At-Surface Reflectance and Albedo from Satellite for

Operational Calculation of Land Surface Energy Balance

Masahiro Tasumi1; Richard G. Allen2; and Ricardo Trezza3
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

Abstract: This paper presents a rapid, operational method for estimating at-surface albedo applicable to Landsat and MODIS satellite
sensors for typical cloud-free, low-haze conditions and sensor view angles less than 20°. At-surface albedo estimates are required input to
various surface energy balance models that are applied operationally. The albedo calculation method was developed using the SMARTS2
radiative transfer model and has been applied in recent versions of the University of Idaho METRIC model as a component of the surface
energy balance for determining evapotranspiration. The albedo procedure uses atmospheric correction functions developed to require only
general humidity data and a digital elevation model. The atmospheric correction functions have a reduced structure to enhance their
operational applicability in routine instantaneous surface energy balances and to estimate evapotranspiration. The method does not require
high levels of knowledge in atmospheric physics and radiation transfer processes, common to traditional radiation transfer models, which
enhances their use by a broad range of agricultural and hydrologic scientists and engineers. The atmospheric correction and surface albedo
estimation procedures are developed primarily for use with Landsat imagery, which does not have an official albedo product. However, the
procedure is also applicable to MODIS imagery that has an official albedo product at the 1 km scale, for situations where full broadband
albedo having 500 m resolution is needed, where albedo is needed for select days having small sensor view angles for reduction of pixel
blurring, or where image striping or reflectance data fallout has occurred in the standard MODIS albedo product. Method results have
been compared to literature values and independent data sets. Test applications against MODIS albedo products in New Mexico, Florida,
and Idaho indicate that the expected error for actual albedo from the developed method is within the interval of −0.035 to +0.033 共95%
confidence level兲, equivalent to a standard error of 0.017, over broad ranges in land surface elevation, humidity, and sun angle.
DOI: 10.1061/共ASCE兲1084-0699共2008兲13:2共51兲
CE Database subject headings: Energy; Satellites; Evapotranspiration; Hydrology.

Introduction coverage than Landsat, MODIS images have become widely used
for earth observation at the moderate spatial resolution of
Landsat has been one of the primary operational earth observation 250– 1000 m.
satellites over the past three decades. With long-term historical Solar energy reflected from the earth’s surface, as observed by
image records and high spatial resolution of 30 m in the short satellite, is impacted by attenuation and scattering by the atmo-
wave bands and 60– 120 m in the thermal band, Landsat thematic sphere between the satellite sensor and the surface target. The
mapper 共TM兲 and enhanced thematic mapper plus 共ETM+ 兲 im- atmospheric effect must be eliminated during calculation of sur-
ages have been widely utilized for both research and nonresearch face reflectance and albedo 共albedo represents the broadband re-
purposes. The high resolution of Landsat makes this satellite flectance over the entire shortwave spectrum兲 and for land surface
highly valuable for agricultural and water resources management, energy balance computations. Radiation transfer models 共RTMs兲
where reflective and thermal information can be retrieved for in- are generally accepted for atmospheric correction of reflected ra-
dividual agricultural fields. On the other hand, the moderate- diation 共Staenz et al. 2002兲. Application of RTM, for example the
resolution imaging spectroradiometer 共MODIS兲 on board the MODTRAN model 共Berk et al. 1999兲, is common in research
Terra and Aqua satellites is a relatively new, but coarser, sensor, applications that require high accuracy of at-surface reflectance
available since 1999. Because MODIS produces highly auto- and integrated albedo. In operational modes of atmospheric cor-
mated, low-cost images having relatively frequent, albeit coarser rection, the related 6S RTM 共Vermote et al. 1997兲 is commonly
applied, and is more computationally efficient than MODTRAN
1
Associate Professor, University of Miyazaki, 1-1, Gakuen Kibanadai- 共e.g., Zhao et al. 2000兲. However, application of these RTM’s in
Nishi, Miyazaki 889-2192, Japan. E-mail: [email protected] routine energy balance computations by public water manage-
2
Professor, Kimberly Research Center, University of Idaho, 3793 N ment agencies, for example, the Idaho Department Water
3600 E., Kimberly, ID 83341. E-mail: [email protected] Resources that routinely applies the METRIC satellite-based
3
Visiting Associate Professor, University of Idaho, 3793 N. 3600 E., evapotranspiration 共ET兲 model 共Kramber 2002; Allen et al. 2005,
Kimberly, ID 83341. E-mail: [email protected] 2007a兲, is not popular for two reasons. First, extensive computa-
Note. Discussion open until July 1, 2008. Separate discussions must
tional requirements of most RTM’s involve the creation of lookup
be submitted for individual papers. To extend the closing date by one
month, a written request must be filed with the ASCE Managing Editor.
tables for standard atmospheric conditions to enable quick atmo-
The manuscript for this paper was submitted for review and possible spheric correction 共Liang et al. 2001兲. Second, ET and energy
publication on May 8, 2006; approved on October 31, 2006. This paper is balance model operators may not have sufficient theoretical back-
part of the Journal of Hydrologic Engineering, Vol. 13, No. 2, February ground and training to manipulate RTMs, since many operators
1, 2008. ©ASCE, ISSN 1084-0699/2008/2-51–63/$25.00. are hydrologists, engineers, or GIS spatial analysts by training.

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 51

J. Hydrol. Eng. 2008.13:51-63.

 Due to these challenges for operational application of RTM, is not image based. By limiting applications of the method to
simple, empirical atmospheric corrections have been developed clear-sky and low-haze conditions 共horizontal visibility ⬎
and frequently used during processing of Landsat images 共Moran ⬃ 12 km兲, the method appears to be applicable with little expert
et al. 2001; Liang et al. 2001, 2002a兲. The traditional dark-object decision making by operators, thereby equipping agricultural and
subtraction 共DOS兲 method and its several improved versions 共e.g., hydrologic engineers with the capability to perform needed atmo-
Chavez 1988; Teillet and Fedosejevs 1995兲 provides simple, spheric corrections with reasonable accuracy for input to evapo-
image-based correction for effects of haze on radiative attenua- transpiration models. The method uses an estimate of precipitable
tion. These techniques are most widely applied in operational water in the atmosphere that is produced from MODIS satellite-
aspects because of the convenience, even though the accuracy is based profiles or estimated, as a second alternative, from surface
often considered to be unacceptable for surfaces having reflec- humidity data only. The low data requirement allows application
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

tances of 0.15 or higher 共Chavez 1996兲. The empirical line

in developing countries where detailed atmospheric parameters
method overcomes the problem of DOS by utilizing an additional
and data required by more-sophisticated models are not widely
calibration point in a high-reflectance area. The method is consid-
available. The method does not require ground calibration targets,
ered to have sufficient accuracy for many applications 共Smith and
and therefore, can be used in historical and regional studies.
Milton 1999; Ben-Dor et al. 2004兲. However, the method requires
one 共bright兲 or two 共bright and dark兲 known ground calibration In this study, two official MODIS reflectance and albedo prod-
targets, which can prove to be a constraint for application to large ucts, MOD09 and MOD43 共Vermote and Vermeulen 1999; Schaaf
regions and for applications involving historical images. Liang et et al. 2002兲 were used for comparison with model estimates and
al. 共2001, 2002a兲 developed a relatively rapid atmospheric correc- to develop calibrations for the small path reflectance terms. The
tion model for Landsat ETM+ that corrected for heterogeneous MOD09 product constitutes at-surface band reflectance, which is
aerosol scattering and surface adjacency effects. The model uti- predominantly, but not entirely, bidirectional 共BD兲. The MOD09
lized histogram matching of reflectances for specific image cali- reflectance products were integrated to broadband albedo during
bration and to remove effects of haze and thin clouds and their this study for comparison with our calculated at-surface albedo.
shadows. The method is operationally applicable to a wide range In addition, actual surface albedo calculated using the MOD43B1
of surface and atmospheric conditions. However, the method re- 16-day albedo product was used as an additional reference basis
quires users to evaluate and optimize the procedures for identify- for comparing with our estimated at-surface albedo. Actual
ing nonhazy regions and to specify the magnitude of a low-pass at-surface albedo, which is predominately directional–hemi-
filter. spherical, is more important in surface energy balance computa-
For MODIS images, calculated surface reflectances and albedo tions than the predominantly BD albedo derived from integration
images have been distributed through NASA/USGS 共Schaaf et al. of the MOD09 product because it is a truer representation of
2002兲. However, corrected albedo products are available for albedo integrated over all directions. For near-nadir sensor view
16-day periods only, and have a minimum of 1 km resolution angles, however, differences are generally small. Preliminary
rather than the 500-m resolution of the original MODIS short- evaluations by Liang et al. 共2002b兲 showed that MOD09 and
wave image. In addition, several problems, including artifact
MOD43 reflectance and albedo products are reasonably accurate
stripes in images, have been reported for MODIS reflectance and
when the images are near-nadir and from clear-sky conditions.
albedo products 共NASA 2005兲. Perhaps even more importantly
for agricultural applications, the inclusion in the 16-day MODIS Liang et al. found error to average about 0.025 for all shortwave
albedo product of reflectances from areas of images having large bands, with largest error 共0.047兲 in MOD09 Band 5 and smallest
view angles causes substantial smearing of the final product, so error 共0.002–0.006兲 in the MOD43 broadband albedo. Both
that the effective pixel resolution increases from the stated 1 km, MODIS products are derived using complex atmospheric correc-
in reality, to approximately 2 km⫻ 3 km. The effects of the tion algorithms that are traceable to the 6S model, which is in turn
smearing in the albedo product are profound when sampling traceable to the MODTRAN correction procedure 共Vermote and
albedo 共and associated derived ET products兲 for individual agri- Vermeulen 1999; Schaaf et al. 2002兲. In this study, we carefully
cultural fields or small land use areas. Therefore, independent reviewed and avoided effects of sensor malfunctioning, and there-
calculation of surface reflectance and albedo is, typically, required fore, the MOD09 and MOD43 MODIS products were considered
with MODIS imagery when albedo for individual days having to be reliable tools for calibration and comparison.
small sensor view angle and thus highest spatial resolution are
obtainable.
The surface albedo estimation method presented in this paper Traditional Albedo Estimation Methods in
was developed to improve the accuracy of net radiation compo- Operational Surface Energy Balance Models
nents in commonly applied instantaneous surface energy balance
and ET estimation models including SEBAL 共Bastiaanssen et al.
In operational application of instantaneous surface energy balance
1998; 2005; Bastiaanssen, 2000兲, METRIC 共Allen et al. 2007b兲,
SEBI 共Roerink et al. 2000兲, and SEBS 共Su 2002; Jia et al. 2003兲. models, only a limited number of applications have utilized pre-
The method attempts to reproduce RTM-based atmospheric cor- cise atmospheric correction and albedo estimation 共e.g., Ma et al.
rections using a reduced equation format introduced by Majumdar 2004兲 due to the physical and technical constraints previously
et al. 共1972兲 for atmospheric transmittance of global solar beam discussed. Instead, many widely used and operationally appli-
radiation that was adopted by the Food and Agricultural Organi- cable models such as SEBAL and SEBI have traditionally applied
zation 共FAO兲 of the United Nations and the Environmental and what may be considered to be overly simplified albedo computa-
Water Resources Institute 共EWRI兲 of the American Society of tion strategies using equations similar to Eq. 共1兲, where broad-
Civil Engineers guidelines for standardized calculation of refer- band at-surface albedo, ␣s, is estimated on a broad-spectrum basis
ence evapotranspiration 共Allen et al. 1998; EWRI 2005兲. Unlike 共Bastiaanssen et al. 1998; Farah and Bastiaanssen 2001; Roerink
the majority of empirical and semiempirical methods, our method et al. 2000; Tasumi et al. 2000兲:

52 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

 ␣t − ␣a
␣s = 共1兲
␶2bb
where ␣t = at-satellite broadband bidirectional albedo; ␣a
= atmospheric 共path兲 induced albedo; and ␶2bb = two-way broad-
band atmospheric transmittance that assumes equal transmissivity
for incoming and outgoing radiation.
The simple broadband method such as in Eq. 共1兲 is beset by
three problems. First, the concept of using a wide-spectrum or
“broadband” transmittance is inappropriate for correction of dis-
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

cretely banded images, where Landsat TM/ETM⫹ and MODIS

land observation sensors have been selectively designed to func-
tion in high-transparency spectral bands. Second, this method
does not consider differences among atmospheric transmittances
of individual sensor bands. Third, the integration of narrowband
to broadband reflectance should be done using band weights that
are proportional to the intensity of solar radiation at the surface
共Starks et al. 1991兲, which is not commonly done with this Fig. 1. Stripe in MOD09 surface reflectance Band 5, observed on
method. Recent research indicates that application of Eq. 共1兲 may August 9, 2004, in southern Idaho
underestimate net radiation as used in surface energy balance
computations and resulting ET estimates over bright, bare soil
surfaces by as much as 15% 共Tasumi et al. 2005兲. method for Step 1 is found in Chander and Markham 共2003兲 and
LPSO 共2006兲, with modifications for surface inclination of the
landscape 共Duffie and Beckman 1991; see the Appendix兲. The
New Albedo Estimation Method developed procedure utilizes Landsat TM/ ETM+ Bands 1–5 and
7, or MODIS Bands 1–7. Integration weights applied in Step 3
The operational method developed here estimates surface albedo must be modified if any of the suggested bands are rejected due to
by separately calculating at-surface reflectance for each satellite sensor malfunctioning, which is a common issue of MODIS im-
sensor band and then integrating these reflectances according to ages 共NASA 2005; See Fig. 1, as an example兲. Modification
intensities of at-surface solar radiation 共incoming兲 within the do- of weights is straightforward with the developed weighting
main of the band. The method does not correct for any bidirec- procedure.
tional reflectance distribution caused by a substantial off-nadir
sensor view angle, such as that occurring with MODIS at large
scan angles, where shadowing within plant canopies or preferen- At-Surface Reflectance
tial forward reflection of the solar beam causes deviation in In Step 2, at-surface reflectance is derived by correcting the at-
reflected signals to the sensor as compared to near nadir position- satellite BD reflectance for scattering and absorption of incoming
ing. Therefore, the equations developed here are most applicable solar radiation and scattered reflected radiation from the surface.
to near-nadir sensor view angles 共within 0 to 20° of nadir, ap- Exoatmospheric solar radiation must be converted to at-surface
proximately兲 unless some type of bidirectional reflectance distri- solar radiation, which is a mixture of directional and hemispheri-
bution function 共BRDF兲 correction is applied. The method was cal components. This is done using a calibrated atmospheric
developed for use with high-resolution 共30 m兲 Landsat images transmittance function. At-satellite observed directional radiance
and applications using near-nadir MODIS images where 500-m is converted to at-surface reflected directional radiance using a
resolution albedo is desired. combination of atmospheric transmittance function and path re-
The albedo calculation process comprises the following steps: flectance. In our procedure we have developed an atmospheric
1. Calculation of at-satellite bidirectional reflectance from at- correction model that requires only an estimate of total precipi-
satellite radiance values assuming the absence of an atmo- table water or, if that is unavailable, general humidity data.
sphere. The at-surface reflectance for band “b,” ␳s,b, is calculated in
2. Calculation of at-surface reflectance from at-satellite BD re- our procedure as
flectance values 共i.e., application of atmospheric correction兲.
The calculated at-surface reflectance is not entirely, but is Rout,s,b ␳t,b − ␳a,b
predominately, BD reflectance, since it is calculated using ␳s,b = = 共2兲
Rin,s,b ␶in,b · ␶out,b
information measured by the satellite sensor, which is a “di-
rectional” sensor. Whereas, at-surface solar radiation is a where Rin,s,b and Rout,s,b = mean at-surface incoming and reflected
mixture of beam 共i.e., directional兲 and diffuse 共i.e., hemi- radiances within band b 共W m−2 ␮m−1兲, and ␶in,b and ␶out,b
spherical兲 components, where the directional component is = effective narrowband transmittances for incoming solar radia-
predominant under clear sky conditions. tion and for shortwave radiation reflected from the surface in
3. Estimation of broadband surface albedo by integrating the band b. Parameter ␳t,b = at-satellite reflectance for band b calcu-
at-surface band reflectances. lated at the top of the atmosphere 共See the Appendix兲, and ␳a,b is
For Landsat images, the processing begins with Step 1. For an atmospheric path reflectance for band b. ␶in,b and ␶out,b account
MODIS images, the initial processing step depends on the prod- for attenuation of both beam and diffuse radiation and are calcu-
uct level of the MODIS image. Processing begins with Step 1 if lated using individually derived equations for each band of the
Level 1 calibrated radiance is used as input, or from Step 3 if satellite. ␳a,b is calculated for each band as a function of
Level 2 daily surface reflectance is used as input. The detailed transmittance.

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 53

J. Hydrol. Eng. 2008.13:51-63.

 The transmittance functions for ␶in,b and ␶out,b follow a format
similar to the broadband 共global兲 transmittance function of Ma-
jumdar et al. 共1972兲 that was adapted by FAO and EWRI stan-
␶out,b = C1 exp 冋 C2 Pair − C3W − C4
cos ␩
册
+ C5 共6兲

dardizations for calculating evapotranspiration 共Allen et al.

1998; EWRI 2005兲 and extended in this study to narrowband where ␩ = sensor view angle relative to the perpendicular from a
applications flat, horizontal surface, which is zero for a nadir view and
␲ / 2 rad for a horizontal view angle. Mean biases caused by real
␶in,b = C1 exp 冋 C2 Pair − C3W − C4
cos ␪
+ C5 册 共3兲
differences between ␶in,b and ␶out,b were incorporated into the ␳a,b
term of Eq. 共3兲 during calibration.
Landsat has a near-nadir view angle, and thus, cos ␩ for the
where Pair = atmospheric pressure 共kPa兲; ␪ = solar zenith angle for entire image can be assumed to be 1. MODIS image sensor angles
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

horizontal surfaces; W = precipitable water in the atmosphere vary each day, occasionally exceeding ␲ / 4 rad 共50°兲, and there-
共mm兲; and C1 – C5 are fitted satellite-dependent constants pre- fore, ␩ changes by image date and with each pixel location. The
sented later in the calibration section. Pair / cos ␪⫽surrogate for sensor-dependent constants C1 – C5 in Eqs. 共3兲 and 共6兲, and the
atmospheric mass and optical path length. atmospheric reflectance of each band, ␳a,b in Eq. 共2兲, were cali-
The solar zenith angle ␪ in Eq. 共3兲, which is relative to the brated using radiative transfer models as described in a following
perpendicular to a horizontal flat surface, is used even for appli- section.
cations in mountainous terrain to properly estimate relative atmo-
spheric thickness. The original broadband transmissivity equation
by Allen et al. 共1998兲 and EWRI 共2005兲, which has a similar form Estimation of Broadband Surface Albedo by
to Eq. 共3兲, contains an atmospheric turbidity term that is difficult Integrating Narrowband At-Surface Reflectances
to calibrate for specific applications and was, therefore, dropped In Step 3, broadband surface albedo is calculated from multiband
from the narrow band equations, which are, therefore, limited to satellite data by integrating band reflectances across the short-
low-haze conditions 共horizontal visibility ⬎ ⬃ 12 km兲. It is recog- wave spectrum. Here, we integrate at-surface band reflectance
nized that atmospheric transmissivity, and thus correction, is sen- following Starks et al. 共1991兲 where
sitive to aerosol levels 共e.g., Wen et al., 1999兲. For high-haze
images contaminated by mineral dust or anthropogenic pollutants,
an image-based correction method should be applied. Horizontal 7
visibility information is, typically, available from nearby airports. ␣s = 兺 关␳s,b · wb兴 共7兲
Mean atmospheric pressure 共kPa兲 is calculated following b=1
EWRI 共2005兲 as

冉 冊 5.26 where wb = weighting coefficient representing the fraction of at-

293 − 0.0065z
Pair = 101.3 共4兲 surface solar radiation occurring within the spectral range repre-
293 sented by a specific band
where 293= standard air temperature 共K兲 for agricultural condi-

冕
tions; and z = elevation above sea level 共m兲. UPb
MODIS total precipitable water maps, available in the MOD05 Rs␭ · d␭
science product 共Gao and Kaufman 1999兲 are recommended as a LOb
共8兲
冕
convenient source of data for W. The MODIS maps are available wb = 4.0
at 1 km spatial resolution. In the absence of profile-based infor- Rs␭ · d␭
mation on W, W can be calculated using measured or estimated 0.3
near-surface vapor pressure from a representative weather station,
for example, using the equation derived for North America by where Rs␭ = at-surface spectral hemispherical solar radiation for
Garrison and Adler 共1990兲 wavelength ␭ 共␮m兲; UPb and LOb = upper- and lower-wavelength
bounds 共␮m兲 assigned to Landsat or MODIS band b, which, in
W = 0.14ea Pair + 2.1 共5兲
our application, includes all regions between the specific satellite
where ea = near-surface vapor pressure 共kPa兲; and W is in milli- bands. Inclusion of wavelength regions between sensor bands in
meters. the determination of weighting provides a more theoretically cor-
Eq. 共5兲 is utilized in standardized evapotranspiration computa- rect estimate of total surface reflectance assuming that reflectance
tions by FAO 共Allen et al., 1998兲 and EWRI 共2005兲. However, W of regions between satellite bands can be approximated by linear
estimated by near-surface point measurement may not explain the interpolation of reflectances for neighboring bands. The regions
spatial variability of W within a satellite image. If Eq. 共5兲 is between satellite bands were arbitrarily divided midway between
applied, the operator should use weather data that represent aver- band edges as shown in Table 1. Overall endpoints of integration
age weather conditions for the image. were chosen as 0.3– 4.0 ␮m, which covers 98% of total at-surface
Parameter ␶in,b can vary from ␶out,b due to impacts of backscat- solar radiation. For the MODIS band coefficients, only the first
tered diffuse radiation originating from radiation from the re- seven bands of MODIS were utilized since these represent sig-
flected surface that is contained in ␶in,b but not in ␶out,b. However, nificant regions of the broad spectrum, are similar to bands of
in this study, outgoing narrowband effective transmittance, ␶out,b, Landsat, have native resolution of 500 m at sensor nadir, and are
was found to be sufficiently well approximated using the same those used by MODIS albedo products 共Schaaf et al. 2002兲. At-
coefficients as for incoming transmittance, but with equivalent surface solar radiation is used for weighting rather than exoatmo-
atmospheric path length calculated for the angle between the land spheric radiation, as is sometimes done, because at-surface values
surface and sensor 共i.e., satellite兲. Therefore, the cosine of the account for the influence of atmospheric attenuation that varies
sensor view angle is used rather than solar zenith angle with wavelength.

54 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

 Table 1. Landsat and MODIS Band Widths and Ranges for LOb and UPb Applied in Eq. 共8兲
Band limits 共␮m兲 Applied LOb–UPb 共␮m兲
a
Band Landsat MODIS Landsat MODIS
1 0.45–0.52 0.620–0.670 0.300–0.520 0.593–0.756
2 0.52–0.60 0.841–0.876 0.520–0.615 0.756–1.053
3 0.63–0.69 0.459–0.479 0.615–0.725 0.300–0.512
4 0.76–0.90 0.545–0.565 0.725–1.225 0.512–0.593
5 1.55–1.75 1.230–1.250 1.225–1.915 1.053–1.439
6 10.4–12.5 1.628–1.652 Thermal 1.439–1.879
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

7 2.08–2.35 2.105–2.155 1.915–4.000 1.879–4.000

a
Official band wavelength for Landsat 5TM. Band limits for Landsat 7 ETM⫹ are slightly different in the following bands: Band 2 共0.53–0.61兲; Band 4
共0.78–0.90兲; and Band 7 共2.09–2.35兲.

Parameter Determination 0.2. The value 0.2 is commonly used in model runs representing
agricultural settings and is used in the RTM to simulate backscat-
The constants C1 – C5 in Eqs. 共3兲 and 共6兲 were calibrated using the ter of reflected radiation to the surface. The U.S. Standard Atmo-
SMARTS2 radiative transfer model developed by Gueymard sphere 1976 共NOAA/NASA/USAF 1976兲 was used to represent
共1995兲. The SMARTS2 model has been distributed by the U.S. humidity and temperature profiles and aerosol conditions, with
National Renewable Energy Laboratory 共NREL兲 and provides humidity profiles modified according to total precipitable water.
spectral exoatmospheric and at-surface solar radiation over small Table 2 summarizes the ranges and increments evaluated for solar
band increments based on user-defined atmospheric and geomet- zenith angle, precipitable water and elevation 共5 ⫻ 5 ⫻ 4 = 100
ric conditions. SMARTS2 has been shown to compare closely combinations兲. These ranges represent a wide range of solar
with MODTRAN simulations over typical clear sky conditions angles and atmospheric water vapor contents expected above both
共Gueymard 1995兲. Spectral transmittance for specific wavelength dry desert and humid terrain. Elevations ranged from sea level to
increments was obtained from model results by ratioing at-surface that of tall mountains. The SMARTS2 outputs were used to cali-
and exoatmospheric values. brate C1 – C5 using multiple regression, where the exponential
SMARTS2 model simulations were made for 100 combina- term of Eq. 共3兲 was regarded as a single variable, Y, with the
tions of solar zenith angle, air water content, and land elevation equation expressed as a simple linear equation ␶in,b = C1 · Y + C5.
for a hypothetical surface having uniform spectral reflectance of C2, C3, and C4 were selected to maximize the R square of the
regression of Y versus ␶in,b and parameters C1 and C5 were solved
to force the final relationship to have slope= 1 and intercept= 0
Table 2. Levels of Solar Zenith Angle, Precipitable Water, and Location relative to the original data.
Elevation Evaluated during SMARTS2 Simulations Calibration constants are summarized in Table 3 for Landsat
Parameter 1 2 3 4 5
and in Table 4 for MODIS. Fig. 2 shows the calculated and pre-
dicted ␶in,b for Landsat. R-square values exceeded 0.99 for each of
Solar zenith angle, 0 0.5548 0.7954 0.9884 1.1593 the six Landsat bands and for each of the seven MODIS bands for
␪ 共rad兲 the entire calibration data set. As evidenced in Fig. 2, the reduced
Precipitable water, 3 5 12 40 60 model of Eq. 共3兲, with the fitted calibration constants, closely
W 共mm兲
reproduced the original SMARTS2 transmittances across the wide
Elevation, z 共m兲 0 1,200 2,400 4,000 —

Table 3. Calibrated Landsat Constants C1 to C5 for Eqs. 共3兲 and 共6兲 and Cb for ␳a,b as Defined in Eq. 共9兲
Coefficient Band 1 Band 2 Band 3 Band 4 Band 5 Band 7
C1 0.987 2.319 0.951 0.375 0.234 0.365
C2 −0.00071 −0.00016 −0.00033 −0.00048 −0.00101 −0.00097
C3 0.000036 0.000105 0.00028 0.005018 0.004336 0.004296
C4 0.0880 0.0437 0.0875 0.1355 0.0560 0.0155
C5 0.0789 −1.2697 0.1014 0.6621 0.7757 0.639
Cb 0.640 0.310 0.286 0.189 0.274 −0.186

Table 4. Calibrated MODIS Constants C1 – C5 for Eqs. 共3兲 and 共6兲 and Cb for ␳a,b as Defined in Eq. 共9兲
Coefficient Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7
C1 1.102 0.451 0.996 1.944 0.318 0.216 0.275
C2 −0.00023 −0.00023 −0.00071 −0.00016 −0.00022 −0.00050 −0.00031
C3 0.000290 0.000550 0.000036 0.000105 0.000640 0.000800 0.004296
C4 0.0875 0.0900 0.0880 0.0540 0.0760 0.0940 0.0155
C5 −0.0471 0.5875 0.0678 −0.8870 0.7100 0.8006 0.7282
Cb 0.262 0.397 0.679 0.343 0.680 0.639 −0.464

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 55

J. Hydrol. Eng. 2008.13:51-63.

Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Comparison of transmittance for Landsat bands estimated by Eq. 共3兲 versus transmittance simulated by SMARTS2 for 100 combinations
of sun angle, precipitable water, and land elevation

range of parametric values. The high R-square values indicate that ␳a,b were parametrized to approximate ␳a,b under other atmo-
the relatively simple format of Eq. 共3兲 is appropriate to explain spheric and sun angle conditions by proportioning them to the
the behavior of atmospheric correction as simulated by estimated amount of one-way scattered and absorbed radiation as
SMARTS2 for the given standard atmosphere over the wide range represented by 1 − ␶in,b, which is the potential source for ␳a,b
of elevation, humidity and solar zenith angles investigated.
Analyses indicate that the transmittances estimated by Eqs. 共3兲 ␳a,b = Cb共1 − ␶in,b兲 共9兲
and 共6兲 are insensitive to assumptions concerning the specific
atmospheric profiles defined by Anderson et al. 共1986兲 and where Cb = determined scaling ratio for band “b.”
NOAA/NASA/USAF 共1976兲 used with SMARTS2 during cali- Coefficients for determining ␳a,b for use with Landsat were
bration. For example, at the broadband level, there are no differ- derived using ␳a,b from MODIS assuming that the estimates for
ences between estimates by the calibrated model 关based on the ␳a,b, if unbiased, are physical functions of the atmosphere, only,
U.S. Standard Atmosphere 1976 共NOAA/NASA/USAF 1976兲兴 and are therefore applicable across satellite platforms. Therefore,
and those by SMARTS2 applied using the standard arctic atmo- coefficients for Landsat were calculated by interpolating MODIS
sphere, and only 0.5% difference when SMARTS2 is operated ␳a,b coefficients within and among commonly associated bands.
using the standard tropic atmosphere. Differences are small partly The calibrated coefficients, Cb, for Eq. 共9兲 are summarized in
because we specify precipitable water and pressure using actual Tables 3 and 4 for Landsat and MODIS, respectively. The nega-
humidity and elevation data in both models. However, the small tive values for Band 7 of both Landsat and MODIS indicate some
differences do indicate general insensitivities to assumptions re- residual calibration error in ␶in,b. The Cb value for Band 7 of both
garding other atmospheric constituents and specific profile shapes Landsat and MODIS is very sensitive to small errors during cali-
for water vapor. bration, because these midinfrared bands represent less than 4%
After the transmittance equations were calibrated, atmospheric of extraterrestrical solar radiation energy, and atmospheric trans-
path reflectance, ␳a,b, was back calculated for each band by in- mittance in these bands is high. Thus, Cb for Band 7 is determined
verting Eq. 共2兲 using the MODIS 共MOD02兲 calibrated radiance by dividing one small number by another. We have elected to
product to represent ␳t,b, using the MODIS at-surface reflectance retain the negative values for Cb, because the absolute error in
product MOD09 as the reference ␳s,b, and using ␶in,b and ␶out,b total albedo caused by error in ␳a,b for Band 7 is on the order of
from Eqs. 共3兲 and 共6兲. The inverted Eq. 共2兲 was applied to a long only 0.001.
series of sampled pixels from a near-nadir, clear-sky MODIS Fig. 3 compares at-surface reflectances calculated with our at-
image over Idaho acquired on August 9, 2004. Values derived for mospheric correction equations to those from MOD09 for 50
␳a,b were numerically stable and essentially constant within each sampled pixels from the August 9, 2004, calibration image from
band over a wide range of surface reflectances. Values for ␳a,b Idaho. Our calculated values utilized a MODIS radiance image
ranged from less than 0.01 共near-infrared bands兲 to 0.08 共blue from the MOD02 product to calculate ␳t,b and applied Eqs. 共2兲,
band兲 across the seven bands of MODIS. The nearly constant 共3兲, 共6兲, and 共9兲 using the constants in Table 4. Land areas that
values derived for ␳a,b over the wide ranges of surface reflectance had nearly uniform soil and vegetation characteristics were
tested indicates relatively accurate calibration of the ␶in,b and ␶out,b sampled to avoid error introduced by georegistration inconsisten-
functions. Following the calibration of ␳a,b values, functions for cies between the two MODIS products. Comparisons at the

56 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Estimated at-surface reflectance for a MODIS Terra image on August 9, 2004, for 50 homogeneous sample surfaces over southern Idaho
关based on Eqs. 共3兲–共5兲兴, compared with the MODIS Level 2 surface reflectance and Level 3 albedo products

broadband level are also shown in Fig. 3, where broadband al- • Case 1: z = 870 m, W = 12 mm, and ␪ = 0.66 rad 共38°兲 共a typical
bedo was calculated from both our model and from the MOD09 setting for Landsat images in southern Idaho兲.
Level 2 reflectance product using Eq. 共7兲 for integration 共the latter • Case 2: z = 4000 m, W = 3 mm, and ␪ = 0 rad. 共an extremely
broadband albedo calculation is termed MOD09 albedo in Fig. 3兲. transparent atmosphere兲.
In addition, the comparative basis for the last graph in Fig. 3 is • Case 3: z = 0 m, W = 60 mm, and ␪ = 1.16 rad. 共66°兲 共a rela-
the MOD43 broadband albedo product representing the 16-day tively low transmittant atmosphere兲.
period encompassing August 9, 2004. Values for Cb in Table 4 As shown in Figure 4, wb did not change measurably over the
were calibrated using this image by constructing regression lines wide range of solar radiation settings. Thus, we propose using
in Fig. 3 having slope of 1.00 and intercept forced to zero. The fixed wb values from Case 1 for all low-haze atmospheric condi-
strong linearity between calculated and MOD09 reflectances is a tions and for all latitude, time, land elevation and atmospheric
strong indication of successful calibration. water contents. The values for Case 1 are listed in Table 5.
The weighting coefficients, wb, used in Eq. 共7兲 were deter-
mined using at-surface spectral data representing clear sky condi-
tions. As defined by Eq. 共8兲, wb is strongly related to the relative Validation of the Developed Method
intensities of at-surface solar radiation in the various bands and
relative positioning of the bands. Our analyses show that variation
Comparison of Atmospheric Correction with MODIS
in solar radiation with latitude and time occurs in nearly the same
Products
proportion across all bands, and thus, values for wb are very stable
over a wide range of solar radiation intensities and solar angles. Reflectances and albedo estimated using MODIS Level 1 cali-
Fig. 4 shows at-surface solar radiation levels derived from brated radiance data as input to Eqs. 共2兲, 共3兲, 共6兲, and 共7兲 are
SMARTS2 and calculated wb for one typical and two extreme compared here against the MODIS official surface reflectance and
combinations of land elevation, precipitable water, and solar albedo products for southern Idaho and New Mexico as tests
angle: under dry atmospheric conditions 共W ranged from about

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 57

J. Hydrol. Eng. 2008.13:51-63.

Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Within-band solar radiation and wb for three solar radiation conditions for Landsat and MODIS shortwave bands

5 to 10 mm兲, and for Florida as a test under humid atmospheric standard procedures 共Berthouex and Brown 1994兲 and was based
conditions 共W ranged from about 12 to 25 mm兲. In the Idaho test, on all samples from the three normal test applications. The pre-
an additional comparison was made for an image having large diction limits are indicated in Fig. 5. The slopes and intercepts of
sensor view angle 共35–42°兲, which is significantly greater than the linear regressions for each test application are summarized in
our recommended application limit of 20°. Table 6 summarizes Table 7. The slopes determined for Band 3 were numerically un-
characteristics of the MODIS images processed for calibration stable 共1.17 for Test-3 and 0.87 for Test EX兲 due to the small
and comparison, including the sensor view angle 共from zenith兲, variability in sampled surface reflectance for this band. However,
sample numbers and conditions of the sampled pixels. Sampled Fig. 5 suggests that all samples in the Band 3 comparison lie
pixels were selected from among relatively homogeneous sur- relatively close to a 1:1 line. The results in Fig. 5 and Table 7
faces to reduce error caused by georegistration differences be- show that surface reflectances calculated using our model agree
tween MODIS products, especially for the BRDF albedo product, closely with the official MODIS products over wide ranges of
which is a 1 km resolution product having composited informa- reflectance in each band and over the wide range of conditions
tion from a large number of view angles 共composited days兲 that represented by the three locations. Maximum error between esti-
causes smearing and makes its effective resolution larger than mated and MOD09 produced reflectances were −0.037 共Band 6兲
1 km. and +0.034 共Band 5兲 for surface reflectance intensities of 0.2
Fig. 5 compares band by band reflectance and computed al- 共95% confidence兲. The generally good agreement between band
bedo estimates from the model against reflectances from the reflectances indicate that the atmospheric corrections made using
MOD09 and albedo from the MOD43 products. The prediction the relatively simple Eqs. 共2兲, 共3兲, and 共6兲 produce nearly the
confidence interval 共at 95% confidence兲 was calculated using same correction as applied in developing the MOD09 surface

Table 5. Weighting Coefficients Based on At-Surface Solar Radiation Derived from SMARTS2
Satellite Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 Total
Landsat 0.254 0.149 0.147 0.311 0.103 — 0.036 1.000
MODIS 0.215 0.215 0.242 0.129 0.101 0.062 0.036 1.000

Table 6. Summary Description of MODIS Images Used for Five Tests of Reflectance Model Applications
Sensor
Image Zenith Number of
number Purpose Date 共degrees兲 samples Conditions of samples
1 Calibration 共Idaho兲 August 9, 2004 0–6 50 From wide-open homogeneous desert bare soils
2 Test 1 共Idaho兲 August 7, 2004 16–27 65 Homogeneous bare soils, heterogeneous agriculture
3 Test 2 共New Mexico兲 April 4, 2002 0–23 48 Homogeneous bare soils, basalt and snow
4 Test 3 共Florida兲 May 4, 2004 0–10 38 Relatively homogeneous natural vegetations
5 Test EX 共Idaho, off sensor-angle兲 August 5, 2004 35–42 65 Homogeneous bare soils, heterogeneous agriculture

58 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Estimated at-surface reflectance for a MODIS Terra image for three test applications over Idaho 共ID兲, New Mexico 共NM兲, and Florida
共FL兲 and one extra test at large sensor view angle 共Test EX兲 using MOD02 radiance images with Eqs. 共2兲, 共3兲, 共6兲, and 共7兲, compared with the
corresponding MODIS Level 2 surface reflectance and Level 3 共MOD09 and MOD43兲 albedo products. The dotted line is the 1:1 relation and
solid lines are the limits for 95% prediction confidence computed for the three normal test applications together 共150 samples兲.

reflectance, which is traceable to the 6S and MODTRAN atmo- The comparisons with the 16-day MOD43 albedo product,
spheric correction models, in both dry and humid climates. which is corrected for BRDF and is composited from as many as
In the broadband level, our estimation closely reproduced at 16 daily images, showed larger differences. Compared with
surface broadband albedo as computed from the MOD09 reflec- MOD43, our estimation fell within −0.035 to + 0.033 共95% CI at
tance products 共Fig. 5兲, estimating within −0.013 and +0.018 of albedo intensities of ⬇0.2兲, representing a standard error of 0.017
the MOD09 albedo 共95% CI兲 over the test applications in Idaho, over the three test applications. Some of this error stems from
New Mexico, and Florida. Estimated albedo compared well with estimation error by our model, which uses a single directional
the MOD09 derived product even for the application having sen- sensor angle 共i.e., a single image having near-nadir view angle兲
sor view angles of 35–42° 共results plotted in Fig. 5兲, since both whereas, actual surface albedo can have a large directional distri-
estimates are predominately bidirectional. The good agreement bution with view angle. Other differences may stem from error in
for all test applications indicates that the developed reflectance the MOD43 product including BRDF corrections, and from time
and albedo estimation equations and calibrated coefficients are and resolution inconsistencies between two products 关i.e., instan-
robust in application, and can be used to estimate the predomi- taneous versus 16-day average and resolution of 500 m versus
nately bidirectional surface albedo 共i.e., non-BRDF corrected兲 ac- 1 km 共effectively 2 km⫻ 3 km due to view angle induced
curately over a wide range of atmospheric conditions and sensor smearing兲兴.
angles. The estimation error in our albedo algorithm is not well quan-

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 59

J. Hydrol. Eng. 2008.13:51-63.

 Table 7. Slope and Intercepts of Linear Regression Lines for Calculated and Official MODIS Band Reflectance and Broadband Albedo-Parenthesized
Values Are Intercepts
Image BB BB
number Description Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 共MOD09兲 共MOD43兲
1 Calibrationa — — — — — — — — —
2 Test 1 共ID兲 1.083 共−0.010兲 1.020 共−0.001兲 1.061 共−0.007兲 1.047 共−0.006兲 1.001 共0.002兲 1.53 共−0.014兲 0.950 共0.018兲 1.024 共−0.004兲 1.039 共−0.019兲
共r2兲 0.914 0.972 0.891 0.931 0.924 0.934 0.951 0.966 0.906
3 Test 2 共NM兲 1.114 共−0.019兲 1.008 共−0.002兲 1.083 共−0.05兲 1.096 共−0.012兲 0.975 共0.005兲 1.060 共−0.015兲 0.990 共0.007兲 1.051 共−0.008兲 1.048 共−0.008兲
共r2兲 0.983 0.985 0.989 0.988 0.968 0.989 0.991 0.985 0.948
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

4 Test 3 共FL兲 0.936 共0.004兲 1.014 共−0.003兲 1.174 共0.009兲 0.968 共0.008兲 0.992 共−0.001兲 0.969 共−0.004兲 0.896 共0.017兲 0.945 共0.012兲 1.082 共−0.001兲
共r2兲 0.909 0.981 0.904 0.912 0.936 0.894 0.891 0.934 0.723
5 Test EXb 0.993 共0.004兲 0.962 共0.011兲 0.869 共0.011兲 0.902 共0.011兲 0.995 共0.004兲 1.010 共0.000兲 0.943 共0.022兲 0.951 共0.011兲 1.017 共−0.020兲
共r2兲 0.966 0.978 0.807 0.944 0.961 0.966 0.965 0.976 0.932
Average of Tests 1–3 共r2兲 1.071 共−0.008兲 1.012 共−0.001兲 1.045 共0.000兲 1.064 共−0.005兲 0.975 共0.007兲 1.071 共−0.021兲 0.980 共0.010兲 1.032 共−0.004兲 1.022 共−0.008兲
0.981 0.976 0.971 0.984 0.948 0.974 0.985 0.983 0.919
a
Intercept was forced to be 0.00 and slope was forced to be 1.00 during calibration 共See Fig. 3兲.
b
Text EX is a test for off sensor zenith angle 共35–42°, which is significantly more than the application limit of 20°兲.

tified from the comparison with MOD43 albedo for these reasons ing coefficients developed by us and by Liang. A primary benefit
and given the relatively small number of test applications. How- of our weighting coefficients stems from their derivation based on
ever, the randomness of differences between the two albedo esti- Equation 共8兲 that allows a new set of coefficients to be derived
mates 共our model versus MOD43兲 indicates that the differences when a band is missing from an image due to corruption, such as
are primarily caused by time inconsistencies between the two shown in Fig. 1. In this case, new coefficients are readily derived
albedo products 共i.e., instantaneous versus 16-day average兲. In using the information in Table 5 by assigning one-half the value
typical Landsat and MODIS-based surface energy balance appli- of the coefficient for the missing band to the coefficients of the
cations, 0.03 error in the surface albedo estimation causes less bands on either side of the missing band. Table 1 provides the
than 2% error in net radiation calculations. Therefore, we con- spectral order of band numbers for the Landsat and MODIS
clude that our developed albedo algorithm provides albedo with satellites.
sufficient accuracy for use in surface energy balance studies that
estimate spatial distribution of ET, such as the METRIC model
Combined Comparison of Atmospheric Correction
共Allen et al. 2007b兲.
and Integration Coefficients with the EWRI General
Transmissivity Model
Comparison of the Integration Procedures for
The development of the simple atmospheric correction algorithms
Broadband At-Surface Albedo with Liang „2000…
for individual satellite sensor bands relied substantially on the
The weighting coefficients, wb, used in this study for integrating SMARTS2 model and developed equations closely reproduce the
reflectances into broadband albedo were derived for use with at- SMARTS2 estimates over wide ranges of solar zenith angle,
surface solar radiation under standard atmospheric conditions. ground elevation and precipitable water for the standard profile
Broadband albedo estimated using our wb values were compared conditions, as was shown in Figure 2. As an independent assess-
to albedo estimated using weighting coefficients derived by Liang ment of SMARTS2, simulations of transmittance by SMARTS2
共2000兲 using regression. Comparisons were made for a Landsat
Path 40 Row 30 in southern Idaho acquired August 22, 2000. Fig.
6 shows comparisons for 250 randomly sampled pixels that in-
cluded irrigated agriculture, bare soil, city, water, desert, basalt
rock, and salt-encrusted surfaces. The same at-surface band re-
flectances were applied for both sets of calculations, thus, the
differences in computed albedo between the two methods are
purely due to differences in wb. The two independently produced
sets for wb produced essentially identical results for nearly all
sampled pixels. However, results systematically deviated for
green agricultural fields having high values for the normalized
difference vegetation index 共NDVI兲. Differences at high NDVI
were caused by the omission of the green band by Liang. It ap-
pears that our theoretically based method for band integration
provides better estimates for surface albedo for the dense, green
agricultural condition. Some random deviation between results
from the two weighting methods was observed over extremely Fig. 6. Estimated broadband albedo using integration functions by
bright surfaces having albedo greater than about 0.4. The scatter Liang 共2000兲 versus those from Eq. 共7兲 and Table 5 for an August 22,
indicates some uncertainty in accuracy of estimates by either 2000, Landsat image in southern Idaho for irrigated agriculture, bare
methods under these extreme conditions. The overall agreement soil, city, water, desert, dark basalt rock, and bright salt-encrusted
between albedo estimates supports the consistency of the weight- surfaces

60 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

 Testing of the Path Reflectance Coefficients
Calculated atmospheric path reflectance, ␳a,b, as determined dur-
ing this study from MODIS data products was, to some degree,
independently evaluated using the MODTRAN radiative transfer
model. The evaluation was made under Idaho summer conditions
and differences between path reflectances calculated by our cali-
brated Equation 共9兲 and derived from MODTRAN are summa-
rized in Table 8. Path reflectances were isolated from MODTRAN
runs by dividing out transmittance estimated using Eqs. 共3兲 and
共6兲. Differences between the two methods were small, ranging as
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

high as 0.024 in some narrow bands and less than 0.006 on a

broadband basis. These differences are considered to be within
accuracy tolerances for estimating surface albedo and they indi-
cate similarities in path reflectance from two mostly independent
atmospheric correction models 共MODTRAN and that used in
MODIS products兲. The path reflectance values from our cali-
Fig. 7. Comparison between broadband transmittance from brated model, when integrated using Eq. 共7兲, are close to literature
SMARTS2 and EWRI for 100 combinations of weather and geomet- values of 0.035–0.038 determined by Bastiaanssen 共2000兲 in Tur-
ric conditions described in Table 5. The regression line and the 95% key and 0.030 determined by Tasumi 共2003兲 in Idaho using
prediction confidence intervals 共shown as solid lines兲 are for typical known dark objects from Landsat satellite images.
situations only.

Conclusions
were compared, in terms of broadband transmittance integrated
over the 0.3 to 4.0 ␮m range, with the broadband transmittance We present a surface albedo estimation method that can be ap-
equation of EWRI 共2005兲, which was calibrated using global plied to Landsat and MODIS imagery for deriving at-surface
pyranometer measurements from 49 quality-controlled agricul- albedo from near-nadir sensor view angles for use in energy bal-
tural weather stations located across the United States 共Itenfisu ance determination of ET. The developed method was calibrated
et al. 2003; EWRI 2005兲. Results of this comparison are shown in using the SMARTS2 radiative transfer model for atmospheric
Fig. 7 where estimated transmittances from the two independent transmittance and MODIS surface reflectance products for atmo-
methods agreed well over the same 100 combinations of location spheric path reflectance. The method is composed of simplified,
elevation, precipitable water, and sun angle as used in develop- explicit equations, with a data requirement of air-humidity and
ment of the narrow band transmittance functions. SMARTS2 digital elevation model 共DEM兲 only. Therefore, the method can
broadband transmittance values averaged only 1% higher than be utilized in operational and routine applications with satellite
those by the EWRI equation, with a standard error of 0.017. Data imagery to estimate incoming and outgoing atmospheric transmit-
points having significantly different values between the two meth- tance for low haze, cloud free conditions. Although the method is
ods occurred only for extreme sets of conditions; for example, at applicable for nonnadir images, the estimation accuracy of at-
4 , 000 m elevation, with extremely high precipitable water, as surface reflectance degrades as sensor view angle increases due to
shown in Fig. 7. These extreme conditions are probably outside the lack of consideration of BRDF. A maximum sensor view
the valid range of the EWRI equation. The close agreement be- angle of 20° is recommended in the absence of BRDF correction.
tween the two independent calculations indicates that SMARTS2 This is not a concern with Landsat images that are always near
transmittances reproduce empirically observed transmittances nadir.
over wide ranges in location and climate as represented by the 49 The developed method was tested and compared against
weather stations at the broadband level. These findings indirectly MODIS at-surface reflectance and albedo products 共MOD09 and
imply accuracy of spectral transmittances estimated by MOD43兲, against a general broadband transmittance equation
SMARTS2 and in turn, accuracy of the narrow band transmit- calibrated using solar radiation data from 49 U.S. locations, and
tance equations and integrated weighting functions developed in against literature values. Although robust validations using actual
this study. The close agreement also indicates that the weighting ground-based measurements have not been conducted, the results
of the atmospheric pressure and precipitable water terms of the of the comparisons indicate good estimation accuracy 共±0.035
EWRI 共2005兲 equation is appropriate. maximum error in actual surface albedo and ±0.018 for bidirec-

Table 8. Comparison between MODIS Calibrated and MODTRAN Calibrated Path Reflectance
Bands Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7 Averagea
Landsat Calibrated 0.079 0.043 0.027 0.011 0.012 — −0.012 0.035
MODTRAN 0.073 0.060 0.034 0.019 0.018 — 0.015 0.041
MODIS Calibrated 0.027 0.009 0.087 0.042 0.005 0.014 −0.012 0.036
MODTRAN 0.032 0.008 0.063 0.032 0.001 0.003 0.009 0.030
Note: Expressed in units of reflectance 共dimensionless兲.
a
Weighted average calculated by wb in Table 6.

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 61

J. Hydrol. Eng. 2008.13:51-63.

 Table 9. Mean Solar Exoatmospheric Spectral Radiation, ESUNb and aspect information derived from a digital elevation model or
共W m−2 ␮m−1兲, for Landsat and MODIS other means using the following equation 共Duffie and Beckman
Band Landsat 4a Landsat 5a Landsat 7b MODISc 1991兲:
1 1,957 1,957 1,969 1,596
cos ␪rel = sin共␦兲sin共␾兲cos共s兲 − sin共␦兲cos共␾兲sin共s兲cos共␥兲
2 1,825 1,826 1,840 974.7
3 1,557 1,554 1,551 2,017 + cos共␦兲cos共␾兲cos共s兲cos共␻兲
4 1,033 1,036 1,044 1,850 + cos共␦兲sin共␾兲sin共s兲cos共␥兲cos共␻兲
5 214.9 215.0 225.7 463.1
6 — — — 232.9 + cos共␦兲sin共s兲sin共␥兲sin共␻兲 共12兲
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

7 80.72 80.67 82.07 92.67 where ␦ = declination of the earth 共positive in summer in the
8 — — 1368 — northern hemisphere兲; ␾ = latitude of the pixel 共positive for the
a
Chander and Markham 共2003兲. northern hemisphere and negative for the southern hemisphere兲;
b
LPSO 共2006兲. s = surface slope where s = 0 for horizontal and s = ␲ / 2 rad for
c
Calculated using the SMARTS2 共Gueymard 1995兲 spectral ESUN data vertical downward slope 共s is always positive and represents the
set and MODIS band sensor response. downward slope in any direction兲; and ␥ = surface aspect angle,
where ␥ = 0 for slopes oriented due south, ␥ = – ␲ / 2 rad for slopes
tional albedo at 95% confidence兲 for use in operational applica- oriented due east, ␥ = + ␲ / 2 rad for slopes oriented due west and
tions of land surface energy balance models. The transmittance ␥ = ± ␲ rad for slopes oriented due north. Parameter ␻ is the hour
equation calibrated in this study closely reproduced the transmit- angle, where ␻ = 0 at solar noon, ␻ is negative in morning and ␻
tance simulated by the SMARTS2 radiative transfer model and is positive in afternoon following the convention of Duffie and
corresponds well with the EWRI 共2005兲 broadband transmittance Beckman 共1991兲.
method. The calibrated atmospheric path reflectances agreed well
with an independent test using MODTRAN and also with litera-
ture values. The theoretical weighting coefficients used to inte- References
grate band reflectances to broadband albedo agree well with the
independently derived method of Liang 共2000兲. The procedure Allen, R. G., Pereira, L. S., Raes, D., and Smith, M. 共1998兲. “Crop evapo-
presented in this paper is applicable to other satellite sensors for transpiration.” FAO Irrigation and drainage Paper, 56, FAO, Rome,
development of specific coefficients for Eqs. 共3兲 and 共6兲 and Italy.
weighting coefficients for Eq. 共7兲. Allen, R. G., Tasumi, M., Morse, A., and Trezza, R. 共2005兲. “A Landsat-
based energy balance and evapotranspiration model in western U.S.
water rights regulation and planning.” J. Irrig. Drain. Sys., 19, 251–
268.
Appendix. Calculation of At-Satellite Bidirectional Allen, R. G., Tasumi, M., Morse, A., Trezza, R., Wright, J. L., Bastiaans-
Reflectance sen, W., Kramber, W., Lorite, I., and Robison, C. W. 共2007a兲.
“Satellite-based energy balance for mapping evapotranspiration with
The digital number 共DN兲 of a Landsat image is convertible to internalized calibration 共METRIC兲—Applications.” J. Irrig. Drain.
radiance 共W m−2 sr−1 ␮m−1兲 using the method and calibration co- Eng., 133共4兲, 395–406.
efficients described in Chander and Markham 共2003兲 for Landsat Allen, R. G., Tasumi, M., and Trezza, R. 共2007b兲. “Satellite-based energy
5 and LPSO 共2006兲 for Landsat 7. MODIS 共MOD02 Level 1b兲 balance for mapping evapotranspiration with internalized calibration
image contains the calibration numbers in the header information 共METRIC兲—Model.” J. Irrig. Drain. Eng., 133共4兲, 380–394.
Anderson, G. P., Clough, S. A., Kneizys, F. X., Chetwynd, J. H., and
of the image.
Shettle, E. P. 共1986兲. “AFGL atmospheric constituent profiles
At-satellite 共BD兲 band reflectance, ␳t,b is calculated from at-
共0 – 120 km兲.” Technical Rep. AFGL-TR-86-0110, Air Force Geophys-
satellite directional radiance as ics Laboratory, Hanscom AFB, Mass.
␲ · Lt,b · d2 Bastiaanseen, W. G. M. 共2000兲. “SEBAL-based sensible and latent heat
␳t,b = 共10兲 fluxes in the irrigated Gediz Basin, Turkey.” J. Hydrol., 229, 87–100.
ESUNb · cos ␪rel Bastiaanssen, W. G. M., Menenti, M., Feddes, R. A., and Holtslag, A. A.
where Lt,b = at-satellite spectral radiance in band b M. 共1998兲. “A remote sensing surface energy balance algorithm for
land 共SEBAL兲: 1. Formulation.” J. Hydrol., 212-213, 198–212.
共W m−2 sr−1 ␮m−1兲; ESUNb = mean solar exoatmospheric radia-
Bastiaanssen, W. G. M., Noordman, E. J. M., Pelgrum, H., Davids, G.,
tion over band b 共W m−2 ␮m−1兲 given in Table 9, ␪rel = solar inci- Thoreson, B. P., and Allen, R. G. 共2005兲. “SEBAL model with re-
dent angle 共or solar zenith angle兲 relative to the land surface motely sensed data to improve water-resources management under
slope; and d = earth–sun distance in astronomical units. actual field conditions.” J. Irrig. Drain. Eng., 131共1兲, 85–93.
Parameter d2 can be calculated as a function of day of year Ben-Dor, E., Kindel, B., and Goetz, A. F. H. 共2004兲. “Quality assessment
using Duffie and Beckman 共1991兲 of several methods to recover surface reflectance using synthetic im-
aging spectroscopy data.” Remote Sen. Environ., 90, 389–404.
1 Berk, A., et al. 共1999兲. “MODTRAN4 radiative transfer modeling for
d2 = 共11兲
1 + 0.033 cos共DOY · 2␲/365兲 atmospheric correction.” Proc. SPIE Optical Spectroscopic Tech-
niques and Instrumentation for Atmospheric and Space Research III,
where DOY= day of year and 共DOY· 2␲ / 365兲 共rad兲.
3756.
The solar incidence angle is the angle between the solar beam Berthouex, P. M., and Brown, L. C. 共1994兲. Statistics for environmental
and a vertical line perpendicular to the land surface. For horizon- engineers, Lewis, Boca Raton, Fla.
tal flat surfaces, ␪rel is simply equivalent to solar zenith angle 共i.e., Chander, G., and Markham, B. 共2003兲. “Revised Landsat-5 TM radiomet-
␲ / 2 minus the solar elevation angle兲. However, for sloping sur- ric calibration procedures and postcalibration dynamic ranges.” IEEE
faces, ␪rel must be calculated pixel by pixel, using surface slope Trans. Geosci. Remote Sens., 41共11兲, 2674–2677.

62 / JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008

J. Hydrol. Eng. 2008.13:51-63.

 Chavez, P. S., Jr. 共1988兲. “An improved dark-object subtraction technique Gonzalez-Dugo, M. P. 共2001兲. “A refined empirical line approach for
for atmospheric scattering correction of multispectral data.” Remote reflectance factor retrieval from Landsat-5 TM and Landsat-7
Sen. Environ., 24, 459–479. ETM⫹.” Remote Sensing Environ., 78, 71–82.
Chavez, P. S., Jr. 共1996兲. “Image-based atmospheric corrections— NASA. 共2005兲. “MODIS land quality assessment—Known product is-
Revised and improved.” Photogramm. Eng. Remote Sens., 62, 1025– sues.” Internet article, 具http://modland.nascom.nasa.gov/cgi-bin/
1036. QA_WWW/newPage.cgi典 共June 19, 2005兲.
Duffie., J. A., and Beckman, W. A. 共1991兲. Solar engineering of thermal National Oceanic and Atmospheric Administration, National Aeronautics
process, 2nd Ed., Wiley, New York. and Space Administration, and United States Air Force 共NOAA/
EWRI Task Committee on Standardization of Reference Evapotranspira- NASA/USAF兲. 共1976兲. U.S. standard atmosphere, 1976, U.S. Gov-
tion of the Environmental and Water Resources Institute of the ASCE ernment Printing Office, Washington, D.C.
共EWRI兲. 共2005兲. The ASCE standardized reference evapotranspira- Roerink., G. J., Su, Z., and Menenti, M. 共2000兲. “S-SEBI: A simple re-
Downloaded from ascelibrary.org by HOWARD UNIV-UNDERGRADUATE LIB on 03/10/15. Copyright ASCE. For personal use only; all rights reserved.

tion equation, R. G. Allen et al., eds., ASCE, Reston, Va. mote sensing algorithm to estimate the surface energy balance.” Phys.
Farah, H. O., and Bastiaanssen, W. G. M. 共2001兲. “Impact of spatial Chem. Earth, Part B, 25共2兲, 147–157.
variations of land surface parameters on regional evaporation: A case Schaaf, C., et al. 共2002兲. “First operational BRDF, albedo and nadir re-
study with remote sensing data.” Hydrolog. Process., 15, 1585–1607. flectance products from MODIS.” Remote Sens. Environ., 83, 135–
Gao, B.-C., and Kaufman, Y. J. 共1999兲. “The MODIS near IR water vapor 148.
algorithm.” MODIS algorithm theoretical background document, Smith, G. M., and Milton, E. J. 共1999兲. “The use of the empirical line
具http://modis-atmos.gsfc.nasa.gov/_docs/atbd_mod03.pdf典. method to calibrate remotely sensed data to reflectance.” Int. J. Re-
Garrison, J. D., and Adler, G. P. 共1990兲. “Estimation of precipitable water mote Sens., 20共13兲, 2653–2662.
over the United States for application to the division of solar radiation Starks, P. J., Norman, J. M., Blad, B. L., Walter-Shea, E. A., and Walthall,
into its direct and diffuse components.” Sol. Energy, 44共4兲, 225–241. C. L. 共1991兲. “Estimation of shortwave hemispherical reflectance 共al-
Gueymard, C. 共1995兲. “SMARTS2, a simple model of the atmospheric bedo兲 from bidirectionally reflected radiance data.” Remote Sens. En-
radiative transfer of sunshine: Algorithms and performance assess- viron., 38, 123–134.
ment.” Professional Paper FSEC-PF-270-95, Florida Solar Energy Staenz, K., Secker, J., Gao, B.-C., Davis, C., and Nadeau, C. 共2002兲.
Center, Univ. of Central Florida, Cocoa, Fla. “Radiative transfer codes applied to hyperspectral data for the re-
Itenfisu, D., Elliott, R. L., Allen, R. G., and Walter, I. A. 共2003兲. “Com- trieval of surface reflectance.” ISPRS J. Photogramm. Remote Sens.,
parison of reference evapotranspiration calculations as part of the 57, 194–203.
ASCE standardization effort.” J. Irrig. Drain. Eng., 129共6兲, 440–448. Su, Z. 共2002兲. “The surface energy balance system 共SEBS兲 for estimation
Jia, L., et al. 共2003兲. “Estimation of sensible heat flux using the surface of turbulent heat fluxes.” Hydrology Earth Syst. Sci., 6共1兲, 85–99.
energy balance system 共SEBS兲 and ATSR measurements.” Phys. Tasumi, M. 共2003兲. “Progress in operational estimation of regional evapo-
Chem. Earth, Part B, 28, 75–88. transpiration using satellite imagery.” Ph.D. dissertation, Univ. of
Kramber, W. J. 共2002兲. Developing evapotranspiration data for Idaho’s Idaho, Moscow, Idaho.
Treasure Valley using surface energy balance algorithm for land Tasumi, M., Allen, R. G., and Bastiaanssen, W. G. M. 共2000兲. “The
(SEBAL), Idaho Dept. of Water Resources, Boise, Idaho. theoretical basis of SEBAL.” Application of the SEBAL Methodology
Landsat, Project Science Office 共LPSO兲. 共2006兲. Landsat 7 science data
for Estimating Consumptive Use of Water and Streamflow Depletion
users’ handbook, Landsat Project Science Office, NASA Goddard in the Bear River Basin of Idaho through Remote Sensing: Final Rep.,
Space Flight Center, Greenbelt, Md., 具http://ltpwww.gsfc.nasa.gov/
Idaho Dept. of Water Resources, Idaho.
IAS/handbook/handbook_toc.html典 共Nov. 15, 2006兲.
Tasumi, M., Trezza, R., Allen, R. G., and Wright, J. L. 共2005兲. “Opera-
Liang, S. 共2000兲. “Narrowband to broadband conversions of land surface
tional aspects of satellite-based energy balance models for irrigated
albedo. I: Algorithms.” Remote Sensing Environ., 76, 213–238.
crops in the semi-arid U.S.” J. Irrig. Drain. Sys., 19, 355–376.
Liang, S., et al. 共2002a兲. “Atmospheric correction of Landsat ETM+ land
Teillet, P. M., and Fedosejevs, G. 共1995兲. “On the dark target approach to
surface imagery—Part II: Validation and applications.” IEEE Trans.
atmospheric correction of remotely sensed data.” Can. J. Remote
Geosci. Remote Sens., 40共12兲, 2737–2746.
Sens., 21共4兲, 374–387.
Liang, S. et al. 共2002b兲. “Validating MODIS land surface reflectance and
Vermote, E. F., Tanre, D., Deuze, J. L., Herman, M., and Morcrette, J. J.
albedo products: Methods and preliminary results.” Remote Sensing
共1997兲. “Second simulation of the satellite signal in the solar spec-
Environ., 83, 149–162.
Liang, S., Fang, H., and Chen, M. 共2001兲. “Atmospheric correction of trum: An overview.” IEEE Trans. Geosci. Remote Sens., 35, 3.
Landsat ETM⫹ land surface imagery—Part I: Methods.” IEEE Trans. Vermote, E. F., and Vermeulen, A. 共1999兲. “Atmospheric correction algo-
Geosci. Remote Sens., 39, 2490–2498. rithm: Spectral reflectances 共MOD09兲.” MODIS algorithm theoretical
Ma, Y., Menenti, M., Tsukamoto, O., Ishikawa, H., Wang, J., and Gao, Q. background document, 具http://modis.gsfc.nasa.gov/data/atbd/
共2004兲. “Remote sensing parametrization of regional land surface heat atbd_mod08.pdf典.
fluxes over arid area in northwestern China.” J. Arid Environ., 57共2兲, Wen, G., Tsay, S.-C., Cahalan, R. F., and Oreopoulos, L. 共1999兲. “Path
257–273. radiance technique for retrieving aerosol optical thickness over land.”
Majumdar, N. C., Mathur, B. L., and Kaushik, S. B. 共1972兲. “Prediction J. Geophys. Res., 104共D24兲, 31321–31332.
of direct solar radiation for low atmospheric turbidity.” Sol. Energy, Zhao, W., Tamura, M., and Takahashi, H. 共2000兲. “Atmospheric and spec-
13, 383–394. tral corrections for estimating surface albedo from satellite data using
Moran, M. S., Bryant, R., Thome, K., Ni, W., Nouvellon, Y., and 6S code.” Remote Sensing Environ., 76, 202–212.

JOURNAL OF HYDROLOGIC ENGINEERING © ASCE / FEBRUARY 2008 / 63

J. Hydrol. Eng. 2008.13:51-63.