis clear sky model. Sol. weather research and forecasting (WRF) model for solar resource assessments
Energy 82 (8), 758–762. and forecasts under clear skies, paper presented at: Proc. Amer. Sol. Energy Soc.,
Irvine, W.M., 1965. Multiple scattering by large particles. Astrophys. J. 142, 1563– Denver, CO.
1575. Sagan, C., Pollack, J.B., 1967. Anisotropic nonconservative scattering and the clouds
Irvine, W.M., 1968. Multiple scattering by large particles II. Optically thick layers. of venus. J. Geophys. Res. 72, 469–477.
Astrophys. J. 152, 823–834. Sengupta, M., Ackerman, T., 2003. Investigating anomalous absorption using surface
Joseph, J., Wiscombe, W., Weinman, J., 1976. The delta-Eddington approximation for measurements. J. Geophys. Res. 108 (D24). http://dx.doi.org/10.1029/
radiative transfer. J. Atmos. Sci. 33, 2452–2459. 2003JD003411.
Kattawar, G., Yang, P., You, Y., Bi, L., Xie, Y., Huang, X., Hioki, S., 2016. Polarization of Sengupta, M., Heidinger, A., Miller, S., 2010. Validating an operational physical
light in the atmosphere and ocean. In: Kokhanovsky, A. (Ed.), Light Scattering method to compute surface radiation from geostationary satellites. In: Proc.
Reviews 10. Springer, Chichester, UK, pp. 3–39. SPIE, San Diego, CA.
King, M.D., Tsay, S.C., Platnick, S.E., Wang, M., Liou, K.N., 1996. Cloud retrieval Sengupta, M., Habte, A., Gotseff, P., Weekley, A., Lopez, A., Molling, C., Heidinger, A.,
algorithms for MODIS: optical thickness, effective particle radius, and 2014. A physics-based GOES product for use in NREL’s national solar radiation
themodynamic phase. In: MODIS Algorithm Theoretical Basis Document database. In: European Photovoltaic Solar Energy Conference and Exhibition,
ATBD-MOD-05. Available on-line at <http://eospso.gsfc.nasa.gov/atbd/modis/ Amsterdam, Netherlands.
atbdmod05new.html>, edited. Skamarock, W., Klemp, J., 2008. A time-split nonhydrostatic atmospheric model for
Kotchenova, S., Vermote, E., Matarrese, R., Klemm, F., 2006. Validation of a vector research and NWP applications. J. Comp. Phys. 227, 3465–3485.
version of the 6S radiative transfer code for atmospheric correction of satellite Skamarock, W., Klemp, J., Dudhia, J., Gill, D., Barker, D., Wang, W., Powers, J., 2005. A
data. Part I: path radiance. Appl. Opt. 45 (26), 6762–6774. description of the advanced research WRF version 2, NCAR Technical note,
Lacis, A., Hansen, J., 1974. A parameterization for the absorption of solar radiation in NCAR/TN-468+STR.
the earth’s atmosphere. J. Atmos. Sci. 31 (1), 118–133. Stamnes, K., Tsay, S.C., Wiscombe, W., Jayaweera, K., 1988. Numerically stable
Lawless, R., Xie, Y., Yang, P., 2006. Polarization and effective Mueller matrix for algorithm for discrete-ordinate-method radiative transfer in multiple scattering
multiple scattering of light by nonspherical ice crystals. Opt. Express 14 (14), and emitting layered media. Appl. Opt. 27 (12), 2502–2509.
6381–6393. Stokes, G.M., Schwartz, S.E., 1994. The atmospheric radiation measurement (ARM)
Li, J., Barker, H., Yang, P., Yi, B., 2015. On the aerosol and cloud phase function program: programmatic background and design of the cloud and radiation test
expansion moments for radiative transfer simulations. J. Geophys. Res. Atmos. bed. Bull. Amer. Meteor. Soc. 75 (7), 1201–1221.
120 (12), 12,128–112,142. Strow, L., Hannon, S., De Souza-Machado, S., Motteler, H., Tobin, D., 2003. An
Liou, K.N., 2002. An Introduction to Atmospheric Radiation, second ed. Academic overview of the AIRS radiative transfer model. IEEE Trans. Geosci. Remote 41
Press, Amsterdam, Boston, xiv, p. 583. (2), 303–313.
Long, C., Ackerman, T.P., Gaustad, K.L., Cole, J.N.S., 2006. Estimation of fractional sky Sun, Z., Liu, J., Zeng, X., Liang, H., 2012. Parametrization of instantaneous global
cover from broadband shortwave radiometer measurements. J. Geophys. Res. horizontal irradiance: cloudy-sky component. J. Geophys. Res. 117 (D14202).
111 (D11204). http://dx.doi.org/10.1029/2005JD006475. http://dx.doi.org/10.1029/2012JD017557.
Lorenz, E., Hurka, J., Heinemann, D., Beyer, H., 2009. Irradiance forecasting for the Sun, Z., Zeng, X., Liu, J., Liang, H., Li, J., 2014. Parametrization of instantaneous global
power prediction of grid-connected photovoltaic systems. IEEE J. Special Top. horizontal irradiance: clear-sky component. Q. J. R. Meteorol. Soc. 140, 267–
Earth Observations Remote Sens. 2, 2–10. 280.
Matrosov, S., Shupe, M., Heymsfield, A., Zuidema, P., 2003. Ice cloud optical Wang, C., Yang, P., Baum, B., Platnick, S., Heidinger, A., Hu, Y., Holz, R., 2011.
thickness and extinction estimates from radar measurements. J. Appl. Meteor. Retrieval of ice cloud optical thickness and effective particle size using a fast
42, 1584–1597. infrared radiative transfer model. J. Appl. Meteorol. Clim. 50 (11), 2283–2297.
Meador, W.E., Weaver, W.R., 1979. Two-stream approximations to radiative transfer Wang, C., Yang, P., Nasiri, S., Platnick, S., Baum, B., Heidinger, A., Liu, X., 2013. A fast
in planetary atmospheres: a unified description of existing methods and a new radiative transfer model for visible through shortwave infrared spectral
improvement. J. Atmos. Sci. 37, 630–643. reflectances in clear and cloudy atmospheres. J. Quant. Specrtrosc. Radiat.
Min, Q., Harrison, L.C., 1996. Cloud properties derived from surface MFRSR Transfer 116, 122–131.
measurements and comparison with GOES results at the ARM SGP site. Wei, H., Yang, P., Li, J., Baum, B., Huang, H., Platnick, S., Hu, Y., Strow, L., 2004.
Geophys. Res. Lett. 23 (13), 1641–1644. Retrieval of semitransparent ice cloud optical thickness from atmospheric
Minnis, P., Sun-Mack, S., Young, D., Heck, P.W., Garber, D.P., Chen, Y., Spangenberg, infrared sounder (AIRS) measurements. IEEE Trans. Geosci. Remote 42 (10),
D.A., Arduini, R.F., Trepte, Q.Z., Smith, W.L., Ayers, J.K., Gibson, S.C., Miller, W.F., 2254–2267.
Hong, G., Chakrapani, V., Takano, Y., Liou, K.N., Xie, Y., Yang, P., 2011. CERES Wendisch, M., Yang, P., 2012. Theory of Atmospheric Radiative Transfer. WileyVCH,
edition-2 cloud property retrievals using TRMM VIRS and TERRA and AQUA Weinheim, Germany.
MODIS data, part I: algorithms. IEEE Trans. Geosci. Remote 49 (11), Xie, Y., 2010. Study of Ice Cloud Properties from Synergetic use of Satellite
4374–4400. Observations and Modeling Capabilities. Texas A&M University, College Station,
Mlawer, E.J., Taubman, S.J., Brown, P.D., Iacono, M.J., Clough, S.A., 1997. RRTM, a TX, p. 150.
validated correlated-k model for the longwave. J. Geophys. Res. 102, 16663– Xie, Y., Liu, Y.G., 2013. A new approach for simultaneously retrieving cloud albedo
16682. and cloud fraction from surface-based shortwave radiation measurements.
Nakajima, T., King, M.D., 1990. Determination of the optical thickness and effective Environ. Res. Lett. 8. http://dx.doi.org/10.1088/1748-9326/1088/1084/044023.
particle radius of clouds from reflected solar radiation measurements. Part I: Xie, Y., Liu, Y.G., Long, C.N., Min, Q.L., 2014. Retrievals of cloud fraction and cloud
theory. J. Atmos. Sci. 47 (15), 1878–1893. albedo from surface-based shortwave radiation measurements: a comparison
Niu, J., Yang, P., Huang, H., Davies, J., Li, J., Baum, B., Hu, Y., 2007. A fast infrared of 16 year measurements. J. Geophys. Res. Atmos. 119 (14), 8925–8940.
radiative transfer model for overlapping clouds. J. Quant. Specrtrosc. Radiat. Xie, Y., Yang, P., Gao, B.C., Kattawar, G.W., Mishchenko, M.I., 2006. Effect of ice
Transfer 103 (3), 447–459. crystal shape and effective size on snow bidirectional reflectance. J. Quant.
Oreopoulos, L., Barker, H.W., 1999. Accounting for subgrid-scale cloud variability in Specrtrosc. Radiat. Transfer 100 (1–3), 457–469.
a multi-layer 1-D solar radiative transfer algorithm. Q. J. R. Meteorol. Soc. 125, Xie, Y., Yang, P., Kattawar, G.W., Minnis, P., Hu, Y.X., 2009. Effect of the
301–330. inhomogeneity of ice crystals on retrieving ice cloud optical thickness and
Perez, R., Kivalov, S., Schlemmer, J., Hemker, K., Renne, D., Hoff, T., 2010. Validation effective particle size. J. Geophys. Res. 114, D11203. http://dx.doi.org/10.1029/
of short and medium term operational solar radiation forecasts in the US. Sol. 2008JD011216.
Energy 84, 2161–2172. Xie, Y., Yang, P., Liou, K.N., Minnis, P., Duda, D.P., 2012a. Parameterization of contrail
Pinker, R.T., Laszlo, I., 1992. Modeling surface solar irradiance for satellite radiative properties for climate studies. Geophys. Res. Lett. 39 (L00F02). http://
applications on a global scale. J. Appl. Meteorol. 31 (2), 194–211. dx.doi.org/10.1029/2012GL054043.
Qiu, J., 2001. An improved model of surface BRDF-atmospheric coupled radiation. Xie, Y., Yang, P., Kattawar, G.W., Minnis, P., Hu, Y.X., Wu, D., 2012b. Determination of
IEEE Trans. Geosci. Remote 39 (1), 181–187. ice cloud models using MODIS and MISR data. Int. Remote Sens. 33, 4219–4253.
Ramanathan, V., 1987. The role of earth radiation budget studies in climate and Yan, B., Stamnes, K., 2003. Fast yet accurate computation of the complete radiance
general circulation research. J. Geophys. Res. 92 (D4), 4075–4095. distribution in the coupled atmosphere-ocean system. J. Quant. Spectrosc.
Rienecker, M., Suarez, M., Gelaro, R., Todling, R., Bacmeister, J., Liu, E., Bosilovich, M., Radiat. Transfer 76 (2), 207–223.
Schubert, S., Takacs, L., Kim, G.K., Bloom, S., Chen, J., Collins, D., Conaty, A., Da Yang, K., Koike, T., Ye, B., 2006. Improving estimation of hourly, daily, and monthly
Silva, A., Gu, W., Joiner, J., Koster, R., Lucchesi, R., Molod, A., Owens, T., Pawson, solar radiation by importing global data sets. Agric. Forest. Meteorol. 137, 43–
S., Pegion, P., Redder, C., Reichle, R., Robertson, F., Ruddick, A., Sienkiewicz, M., 55.