Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261179954

Revising ASHRAE climatic data for design and standards - Part 2: Clear-sky solar
radiation model

Article in ASHRAE Transactions · January 2013

CITATIONS READS

8 1,948

2 authors, including:

Chris Gueymard
Solar Consulting Services
317 PUBLICATIONS 7,895 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Clear-sky irradiance modeling for solar applications View project

Solar radiation time series forecasting View project

All content following this page was uploaded by Chris Gueymard on 21 June 2014.

The user has requested enhancement of the downloaded file.


ASHRAE Trans., 119(2), 194-209, 2013

Revising ASHRAE Climatic Data


for Design and Standards –
Part II, Clear-Sky Solar Radiation Model
(1613-RP)

Christian A. Gueymard, PhD Didier Thevenard, PhD, P.Eng.


Member ASHRAE

ABSTRACT
The first part of this paper (Revising ASHRAE Climatic Data for Design and Standards – Part I, Overview and Data (1613-RP)) provided
a summary of the changes made to the tables of climatic design conditions in the 2013 Handbook – Fundamentals. This second part describes
changes that are made to the clear-sky solar radiation model. The model provides a simple way to calculate solar irradiance components from a
pair of location-specific parameters for any location in the world, and is used in particular to evaluate cooling loads in buildings.
The model was first introduced in the 2009 Handbook but frequently exhibited an apparent bias (direct normal irradiance too low, diffuse irra-
diance too high), which presumably resulted from a high bias in the aerosol data sets used for its derivation. This paper explains how the bias was
corrected for the 2013 Handbook. Various sources of gridded aerosol data, derived from satellite observations, were combined and calibrated with
sunphotometric data from 652 ground stations. A statistical analysis was performed to determine the most appropriate statistical estimator of
aerosol optical depth to use. It was found that because of the log-normality distribution of that variable over monthly periods, using its median
(rather than its mean) translates into irradiance values that are more representative of average conditions. A simple linear correlation was estab-
lished to correlate the median aerosol optical depth to its mean. Finally, the derivation of a condensed set of equations, which constitutes the clear-
sky model as it appears in the Handbook, was revised to cover a larger set of aerosol and surface albedo conditions.
The clear-sky model has been validated against clear-sky solar irradiance data from a number of research-class stations, including Darwin, Aus-
tralia; Golden, CO, USA; and Xianghe, China. Based on the analysis summarized here, the clear-sky model is found to be in reasonable
agreement with measured values for these stations, even under very hazy conditions.

INTRODUCTION

Over the years, various chapters of the Handbook - Fundamentals (HOF) have provided a method to estimate clear-
sky (cloud-free) irradiance values for use with cooling load estimation methods and other applications. Regional/seasonal
values of clearness number, required to use the clear-sky model, were only provided graphically for the continental U.S., and

Christian A. Gueymard is president of Solar Consulting Services, Colebrook, NH. Didier Thevenard is principal of Numerical Logics Inc., Wa-
terloo, ON.
could not be easily used, or could not even be used at all outside of the U.S. A history of this clear-sky solar radiation mod-
el, and a discussion about its shortcomings and accuracy, can be found in (Gueymard and Thevenard, 2009; Gueymard.
2012a). Based on the findings of 1453-RP (Thevenard and Gueymard, 2010), the 2009 edition of HOF introduced a com-
pletely new clear-sky radiation model that had the capability to evaluate solar irradiance components based on locally-
specified atmospheric conditions for any location in the world. The rationale for an updated clear-sky model, as well as the
details about the development of that new model and of the underlying datasets of atmospheric inputs, was described in
(Gueymard and Thevenard, 2009). Based on that study, the new ASHRAE clear-sky radiation model introduced in HOF
2009 consists of a highly simplified, or “condensed”, reformulation of the high-performance REST2 model (Gueymard,
2008). The latter model can be considered “state-of-the-art”, and has undergone extensive validation and intercomparison
(Gueymard, 2012a). The simplification leading to the condensed ASHRAE model was introduced to prevent excessive
computation burden, while maintaining as much of the generality and accuracy of the original REST2 model as possible.
A preliminary validation of the condensed model appeared in (Gueymard and Thevenard, 2009), based on measured
radiation data from only a few stations. It was found that the predicted direct normal irradiance (DNI) was often too low,
whereas the diffuse horizontal irradiance (DIF) was often too high. Using data from more experimental stations, further
validation was undertaken by the authors after that publication, and confirmed that these biases could indeed be considered
relatively general, albeit with varying magnitude. This prompted a special study (within 1613-RP) to remedy the situation, as
described in what follows. The two main directions of research were (i) the development of a more robust formulation of
the condensed REST2 model (see Section 2); and (ii) the development of a better input dataset for the aerosol optical depth
(AOD, see Section 3).
Suspicion that a high bias in the AOD data used as inputs could explain the prediction biases (too low DNI and too
high DIF) resulted from the well-known fact that AOD has a strong and opposite effect on these two irradiance compo-
nents. In particular, the impact of AOD on DNI has been quantified at various time scales (Gueymard, 2003, 2012b), and
found quite high. The radiative effects of other atmospheric constituents, such as water vapor or ozone, results in extinc-
tion in both DNI and DIF of similar relative magnitude, and thus would not explain the observed biases of opposite signs.
The propagation of errors from these inputs to DNI is also much less than that caused by AOD (Gueymard, 2003). Conse-
quently, only the atmospheric aerosol data have been scrutinized and modified in this study, compared to the input datasets
used in the 2009 version of the model.
The results of this work have been incorporated into an updated version of the clear-sky model in the 2013 HOF, as
described in the following sections.

2. CONDENSED MODEL MODIFICATION

2.1 Aerosol optical depth

The REST2 model requires two aerosol-related inputs: the AOD at 1 µm, better known as the Ångström turbidity co-
efficient, ß, and the Ångström exponent, α, that characterizes the spectral variation of AOD through Ångström’s law:

τ aλ = ß λ - α (1)

where τa is the spectral AOD at wavelength λ (µm). Since most sources of gridded aerosol data are now expressed in terms
λ

of AOD at 550 nm, τa550, ß can be simply derived from them using

ß = τa550 0.55α. (2)


In what follows, the term AOD is often employed in a generic way to refer to both τa550 (or ß) and α, since the com-
bination of these two quantities through Eq. (1) allows evaluation of AOD at any wavelength.

2.2 Model overview

Based on extensive research (Gueymard and Thevenard, 2009; Thevenard and Gueymard, 2010), a new form of solar
irradiance model has been introduced in the 2009 HOF. This “condensed” clear-sky model is described by only two main
equations:

Eb = Eo exp[–τb ∙mab] (3)


Ed = Eo exp[–τd ∙mad] (4)
where
Eb = beam normal irradiance (measured perpendicularly to the sun rays)
Ed = diffuse horizontal irradiance (measured on a horizontal surface)
Eo = extraterrestrial normal irradiance (based on a solar constant of 1367 W/m2 [433.3 BTU/hr ft2])
m = air mass (a pure function of the sun’s zenith angle; Kasten and Young, 1989)
τb and τd = beam and diffuse pseudo optical depths
ab and ad = beam and diffuse air mass exponents.

In equations (3) and (4), τb and τd are site-specific numerical coefficients that, from a physical standpoint, represent the
“effective” optical depth of the total atmosphere at any given time. For the engineering applications envisioned here, they
are defined on a mean monthly basis—more precisely for the 21st day of each calendar month. The air mass exponents are
obtained through empirical relationships given in the 2009 HOF:

𝑎! = 𝜅! + 𝜅!! ∙ 𝜏! + 𝜅!" ∙ 𝜏! + 𝜅!!" ∙ 𝜏! ∙ 𝜏! (5)


a! = κ! + κ!" ∙ τ! + κ!! ∙ τ! + κ!"! ∙ τ! ∙ τ! (6)

where the eight coefficients κi are numerical constants, obtained through a fit procedure to match results from the full
REST2 model. To that end, a reference dataset is first obtained by running REST2 for a wide selection of conditions; then
these results are fitted to equations (5) and (6). In other words, the condensed model consists of a simple parameterization
of the REST2 model’s results.

2.3 Development of revised coefficients

In addition to the total ozone depth (set constant to 0.35 cm), the inputs to REST2 (version 8.3) that are varied to ob-
tain a reference dataset or “look-up table”, consist of:

Zenith angle (°)


Atmospheric pressure (hPa)
Precipitable water w (cm)
Ångström’s turbidity coefficient ß (unitless)
Ångström’s wavelength exponent α (unitless)
Surface albedo (unitless).

The specific inputs used for the 2009 HOF are summarized in Table 1. These combinations of inputs lead to a very
large number of calculations (14,040) from which the condensed model was derived. As it was later realized, the values cho-
sen were not ideal in the sense that they could not cover the whole range of inputs for all possible weather stations com-
piled in the 2009 HOF. In particular, only a single value of wavelength exponent, α, and of surface albedo were used. A
more realistic set was thus developed for the 2013 HOF, as summarized in the last column of Table 1. Note, in particular,
that the sun geometry is now defined much more precisely (by the zenith angle rather than date/time/latitude) than in the
previous version.

Table 1 – Inputs used with REST2 to derive the reference datasets for the 2009 and 2013 HOF.
Input 2009 HOF range 2013 HOF range
Month 1 to 12 Not set
Time 0 to 12 Not set
Latitude (°) 0, 15, 30, 45, 60, 75 Not set
Zenith angle (°) Set from above 0 to 85° in 0.5° increments
Station pressure (hPa) 750, 1013.25 975
Precip. water (cm) 0.5, 1, 2, 6 0.25, 0.5, 1.5, 3.5
0.01, 0.02, 0.03, 0.05, 0.07, 0.1, 0.15, 0.01, 0.03, 0.06,
Turbidity coeff. ß
0.2, 0.3, 0.5 0.1, 0.2, 0.3
Wavelength exp. α 1.3 0.6, 1.0, 1.3, 1.6
Surface albedo 0.2 0.05, 0.1, 0.2, 0.3

The least-squares fits developed with this new set of inputs from the database of irradiance results just described
(69,120 combination of inputs in total) led to a re-derivation of equations (5) and (6). Although the form of the equations
remains unchanged, the values of the eight multilinear regression coefficients 𝜅! are updated, and these equations now take
the form:

𝑎! = 1.454 − 0.406 ∙ 𝜏! − 0.268 ∙ 𝜏! + 0.021 ∙ 𝜏! ∙ 𝜏! (5a)


𝑎! = 0.507 + 0.205 ∙ 𝜏! − 0.080 ∙ 𝜏! − 0.190 ∙ 𝜏! ∙ 𝜏! . (6a)

Users are cautioned that these expressions, developed specifically for the 2013 HOF, differ from those used in the
2009 HOF. Hence, since the derivation of the numerical values of τb and τd depends on the expressions used for 𝑎! and
𝑎! , the equations from the 2009 HOF should only be used with the values of τb and τd published in the 2009 HOF; and
vice versa, equations from the 2013 HOF should only be used with the values of τb and τd published in the 2013 HOF.

2.4 Adjustment of coefficients 𝝉𝒃 and 𝝉𝒅

By design, the condensed model exactly matches the predictions of the full REST2 model at a zenith angle of 0°.
However, at zenith angles above 45° the model may deviate significantly from REST2, which is unfortunate since this range
of angles is of practical interest in mid-latitude climates. To correct for this, the coefficients τb and τd are slightly adjusted to
minimize the difference between the condensed and full REST2 models over the whole range of zenith angles from local
solar noon to 85°. The adjustment is done algorithmically by exploring the values of τb and τd around the default values
provided by the condensed model, and selecting the ones that minimize the sum of absolute differences between the con-
densed model and the REST2 model for all zenith angles in the range of interest.
The use of the new equations (5a) and (6a), and of adjusted values of τb and τd, enable the condensed model to closely
follow the full REST2 model for most stations. Figure 1 illustrates this for Atlanta, GA (USA) over the four seasons. Simi-
lar agreement is found at most other sites, although it should be noted that the agreement is not as good at very high-
latitude sites, or in areas with high aerosol loading (i.e., high ß). This latter case is illustrated in Figure 2 for Karachi, Paki-
stan: the fit for diffuse irradiance, for example, deviates significantly from the full REST2 model, particularly in the early
morning and late afternoon in summer. This is caused by the various simplifications introduced by the condensed model
and the empirical nature of its coefficients. This loss of accuracy under extreme conditions can be seen as a minimal “price
to pay” for the convenience and otherwise considerable advantages of using a simple “engineering” model.

Atlanta,  GA,  USA


1000 317
Eb,  fit
900 Ed,  fit 285
1
Eb,  full  REST2
800 Ed,  full  REST2 254

Clear  sky  irradiance  (BTU/h-­‐ft2)


700 222
Clear  sky  irradiance  (W//m2)

600 190

500 159
3
400 127
4
300 95

200 63
2 2
100 32
4
0 1,3 0
3 6 9 12 15 18 21
Local  solar  time

Figure 1 Comparison of condensed and full REST2 model for Atlanta, GA, USA, for (1) March 21, (2) June 21,
(3) September 21 and (4) December 21. Eb is beam normal irradiance, Ed is diffuse horizontal irradiance. The
term ‘fit’ refers to the condensed model described by equations (3)-(6).

Karachi,  Pakistan
900 285
Eb,  fit

800 Ed,  fit 254


Eb,  full  REST2
4
700 Ed,  full  REST2 222
1
Clear  sky  irradiance  (BTU/hr-­‐ft2)
Clear  sky  irradiance  (W//m2)

600 190

3
500 159
2

400 127
2
3
300 95
1
200 63
4
100 32

0 0
3 6 9 12 15 18 21
Local  solar  time

Figure 2 Comparison of condensed and full REST2 model for Karachi, Pakistan, for (1) March 21, (2) June 21,
(3) September 21 and (4) December 21. Eb is beam normal irradiance, Ed is diffuse horizontal irradiance. The term
‘fit’ refers to the condensed model described by equations (3)-(6).
3. AEROSOL DATA

It is well established that, under cloudless conditions, AOD is the atmospheric variable that has the strongest effect on
both DNI and DIF (see, e.g., Gueymard, 2003; 2008). Therefore, the accuracy of DNI and DIF directly depends on that of
AOD. Improving the accuracy of the worldwide AOD datasets that are needed here constitutes a challenge because of the
limited number of high-quality ground observations, which are too scarce for the task at hand. Only remote-sensed obser-
vations from satellites or model calculations can provide AOD data with the necessary worldwide geographic coverage.
However, these sources of data are known to have biases, sometimes large, with strong temporal and/or spatial variations.
A methodology to develop a worldwide dataset of monthly-mean AOD (in terms of both τa550 and α), while reducing the
biases in the original data, was described in (Gueymard and Thevenard, 2009). A similar methodology has been used here,
but with a number of improvements, which are described in the following subsections.

3.1 Sources and accuracy of aerosol data

A detailed review of the usable sources of aerosol turbidity data was provided in (Gueymard and Thevenard, 2009).
The same sources of data are used here, but with multiple improvements: (i) the ground-based sunphotometric observa-
tions (used as “ground truth” and reference for calibration of gridded data) are now available at more sites and for longer
periods; (ii) the gridded satellite-derived data under cloud-free conditions are now based on upgraded retrieval algorithms,
and now include retrievals from the MISR instrument in addition to the two MODIS instruments that were used before;
and (iii) two new modeled datasets have been added to the pool of gridded data, namely an AeroCom climatology (Kinne,
2009) and a climatology derived from the MACC reanalysis model (Bellouin et al., 2012; Benedetti et al., 2009; Morcrette et
al., 2009).
Ground observations of AOD constitute the most accurate source of data, with an absolute accuracy estimated at
±0.015 in the units of τa550 (Holben et al., 2001). This good and consistent accuracy explains why this type of measurement
is always considered as “ground truth” to validate other sources of aerosol data. The accuracy of AOD retrievals from the
spaceborne MODIS instruments has been estimated by (Levy et al., 2010) at ±(0.05+0.15∙τa550) for the high-resolution data,
and at ±(0.06+0.06∙τa550+0.29∙τa5502) for data aggregated at 1x1° resolution (Ruiz-Arias et al., 2012). These are global aver-
ages with significant geographical variations (Ruiz-Arias et al., 2012). The accuracy of modeled AOD data is not known
precisely, but can be assumed to be similar to the MODIS result just mentioned in the case of the MACC model, since the
latter is constrained by MODIS observations.
Evaluating the uncertainty in AOD data is important in the context of modeling solar irradiance because of the strong
dependence of the latter on the former, particularly in the case of DNI under cloudless skies. This dependence, and the
resulting propagation of errors, has been studied in detail (Gueymard, 2003, 2005, 2012a). As a tentative conclusion of these
studies, the following rule-of-thumb can be formulated: if an accuracy of ±10% in DNI is desired, that in τa550 should be no
larger than ±0.10. This translates into a relatively stringent requirement about the quality of the AOD data, which also con-
stitutes an important justification for the present detailed study.

3.2 Spatial and temporal resolution

The previous monthly datasets of both ß and α were developed on a relatively coarse 1°x1° grid. Taking into account
the better native resolution of the MISR data, all AOD data (at 550 nm in the case of MODIS and at four wavelengths in
the case of MISR) were regridded to the MISR resolution, i.e., 0.5°x0.5°, which corresponds to grid boxes of ≈55x55 km at
the equator and ≈40x55 km at a latitude of ±45°. This finer resolution allows a better comparison between gridded data,
for which the AOD value is averaged over a grid box, and ground observations, which rather correspond to spot locations.
Increasing the spatial resolution of gridded data is important because the aerosol field is far from homogeneous, particularly
over complex topography. Since AOD is also very sensitive to elevation, the same scale height correction as developed be-
fore (Gueymard and Thevenard, 2009) has been applied to AOD when comparing gridded and ground (spot) data, using a
world digital elevation model (DEM) at 0.5° resolution.
The aerosol field is also highly dependent on time, with significant interannual variability in particular. Therefore, the
temporal resolution of the AOD data has also been improved, now using mean monthly values for each year of the 12-year
period from 2000 to 2011, during which retrieved AOD data from MISR and MODIS are available. Sunphotometer sta-
tions with as little as one single month of valid data (i.e., at least 4 days of data in any single month) during that period were
used for ground truthing, inasmuch as their scale height correction was not more than ±35%. This eliminated some poten-
tial control stations located in high mountain areas. A total of 652 stations and 19,661 station-month data points survived
the filtering process (Table 2). The number of monthly data points per station varies considerably from one station to the
next, between 1 and 130, for a world average of ≈30. The 12 annual datasets that resulted from this analysis were ultimately
combined to derive long-term mean-monthly AOD averages over the period 2000–2011. This means that the irradiance
data that were eventually produced through this study are representative of the most recent past. This is of importance in
the present context of climate change, since significant and regionally varying decadal trends in irradiance have been noted
(e.g., Wang et al., 2012; Wild, 2009; Wild et al., 2005).

Table 2: Number of sunphotometric stations and station-months used as ground truth.


Avg. months per
Continent Stations Station-months
station
Africa (northern) 83 2700 32.5
Africa (southern) 26 683 26.3
Asia 126 3150 25.0
Australasia (inland) 4 196 49.0
Australasia (coastal) 15 454 30.3
Europe 132 4902 37.1
North America (southwest) 24 799 33.3
North America (excl. southwest) 182 5095 28.0
South America 60 1682 28.0
WORLD 652 19661 30.2

The nominal wavelength used here for AOD is 550 nm for MODIS and 555 nm for MISR. Considering that the
MODIS data were used much more often than the MISR data, and that the difference between τa550 and τa555 is negligible
compared to the uncertainty of the method, all results have been obtained for the single 550-nm wavelength. The wave-
length exponent, α, was calculated with the method described by (Gueymard, 2008), i.e., by linearly fitting the log-log trans-
form of Eq. (1), using the four available channels (443, 555, 670 and 865 nm) of the MISR sensor. In the case of MODIS,
the α data provided by NASA were used instead. In what follows, the methodologies used to evaluate τa550 and α are exact-
ly the same, with the only exception that no scale height correction was applied on α, like in the authors’ previous study
(Gueymard and Thevenard, 2009).

3.3 Regional and seasonal analyses

Retrieval algorithms used to convert radiance observations by satellite-based sensors into AOD data must make as-
sumptions about the most likely type of aerosols being sensed. Similarly, aerosol transport models analyze the production,
transport and chemical transformation of different aerosol species, such as sulfates, organic matter, sea salt, black carbon or
dust, treated separately. In-depth validation studies have shown that the performance of both satellite retrievals and
transport aerosol models was strongly affected by the prevalent aerosol species, both on a regional and seasonal basis (e.g.,
Kinne et al., 2006).
Based on a preliminary assessment of the spatial and temporal performance of satellite retrievals, it became clear that
the analysis could normally be done on a continental scale. The uneven ground-truth data distribution over space justified a
continental scale, but with some sub-scaling wherever necessary. In particular, the southwest North America had to be sep-
arated from the rest of that continent because of the excessive bias in satellite retrievals observed there, and later confirmed
by a different study (Ruiz-Arias et al., 2012). For the same reason, Australasia was analyzed separately for the central (de-
sert) part of Australia and all coastal areas, whereas Africa was separated into two parts, north and south of the equator. In
total, nine continental or sub-continental areas were finally considered. (No specific analysis was conducted for Antarctica
because of the extreme paucity of ground-truth data there and the marginal importance that solar radiation data would have
for any ASHRAE application over that continent, which would be highly unlikely anyway; AOD data have been developed
for Antarctica anyway, but by spatial extrapolation, and thus with possibly larger uncertainty.)
Although a month-by-month analysis would have been desirable (since the goal of this study is to produce irradiance
data on a monthly basis), this would have caused statistical problems due to the relative paucity of ground data during some
months (particularly winter at high latitudes) and their uneven distribution over time. Thus, the monthly data were normally
pooled into two seasons (summer and winter), based on the fact that the aerosol load is generally more pronounced in
summer than winter. An exception had to be considered for southern Africa, due to the existence of three contrasting aero-
sol seasons, as far as their species and magnitude are concerned: summer (Dec–Apr) with normally moderate AODs, dust
season (May–Jun) with strong occasional dust storms in some areas, and fire season (Jul-Nov) with very large AODs caused
by smoke over large areas.

3.4 Bias correction, data gap filling and mapping

For each continental area and each season, the data points from each source of data described in Section 3.1 were
compared to the ground observations, or “ground-truth dataset”. Examples of such comparisons corresponding to summer
conditions are shown in Figure 3 for Europe and Asia. Although very low AOD values are observed over both continents,
Europe has very few cases with τa550>0.40, whereas such cases are common in Asia, due to intense dust and/or anthropo-
genic pollution loads over many areas. A large scatter, which tends to increase with τa550, is also apparent in both cases. This
results from three different effects: (i) the absolute uncertainty in spaceborne data that increases with AOD, as mentioned
in Section 3.1; (ii) the spatial distortion due to spatial variability of aerosols and imperfect comparison between an average
grid-cell value and a spot measurement; and (iii) a temporal distortion due to the temporal variability of aerosols and the
highly likely non-coincidence between the ground-truth measurement periods and the overpass times of the polar-orbiting
satellites.
The noise (random differences) apparent in Figure 3 is unavoidable when validating the existing sources of gridded
data on a monthly basis and at their nominal spatial resolution. What is most important, however, is to remove as much
bias as possible in the gridded data. This is achieved here by correcting the raw gridded data with a correction function cor-
responding to the simple linear regression between it and ground truth, or by using a multilinear regression involving two
complementary gridded sources of data. With this process, the best source(s) of raw gridded data are used first, in decreas-
ing order of performance. In case of missing data, the next best source is used, and the process is repeated until it is certain
that all grid cells are appropriately covered. The process is also extended geographically from each continent to cover its
surrounding oceanic areas, since there are many scattered islands in need of radiation data.
The correction and supplementation process described above is repeated for each month of the 2000–2011 period,
for both τa550 and α, and for each continental area. The corresponding long-term average values are then calculated from
these 12-year datasets. Monthly maps of τa550 and α for the world for July are shown in Figures 4 and 5, respectively. Areas
with very high aerosol loading appear as “hot spots” in Figure 4. These are particularly intense over parts of Asia, as could
be expected. However, hot spots over other areas (such as north and central Africa or eastern South America) also appear
during other months, since the underlying sources of aerosols have a strong seasonality.
2 2
Monthly AOD Europe Monthly AOD Asia
MODIS Aqua MODIS Terra
1.8 Summer 1.8 Summer

1.6 1.6

1.4 1.4
AOD550 Ground Truth

AOD550 Ground Truth


1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 Linear correction: 0.2 Linear correction:


y = 0.056432 + 0.66966x R= 0.74432 y = 0.073911 + 0.76905x R= 0.82682

0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
AOD550 Gridded Data AOD550 Gridded Data

Figure 3 Comparison between ground-truth sunphotometric data and satellite observations of monthly-mean AOD at 550 nm for
Europe in summer (left plot) and Asia in summer (right plot).

3.5 Statistical analysis

As pointed out in the Introduction, biases of an apparently systematic nature were found in the monthly values of
DNI and DIF predicted by the 2009 model. A logical explanation was that this could be primarily caused by a high bias in
the AOD input to the model. Since further tests showed that the model’s predictions were correct on an instantaneous ba-
sis, and since the AOD dataset was produced in a way to avoid bias, the remaining possible explanation was that some sta-
tistical attribute of the AOD distribution could be at play. The hypothesis that resulted is that, to obtain the monthly mean
DNI and DIF accurately, the monthly mean AOD is not the correct statistic to use. The formulation of this hypothesis
stemmed from the various studies that emphasized that AOD was not normally distributed over time, but was rather log-
normally distributed (O’Neill et al., 2000; Liu et al., 2008; Pan et al., 2010; Anderson et al., 2012). Indeed, the ground-truth
collected for this study confirmed the log-normality of the τa550 and ß distributions. Figure 6 shows examples of monthly fre-
quency distributions for one very clear site (Sevilleta, New Mexico) and one very hazy site (Kanpur, India). On an annual
basis the frequency distribution still appears asymmetric and log-normal, but much smoother because the number of data
points is then about an order of magnitude larger. All the monthly or annual frequency distributions are found right-tailed,
i.e., they all have a positive skewness. A direct and important consequence of this positively skewed log-normality is that the
mean is larger than the median and, even more so, than the mode. The latter would be the statistic of choice if the most fre-
quent aerosol conditions and thus the clearest possible conditions were to be simulated. This avenue was actually selected in
a previous ASHRAE project to obtain the largest possible DNI with a large probability of occurrence (Gueymard, 2007).
For the present study, the goal was to obtain monthly mean irradiance data under clear-sky conditions, thus leaving the sta-
tistical issue (regarding the most appropriate conditions for cooling load calculations) open.
 

Figure 4 Long-term mean AOD at 550 nm over the world for July.

Figure 5 Long-term mean Ångström exponent (α) over the world for July.
Table 3: Information on 9 sunphotometer stations used in the statistical study.
Mean Mean Mean
Valid
Station Country Lat. Long. Elev. (m) Annual Annual Annual
months
ß α w (cm)
Banizoumbou Niger 13.54 2.67 250 0.438 0.353 2.305 12
Belsk Poland 51.84 20.79 190 0.086 1.422 1.493 10
Bonanza Creek, AK USA 64.74 -148.32 150 0.068 1.256 1.197 7
El Arenosillo Spain 37.11 6.73 20 0.082 0.979 1.756 12
FORTH-Crete Greece 35.33 25.28 20 0.102 1.102 1.883 12
Kanpur India 26.51 80.23 123 0.351 0.938 2.434 12
Moscow Russia 55.70 37.51 192 0.086 1.460 1.381 11
Sede Boker Israel 34.78 30.86 480 0.116 0.919 1.352 12
Solar Village Saudi Arabia 24.91 46.40 764 0.277 0.522 1.237 12
ALL — — — — 0.194 0.943 1.727 100

For each of the 9 stations selected for this statistical study (Table 3), the daily AOD data points (from ground truth)
have been sorted by month. The mean, median and mode of the measured daily ß values have been calculated for each
month. For each individual day for which measured values of AOD (reduced here to the two Ångström coefficients ß and
α) and precipitable water, w, were available, the sub-hourly and daily irradiances were calculated with REST2, using also
climatological values of ozone and nitrogen dioxide. Comparisons between the predicted hourly or daily values and the cor-
responding measured irradiances at Sede Boker, Israel (where nearly all days are cloudless in summer) showed excellent
agreement, thus validating the method. The twelve long-term monthly DNI averages were thus calculated for each station,
to represent the reference mean daily DNI. Such reference results were obtained for a total of 100 station-months (Table 3).

30
τ
a550
Frequency distribution
Sevilleta, New Mexico
25 Nov, 1994–2012 of daily values

Mean 0.039
Median 0.033
Mode 0.032
20
Frequency (%)

15

10 Kanpur, India
May, 2001–2012
Mean 0.708
Median 0.659
5
Mode 0.563

0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
τ
a550

Figure 6 Frequency distributions and summary statistics of the daily AOD at 550 nm during a calendar month for
Sevilleta, New Mexico and Kanpur, India.
In a second step, the process was repeated three more times, but now replacing the actual daily value of ß by its long-
term monthly mean, median or mode, while keeping all other daily inputs the same as in the initial run. These three alternate
calculations are “static” since they do not take the daily variability of ß into account, in contrast with the “dynamic” calcula-
tion of the reference DNI results just described.
The monthly-average DNI values derived from the monthly mean ß data are found systematically lower than those
obtained with the reference data, based on daily values of ß. This is consistent with the observations made about the results
of ASHRAE 1453-RP that were mentioned earlier. This shows that, indeed, the monthly mean ß is not the appropriate sta-
tistics to use for the problem at hand (Figure 7). Conversely, the monthly DNI values obtained when using the monthly
modal value of ß are systematically too high. A much better agreement is found when using the monthly median value of ß
(Figure 7). To find the “effective” monthly values of ß that would yield a nearly perfect agreement between the static and
dynamic methods, linear interpolation has been used. The relationship between these effective ß values and their monthly
mean, median and modal counterparts is shown in Figure 8. Despite the noise inherent to the method, the effective ß values
appear to follow a linear relationship with the mean ß, with a slope of ≈0.8 (Figure 8).

15
9 Sites

DNI from mean β


DNI from median β
Monthly DNI from monthly β

DNI from mode β


10

y = 0.94766x
y = 1.0211x
y = 1.075x

0
0 5 10 15
Monthly DNI from daily β (Reference)

Figure 7 Monthly DNI obtained with various monthly statistics of ß vs. reference values based on individual daily values.

In a final processing step, each grid cell was subjected to Eq. (2) so as to derive the mean ß from the mean τa550 and
the mean α data, both obtained per Section 3.4. The effective ß was then obtained as 0.8 times the mean ß. Finally, these
values were interpolated over time to produce data representative of the 21st of each month. The latter interpolation was
also used for α.
0.8
9 Sites
y = m0*(m1 + m2 * M0)/(1+m4*...
0.7 Effective monthly β Value Error
Monthly median β m1 0.91948 0.017618
0.6 Monthly mode β m2 -0.26095 0.037802
m4 -0.067938 0.010031
Chisq 0.035286 NA

Calculated β
0.5 R 0.99996 NA

0.4
y = m0*(m1 + m2 * M0)/(1+m4*...
Value Error
0.3 m1 0.71151 0.045285
m2 0.70361 0.36981
0.2 m4 0.82082 0.46673
Chisq 0.036695 NA
+ ... M8*x8 + M9*xNA
Y = M0R+ M1*x0.99064 9
0.1
M0 0.0065512
M1 0.49056
0 M2 0.057462
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 M3 0.04385
Monthly mean β R 0.99999

Figure 8 Effective monthly ß derived from the reference monthly DNI, median ß and modal ß, as a function of the mean monthly ß.

4. FINAL PRODUCTION

The final production of solar radiation data for the 2013 HOF was done is several stages. First, gridded dataset of
REST2 inputs were prepared. Data of single-scattering albedo, water vapor, surface albedo, ozone and elevation were iden-
tical to those used for the 2009 HOF; details about these datasets can be found in Gueymard and Thevenard (2009). The
only difference was the additional use of 0.5° gridded elevation data along with the 1° data used previously. The aerosol
data (ß and α) were derived as explained in Section 3. Then, individual fits were developed for each station using the full
REST2 model and updated equations (5a) and (6a). That is, for each month, the full REST2 model was run with inputs
interpolated from the sets above, over the whole range of zenith angles from local solar noon to 85°; 𝜏! and 𝜏! were cho-
sen so as to minimize the overall difference between the full REST2 model and the condensed model. The tabulated values
of the site-specific 𝜏! and 𝜏! coefficients were then kept for inclusion along with other climatic design conditions in the
Handbook. All sites are provided with monthly values of 𝜏! and 𝜏! ; the only exceptions are sites at extreme latitudes where
some values may appear as N/A because of polar night.

5. VALIDATION OF THE NEW RADIATION MODEL

For validation purposes, solar irradiance data have been obtained from a number of research-class stations that follow
the most rigorous quality standards. Results for three of them are used here (Table 4). Normally, validation studies involve
the comparison of modeled estimates and measured data points for the same exact moment, out of which summary statis-
tics such as mean bias difference, root mean square error, or uncertainty at 95% are calculated (e.g., Gueymard, 2003, 2008,
2012a). Here, however, the modeled values refer exclusively to the 21st of each month, whereas the measured values are for
any day of the month. This introduces a source of unavoidable noise in the comparison. Additionally, some systematic er-
rors may result from the fact that the atmospheric data that are used as inputs to the model (particularly τb and τd) are each
evaluated as a spatial average over a gridbox, whose size is relatively large (0.5x0.5° or 1x1°, depending on input variable)
compared to the point-like footprint of a ground-based radiometric station. This introduces an unavoidable variance when
the atmospheric conditions are not constant over a gridbox, which is typically the case over complex topographic areas
(mountainous or coastal areas, for instance). To avoid confusion or misinterpretations of the present results compared to
others of the literature, only graphical results are presented in what follows.
The Darwin station is part of the Australian Bureau of Meteorology’s network. The Golden, Colorado station is the
main radiometric station maintained by the National Renewable Energy Laboratory (NREL). The Xianghe, China station is
part of the Baseline Solar Radiation Network (BSRN). At each station, the two radiation components (DNI and DIF) are
provided at high frequency (1-min time step). DNI is measured with a first-class pyrheliometer, whereas DIF is measured
with a first-class pyranometer with low thermal imbalance, and equipped with a tracking shade to permanently block the
sun within a subtended cone roughly equal to that of the pyrheliometer’s field of view. (The global horizontal irradiance is
also routinely measured, with an unshaded pyranometer, but is used here only for quality control purposes.) After filtering
out cloudy occurrences, the mean and standard deviation, σ, of each irradiance component over the whole measurement
period have been calculated at 30-min intervals for each calendar month. A few examples of these results are presented in
what follows. These examples are for the summer season at three sites (January at Darwin, July at Golden, and June at
Xianghe) and the winter season (July) at Darwin. For these months, the effective ß (as obtained from Section 3.5) is relative-
ly low at Darwin (0.07 in January and 0.03 in July) and Golden (0.04), but in contrast is quite high (0.32) at Xianghe. Note
that Darwin is the only station of the three to be part of the database of 6,443 stations for which monthly weather data have
been derived during the course of 1613-RP for inclusion in HOF 2013.

Table 4: Information on 3 radiometric stations used in the validation study.


Station Darwin Golden Xianghe
Lat. (°) -12.424 39.742 39.754
Long. (°) 130.892 -105.18 116.962
Elev. (m)/(ft) 29 / 95 1829 / 6000 36 / 118
Time zone (hr) 9.5 -7 8
Measurement period 2004–2011 2004–2012 2005–2010
𝜏! 𝜏! 𝜏! 𝜏! 𝜏! 𝜏!
Jan 0.460 2.497 0.261 2.559 0.394 2.193
Feb 0.442 2.516 0.256 2.538 0.447 2.043
Mar 0.397 2.579 0.273 2.476 0.534 1.811
Apr 0.363 2.585 0.270 2.533 0.651 1.566
May 0.337 2.572 0.275 2.542 0.724 1.451
Jun 0.321 2.552 0.282 2.551 0.761 1.416
Jul 0.321 2.533 0.310 2.495 0.669 1.613
Aug 0.363 2.408 0.320 2.487 0.595 1.780
Sep 0.416 2.345 0.298 2.585 0.574 1.809
Oct 0.463 2.320 0.275 2.665 0.510 1.941
Nov 0.470 2.394 0.265 2.581 0.422 2.193
Dec 0.488 2.395 0.261 2.575 0.388 2.264

In addition to the predictions obtained with the HOF 2013 model presented here, those obtained with the models
from previous HOF editions (1967–2001, 2005 and 2009) are also included for comparative purposes. (“How significant
are the changes induced by the evolution of the ASHRAE clear-sky model?” and “Is there a compelling reason to adopt the
newer model over my area of interest?” are typical questions that have been frequently asked by practicing engineers during
the last few years.) The 2005 model is basically the same empirical model as that first proposed in 1967 (whose history and
algorithmic details are provided in Gueymard and Thevenard, 2009), but with slightly modified monthly coefficients. The
four models analyzed here can thus be classified into two groups: the older, purely empirical models of the 1967 to 2005
HOF editions, and the newer, semi-physical models of the 2009 and 2013 editions.
The southern hemisphere location of Darwin makes the predictions of the HOF 1967–2005 models appear out of
phase, with DNI too high in summer and too low in winter (Figure 9), and vice versa for DIF. This is because these models
have been developed empirically from measured data in the U.S. exclusively, and thus for the seasonal pattern of the north-
ern hemisphere only, which is opposite to that of the southern hemisphere. (In any hemisphere, the normal seasonal pat-
tern is characterized by low ß and w in winter, and higher ß and w in summer, with relatively smooth sinusoidal-type varia-
tions along the year; more complex patterns do occur in those regions where intense influx of aerosols exist and have a dif-
ferent, highly-peaked seasonal pattern.) The July results at Darwin display a smooth diurnal variation of the measured DNI,
with relatively small standard deviations (and hence, day-to-day and interannual variability). All mid-day model predictions
are in remarkably close agreement, and within the ±1σ limits. The 2013 model may tend to overestimate here, whereas the
2009 model may be in closer agreement with the measured data. The implied caution in the statement just above is neces-
sary because the comparisons between predicted and measured data presented here are not perfect since the latter cover all
days in a calendar month, whereas the former are specified for the 21st of that month. This discrepancy creates a bias,
whose magnitude is normally small, but actually depends on the specific station and month. For the summer conditions
(January) at Darwin, a much larger variability in the measured data is found. This may be explained by the higher ß, and
much more frequent cloudy periods than in winter. Larger ß in the mid afternoon than in the morning, and/or occurrences
of undetected sub-visual or very thin cirrus clouds are also possible, leading to DNI’s magnitude being lower, and its stand-
ard deviation larger, than in the morning. The 2013 model appears to have the closest agreement here, whereas the older
1967–2005 models are significantly overestimating.

1200 380 1200 380


Clear-sky direct normal irradiance Clear-sky direct normal irradiance
Darwin, Jul. 2004–2011 Darwin, Jan. 2004–2011
340 340
1000 1000
300 300

Irradiance (BTU/hr ft )
Irradiance (BTU/hr ft )

260

2
2

260 800
800
Irradiance (W/m )
Irradiance (W/m )

2
2

220 220
600 600 180
180

140 140
400 400
100 100
Measured (mean±1σ) Measured (mean±1σ)
HOF 2013 HOF 2013
200 60 200 HOF 2009 60
HOF 2009
HOF 2005 HOF 2005
HOF 1967–2001 20 HOF 1967–2001 20
0 0
4 6 8 10 12 14 16 18 20 4 6 8 10 12 14 16 18 20
Local Standard Time Local Standard Time

Figure 9 Comparison between the measured and predicted DNI at Darwin, Australia for January and July. Error bars represent the
standard deviation of irradiance over the whole measurement period (2004–2011) and for each 30-min diurnal period.

The high-elevation Golden station experiences a clean and dry atmosphere, and therefore ß and w are normally low
there in summer, and even lower in winter. Results for DNI and DIF appear in Figure 10 for the month of July. Interest-
ingly, the 2013 model appears better in the morning, whereas the 2009 model appears better in the afternoon. The distinct
asymmetry in the measured data is a specificity of that station. It is located on a mesa above the city of Golden, and within
the agglomeration of Denver, which is known for frequent occurrences of photochemical smog. During morning hours, the
incident solar radiation is intense and creates an upward (and locally, upslope) air stream, which brings with it moisture and
aerosols from the city below. This also creates turbulence and an expansion of the aerosol mixing layer, which reaches a
new equilibrium shortly before local noon, on average. Since ß increases, DNI decreases and DIF simultaneously increases,
as could be expected. The condensed model is not designed to handle such diurnal variations, but the REST2 model is able
to do so without difficulty, provided instantaneous values of its most important input data are available.

1200 380 500


Clear-sky direct normal irradiance Clear-sky diffuse irradiance
Golden, Jul. 2004–2012 Golden, Jul. 2004–2012
340 140
1000 Measured (mean±1σ)
300 400 HOF 2013
HOF 2009 120

Irradiance (BTU/hr ft )

Irradiance (BTU/hr ft )
HOF 2005

2
800 260 HOF 1967–2001
Irradiance (W/m )

Irradiance (W/m )
2

2
100
300
220
600 80
180
200 60
140
400
100 40
Measured (mean±1σ)
HOF 2013 100
200 HOF 2009 60 21
HOF 2005
HOF 1967–2001 20
0 0 1
4 6 8 10 12 14 16 18 20 4 6 8 10 12 14 16 18 20
Local Standard Time Local Standard Time

Figure 10 Comparison between the measured and predicted DNI (left plot) and DIF (right plot) at Golden, Colorado for July. Error
bars represent the standard deviation of irradiance over the whole measurement period (2004–2012) and for each 30-min
diurnal period.

The case of Xianghe is markedly different from that of Golden because of the former’s very high aerosol load most of
the time. Such conditions are typical over many parts of the world, particularly in Asia, where considerable economic devel-
opment and building activity exist, including solar applications. The inadequacy of the older HOF models of 1967–2005 is
particularly obvious here, both for DNI and DIF. Due to the predominant strong-haze conditions, DIF is as high as about
half of DNI around noon, on average. It can be actually larger than DNI during days with peak haze conditions. As Figure
11 demonstrates, the diurnal variations of the mean DNI and DIF is not as smooth as in those in Figure 9 or 10, and their
standard deviations are much larger. All this is primarily caused by the strong diurnal, day-to-day and interannual variability
in ß. As was recently found (Gueymard, 2012b), such variability is roughly proportional to the long-term average ß, which is
≈10 times larger at Xianghe than at Golden on an annual basis, for instance. Other possible causes of unevenness are: cloud
interference (it is difficult to detect thin clouds when the sky is very hazy), and statistical interference (the number of total
clear-sky cases is highly variable during the average day, therefore some 30-min periods may have a larger fraction of clearer
conditions than others). The overestimation of the 2013 model during parts of the morning and afternoon can be explained
by the curve fitting issues mentioned in Section 2.4. Despite all these sources of discrepancy under high-turbidity condi-
tions, the 2013 model appears reliable, and even significantly better than the 2009 model.
1200 380 500 Clear-sky diffuse irradiance
Clear-sky direct normal irradiance Xianghe, Jun. 2005–2010
Xianghe, Jun. 2005–2010
340 140
1000
300 400
120

Irradiance (BTU/hr ft )

Irradiance (BTU/hr ft )
2

2
800 260
Irradiance (W/m )

Irradiance (W/m )
2

2
100
220 300

600 80
180
200 60
140
400
100 40
Measured (mean±1σ) 100 Measured
200 HOF 2013 60 HOF 2013 21
HOF 2009 HOF 2009
HOF 2005 20 HOF 2005
HOF 1967–2001 HOF 1967–2001
0 0 1
4 6 8 10 12 14 16 18 20 4 6 8 10 12 14 16 18 20
Local Standard Time Local Standard Time

Figure 11 Comparison between the measured and predicted DNI (left plot) and DIF (right plot) at Xianghe, China for June. Error
bars represent the standard deviation of irradiance over the whole measurement period (2005–2010) and for each 30-min
diurnal period.

6. CONCLUSION

The clear-sky solar radiation model introduced in the 2009 Handbook – Fundamentals was revised for the 2013 edi-
tion. Revisions included changes to the condensed model formulation and the use of better aerosol data.
The general shape of the clear-sky model equations appearing in the Handbook is unchanged. The model is still based
on two simple equations to derive beam and diffuse clear-sky irradiance, given only two monthly parameters (beam and
diffuse pseudo optical depths τb and τd) specific to each location. However, the correlations providing the beam and diffuse
air mass exponents ab and ad have been re-derived using a more extensive range of atmospheric conditions. In addition, a
new procedure has been developed to adjust τb and τd for each station, to minimize the difference between the predictions
of the model and those of the full REST2 model from which it was originally derived.
The atmospheric aerosol data used to derive the model have been significantly scrutinized and modified. The impetus
was that some clear-sky irradiance values in the 2009 model appeared somewhat biased towards low beam and high diffuse
irradiances relative to both ground-based observations and clear-sky model results from previous editions of the HOF; this
could be most readily explained by aerosol optical depth values being somewhat too high. An effort was made to build a
more accurate and representative dataset for aerosols. Several sources of satellite-derived, gridded aerosol turbidity data
were used. They were compared to ground-based observations obtained from 652 sunphotometric stations covering the
whole globe and representing a total of 19,661 station-months of observations. Most of the potential bias in satellite data
was corrected on a regional and seasonal basis, while all missing data periods were appropriately filled. This enabled the
derivation of a consistent, accurate set of long-term monthly mean gridded aerosol data for the whole world.
In addition, a statistical analysis was conducted to explain the bias in the clear-sky beam and diffuse radiation values
calculated for the 2009 Handbook. It was found that the aerosol optical depth does not follow a normal distribution, but
rather a log-normal one. For that reason, the use of the median aerosol optical depth appears to provide better results if one
wants to calculate mean clear-sky irradiance conditions at a given location. Based on data from 9 sunphotometer stations, a
correlation between median and mean aerosol optical depth was established. It enables the calculation of the median from
the mean gridded aerosol data derived during the previous step.
The new radiation model, incorporating changes to aerosol data and the change in formulation, was tested against
clear-sky solar irradiance data from various radiometric stations. These include Darwin, Australia; Golden, CO, USA; and
Xianghe, China, chosen here to represent a large range of atmospheric conditions, from generally very clear to very hazy.
The model was found to be in generally good agreement with ground-based observations, although the agreement varies
depending on time of day and climatic conditions specific to each station (including their interannual variability).
The revised model, and its accompanying new sets of pseudo optical depths τb and τd derived for 6,443 stations from
the updated gridded aerosol data, have been incorporated in the 2013 Handbook – Fundamentals. The impact of the updat-
ed, more accurate clear-sky model on building energy calculations is beyond the scope of this contribution, and thus has not
been investigated in detail. It can be anticipated that such impacts depend on the configuration of any given building and its
climate. Results of preliminary calculations (conducted by Steven F. Bruning and kindly provided to us through Pers.
Comm., 2012) show that, most generally, the changes have only a small impact on the cooling loads of model office build-
ings under various climate conditions.

REFERENCES

Anderson J.C., Wang J., Zeng J., Petrenko M., Leptoukh G.G. and Ichoku C., Accuracy assessment of Aqua-MODIS aero-
sol optical depth over coastal regions: importance of quality flag and sea surface wind speed. Atmos. Meas. Tech. Dis-
cuss., 5, 5205–5243, 2012.
Bellouin N., Quaas J., Morcrette J.J. and Boucher O., Estimates of aerosol radiative forcing from the MACC re-analysis.
Atmos. Chem. Phys. Discuss., 12, 20073–20111, 2012.
Benedetti, A., Morcrette, J.J., Boucher, O., Dethof, A., Engelen, R.J., Fisher, M., Flentje, H., Huneeus, N., Jones, L., Kaiser,
J. W., Kinne, S., Mangold, A., Razinger, M., Simmons, A. J., and Suttie, M., Aerosol analysis and forecast in the
ECMWF integrated forecast system: 2. Data assimilation. J. Geophys. Res., 114, D13205, doi:10.1029/2008JD011115,
2009.
Gueymard, C.A., Direct solar transmittance and irradiance predictions with broadband models. Pt. 2: Validation with high-
quality measurements. Solar Energy, 74, 381–395, 2003. Corrigendum: Solar Energy, 76, 515, 2004.
Gueymard, C.A., Importance of atmospheric turbidity and associated uncertainties in solar radiation and luminous efficacy
modelling. Energy 30, 1603–1621, 2005.
Gueymard, C.A, Advanced solar irradiance model and procedure for spectral solar heat gain calculation. ASHRAE Trans.,
113(1), DA-07-016 (RP-1143), 2007.
Gueymard, C.A., REST2: High performance solar radiation model for cloudless-sky irradiance, illuminance and photosyn-
thetically active radiation—validation with a benchmark dataset. Solar Energy, 82, 272–285, 2008.
Gueymard C.A., Clear-sky irradiance predictions for solar resource mapping and large-scale applications: Improved valida-
tion methodology and detailed performance analysis of 18 broadband radiative models. Solar Energy, 86, 2145-2169,
2012a.
Gueymard C.A., Temporal variability in direct and global irradiance at various time scales as affected by aerosols. Solar En-
ergy, 86, 3544-3553, 2012b.
Gueymard, C.A. and Thevenard, D., Monthly average clear-sky broadband irradiance database for worldwide solar heat gain
and building cooling load calculations. Solar Energy, 83, 1998–2018, 2009.
Holben B.N. et al., An emerging ground-based aerosol climatology: Aerosol optical depth from AERONET. J. Geophys
Res., 106, D11, 12 067-12 097, 2001.
Kasten F. and Young T., Revised optical air mass tables and approximation formula. Applied Optics 28, 4735–4738, 1989.
Kinne S. et al., An AeroCom initial assessment – optical properties in aerosol component modules of global models. At-
mos. Chem. Phys., 6, 1815–1834, 2006.
Kinne S., Remote sensing data combinations: superior global maps for aerosol optical depth. In: Satellite Aerosol Remote
Sensing over Land, A.A. Kokhanovsky and G. de Leeuw, Eds., Springer, 2009.
Levy, R., Remer, L., Kleidman, R., Mattoo, S., Ichoku, C., Kahn, R., and Eck, T., Global evaluation of the Collection 5
MODIS dark-target aerosol products over land, Atmos. Chem. Phys., 10, 10 399–10 420, 2010.
Liu J., Zheng Y., Li Z. and Wu R., Ground-based remote sensing of aerosol optical properties in one city in Northwest
China. Atmos. Res., 89, 194-205, 2008.
Morcrette, J.J., Boucher, O., Jones, L., Salmond, D., Bechtold, P., Beljaars, A., Benedetti, A., Bonet, A., Kaiser, J. W., Raz-
inger, M., Schulz, M., Serrar, S., Simmons, A. J., Sofiev, M., Suttie, M., Tompkins, A. M., and Untch, A., Aerosol analy-
sis and forecast in the ECMWF integrated forecast system: forward modeling. J. Geophys. Res., 114, D06206,
doi:10.1029/2008JD011235, 2009.
O’Neill N.T., Ignatov A., Holben B.N. and Eck T.F., The lognormal distribution as a reference for reporting aerosol optical
depth statistics; Empirical tests using multi-year, multi-site AERONET sunphotometer data. Geophys. Res. Lett., 27,
3333-3336, 2000.
Pan L. et al., Aerosol optical properties based on ground measurements over the Chinese Yangtze Delta Region. Atmos.
Environ., 44, 2587-2596, 2010.
Ruiz-Arias J.A., Dudhia J., Gueymard C.A. and Pozo-Vazquez D., Assessment of the Level-3 MODIS daily aerosol optical
depth in the context of surface solar radiation prediction and numerical weather modeling. Atmos. Chem. Phys. Dis-
cuss., 12, 23219-23260, 2012.
Thevenard D. and Gueymard C.A., Updating the ASHRAE climatic data for design and standards. ASHRAE Trans.,
116(2), AB-10-011 (RP-1453), 2010.
Wang K., Dickinson R.E., Wild M. and Liang S., Atmospheric impacts on climatic variability of surface incident solar radia-
tion. Atmos. Chem. Phys. Discuss., 12, 14 009–14 042, 2012.
Wild, M., Global dimming and brightening: a review. J. Geophys. Res., 114, D00D16, doi:10.1029/2008JD011470, 2009.
Wild, M., Gilgen, H., Roesch, A., Ohmura, A., Long, C. N., Dutton, E. G., Forgan, B., Kallis, A., Russak, V., and Tsvetkov,
A., From dimming to brightening: decadal changes in solar radiation at Earth’s surface. Science, 308, 847–850, 2005.

View publication stats

You might also like