Vous êtes sur la page 1sur 14

Remote Sensing of Environment 225 (2019) 16–29

Contents lists available at ScienceDirect

Remote Sensing of Environment


journal homepage: www.elsevier.com/locate/rse

Validation of Collection 6 MODIS land surface temperature product using in T


situ measurements

Si-Bo Duana, Zhao-Liang Lib,a, , Hua Lic, Frank-M. Göttsched, Hua Wue, Wei Zhaof, Pei Lenga,
Xia Zhangb, César Collg
a
Key Laboratory of Agricultural Remote Sensing, Ministry of Agriculture/Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural
Sciences, Beijing 100081, China
b
School of Land Resources and Urban-rural Planning, Hebei GEO University, 050031, Hebei, China
c
State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China
d
Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, 76344 Eggenstein-Leopoldshafen, Germany
e
State Key Laboratory of Resources and Environment Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences,
Beijing 100101, China
f
Institute of Mountain Hazards and Environment, Chinese Academy of Sciences, Chengdu 610041, China
g
Department of Earth Physics and Thermodynamics, Faculty of Physics, University of Valencia, 46100 Burjassot, Spain

A R T I C LE I N FO A B S T R A C T

Keywords: Land surface temperature (LST) is an important physical quantity at the land-atmosphere interface. Since 2016
Land surface temperature the Collection 6 (C6) MODIS LST product is publicly available, which includes three refinements over bare soil
MODIS surfaces compared to the Collection 5 (C5) MODIS LST product. To encourage the use of the C6 MODIS LST
Temperature-based validation method product in a wide range of applications, it is necessary to evaluate the accuracy of the C6 MODIS LST product. In
Split-window algorithm
this study, we validated the C6 MODIS LST product using temperature-based method over various land cover
In situ measurements
types, including grasslands, croplands, cropland/natural vegetation mosaic, open shrublands, woody savannas,
and barren/sparsely vegetated. In situ measurements were collected from various sites under different atmo-
spheric and surface conditions, including seven SURFRAD sites (BND, TBL, DRA, FPK, GCM, PSU, and SXF) in the
United States, three KIT sites (EVO, KAL, and GBB) in Portugal and Namibia, and three HiWATER sites (GBZ,
HZZ, and HMZ) in China. The spatial representativeness of the in situ measurements at each site was separately
evaluated during daytime and nighttime using all available clear-sky ASTER LST products at 90 m spatial re-
solution. Only six sites during daytime are selected as sufficiently homogeneous sites despite the usually high
spatial thermal heterogeneity, whereas during nighttime most sites can be considered to be thermally homo-
geneous and have similar LST and air temperature. The C6 MODIS LST product was validated using in situ
measurements from the selected homogeneous sites during daytime and nighttime: except for the GBB site, large
RMSE values (> 2 K) were obtained during daytime. However, if only satellite LST with a high spatial thermal
homogeneity on the MODIS pixel scale are used for LST validation, the best daytime accuracy (RMSE < 1.3 K)
for the C6 MODIS LST product is achieved over the BND and DRA sites. Except for the DRA site, the RMSE values
during nighttime are < 2 K at the selected homogeneous sites. Furthermore, the accuracy of the C6 MODIS LST
product was compared with that of the C5 MODIS LST product during nighttime at the selected homogeneous
sites. Except for the GBB site, there are only small differences (< 0.4 K) between the RMSE values for the C5 and
C6 MODIS LST products.

1. Introduction et al., 2014, 2017a; Sandholt et al., 2002; Weng, 2009; Sobrino and
Jiménez-Muñoz, 2014; Sobrino et al., 2016). Satellite-derived LST was
Land surface temperature (LST) is an important climate variable widely used in a variety of applications, including land cover and land-
related to surface energy and water balance (Li et al., 2013b). It is also a cover change analysis (Lambin and Ehrlich, 1997), estimation and
key physical parameter for various studies, including hydrology, cli- parameterization of surface fluxes (Lu et al., 2013), climate change
matology, the environment, and ecology (Anderson et al., 2008; Duan studies (Hansen et al., 2010), vegetation monitoring (Kogan, 2001),


Corresponding author at: School of Land Resources and Urban-rural Planning, Hebei GEO University, 050031, Hebei, China.
E-mail address: lizhaoliang@caas.cn (Z.-L. Li).

https://doi.org/10.1016/j.rse.2019.02.020
Received 15 September 2018; Received in revised form 19 February 2019; Accepted 22 February 2019
0034-4257/ © 2019 Elsevier Inc. All rights reserved.
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

urban climate studies (Voogt and Oke, 2003), and drought monitoring Ta + 16 K.
and surface soil moisture estimation (Wan et al., 2004a; Leng et al., To improve the accuracy of the MODIS LST product, three refine-
2014; Zhao et al., 2017a, 2017b). LST has been identified as an im- ments of the C6 MODIS GSW algorithm over bare soil surfaces were
portant Earth Surface Data Record (ESDR) by NASA. Furthermore, LST introduced (Wan, 2014): 1) separate sets of coefficients for retrieving
has been accepted and defined as an Environmental Climate Variable daytime and nighttime LST over bare soil surfaces in the hot and warm
(ECV) by the Global Climate Observing System (GCOS). bare soil zone (latitude range: −38° to 49.5°). The C5 GSW algorithm
The Moderate Resolution Imaging Spectroradiometer (MODIS) only uses one set of coefficients to retrieve daytime and nighttime LST
sensor is one of the key instruments on the Terra and Aqua platforms of for each group of land cover types. 2) Emissivity differences in MODIS
the NASA Earth Observing System. The Aqua MODIS can provide ob- bands 31 and 32 over bare soil surfaces were adjusted. 3) A term
servational overlap and continuity in conjunction with the Advanced containing the quadratic difference between brightness temperatures in
Very High Resolution Radiometer (AVHRR) onboard the NOAA's op- bands 31 and 32 was added to the original GSW algorithm to improve
erational polar-orbiting satellites and the Visible Infrared Imaging LST retrieval accuracy over bare soil surfaces in the hot and warm bare
Radiometer Suite (VIIRS) onboard the Suomi National Polar-Orbiting soil zone. The refined MODIS GSW algorithm is expressed as:
Partnership (S-NPP) satellite. Two LST retrieval algorithms for MODIS
1−ε Δε T + T32
are currently operational: the generalized split-window (GSW) algo- Ts = b0 + ⎛b1 + b2 + b3 2 ⎞ 31
rithm (Becker and Li, 1990; Wan and Dozier, 1996), which generates ⎝ ε ε ⎠ 2
the MOD11A LST product at 1-km resolution, and the physics-based 1−ε Δε T31 − T32
+ ⎛b4 + b5 + b6 2 ⎞ + b7 (T31 − T32 )2
day/night algorithm (Wan and Li, 1997), which generates the MOD11B ⎝ ε ε ⎠ 2 (2)
LST product at 5-km (Collection 4, C4) and 6-km (Collection 5, C5)
Of the three refinements the emissivity adjustment is considered to
resolutions.
be most important. Taking into account that the GSW algorithm is more
The C4 and C5 MODIS LST products were validated using tem-
sensitive to changes in the difference between MODIS emissivities in
perature-based (T-based) and radiance-based (R-based) methods over
bands 31 and 32 than to changes in their mean (Li and Becker, 1993),
various sites, including bare soil, grassland, silt playa, cropland, and
Wan (2014) only adjusted emissivity differences for bare soil pixels and
inland water (Wan et al., 2002, 2004b; Wan and Li, 2008; Wan, 2008,
kept the corresponding emissivity mean constant. The emissivity ad-
2014; Coll et al., 2005, 2009). The results indicated that the accuracy of
justment model was developed by comparing actual brightness tem-
the MODIS LST product was better than 1 K at most sites, with the
perature differences between MODIS bands 31 and 32 with simulated
exception of bare soil sites. The C5 MODIS LST product over arid and
brightness temperature differences, which were estimated from column
semi-arid areas was reported to significantly underestimate LST, which
water vapor and surface air temperature; then the corresponding
can be mainly attributed to the overestimation of surface emissivity
emissivity in band 31 was adjusted with a value between ± 0.0063 and
with the classification-based emissivity method (Hulley and Hook,
the same amount, but with opposite sign, was added to band 32
2009). Li et al. (2014a) and Yu et al. (2014) evaluated the C5 MODIS
emissivity. A detailed description of the emissivity adjustment model
LST product using ground-based measurements in a northern arid area
can be found in Wan (2014).
of China with root mean square error (RMSE) values larger than 3 K. Liu
The MODIS LST product during the period of 2013–2014 was
et al. (2015) validated the C5 MODIS LST product using in situ LST over
downloaded from the EARTHDATA website (https://search.earthdata.
a desert site in Gobabeb, Namibia, with an RMSE value of 3.5 K.
nasa.gov/). The MODIS Reprojection Tool (MRTSwath) was used to
The C6 MOD11 products are publicly available to the user com-
transform the MOD11_L2 and MYD11_L2 product from HDF-EOS swath
munity since 2016. Assessing the accuracy of these LST products will
format to geographic latitude and longitude and the science data set
help encourage their use across a wide range of applications. The main
layers LST, LSE, and quality control (QC) were extracted.
objective of this study is, therefore, to validate the C6 MOD11 products
with in situ measurements over various land cover types, including
grasslands, croplands, cropland/natural vegetation mosaic, open 3. Validation sites
shrublands, woody savannas, and barren/sparsely vegetated. The rest
of this paper is organized as follows: Sections 2 and 3 introduce the 3.1. SURFRAD sites
MODIS GSW algorithms and validation sites, respectively, Section 4
describes the used methodologies, and Sections 5 and 6 present and The Surface Radiation Budget Network (SURFRAD) was established
discuss the validation results. The last section provides some conclu- in 1993 with a primary objective of supporting climate research with
sions. accurate, continuous, long-term measurements of the surface radiation
budget over the United States (Augustine et al., 2000). Seven SURFRAD
sites operating in climatologically diverse regions and representing
2. MODIS GSW algorithms and refinements
various land cover types were selected in this study. Fig. 1 shows
photographs of the seven SURFRAD sites, and Table 1 provides specific
The MODIS level-2 LST product (MOD11_L2 for Terra and
details of the seven sites. The SURFRAD stations provide quality-con-
MYD11_L2 for Aqua) was retrieved with the GSW algorithm from
trolled measurements of broadband hemispherical upwelling and
brightness temperatures in MODIS bands 31 and 32 (Becker and Li,
downwelling longwave radiation along with meteorological parameters
1990; Wan and Dozier, 1996). The GSW algorithm is written as:
every 3 min (before 2009) or every minute (after 2009). Upwelling and
1−ε Δε T + T32 downwelling longwave radiation is measured from 10 m height by two
Ts = b0 + ⎛b1 + b2 + b3 2 ⎞ 31 pyrgeometers (Eppley Precision Infrared Radiometer, spectral range
⎝ ε ε ⎠ 2
1−ε Δε T31 − T32 3–50 μm) with an estimated uncertainty of ± 5 W/m2. The spatial re-
+ ⎛b4 + b5 + b6 2 ⎞ solution of the pyrgeometer measurements is approximately
⎝ ε ε ⎠ 2 (1)
70 × 70 m2 (Guillevic et al., 2014). The SURFRAD measurements are
where Ts is the LST, and ε and Δε are the mean and the difference of the established reference datasets for long-term LST validation (Li et al.,
emissivities in bands 31 and 32, respectively. The regression coeffi- 2014b) and have been used to evaluate various satellite-derived LST
cients bk (k = 0–6) depend on the viewing zenith angle (VZA), surface product, e.g., ASTER (Wang and Liang, 2009), AATSR (Ghent et al.,
air temperature (Ta), and atmospheric column water vapor (CWV). 2017), MODIS (Li et al., 2014b; Wang and Liang, 2009), and VIIRS
These coefficients were derived from regression analysis of radiative (Guillevic et al., 2014; Liu et al., 2015).
transfer simulation data for LST values varying from Ta-16 K to The surface broadband emissivity at the SURFRAD sites was

17
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

BND TBL DRA

FPK GCM PSU

SXF

Fig. 1. The seven SURFRAD sites.


(Source: NOAA Earth System Research Laboratory).

estimated from ASTER emissivity product using a spectral-to-broad- 3.2. KIT sites
band linear regression equation (Cheng et al., 2013):
In situ measurements from three permanent LST validation stations,
εb = 0.197 + 0.025ε10 + 0.057ε11 + 0.237ε12 + 0.333ε13 + 0.146ε14 (3) i.e., EVO (Evora), KAL (Farm Heimat), and GBB (Gobabeb) operated by
the Karlsruhe Institute of Technology (KIT, Germany) were used to
where ε10–ε14 are the narrow-band surface emissivities of the ASTER validate the MODIS LST product. Fig. 2 shows photographs of the three
bands 10–14, respectively. The surface broadband emissivities are KIT sites and Table 1 provides specific details of the three sites.
0.968, 0.972, 0.967, 0.973, 0.971, 0.970, and 0.970 for BND, TBL, DRA, EVO LST validation station is located approximately 12 km south-
FPK, GCM, PSU, and SXF, respectively. west of the town of Evora in the Alentejo region of Portugal. The
dominant vegetation types at the station are isolated groups of ever-
green oak trees and grassland (Kabsch et al., 2008; Trigo et al., 2008).

Table 1
Details of the validation sites used in this study.
No. Site IDa Latitude Longitude Elevation Land cover type

1 BND 40.0516° N 88.3733° W 230 m Croplands


2 TBL 40.1256° N 105.2378° W 1689 m Grasslands
3 DRA 36.6232° N 116.0196° W 1007 m Open shrublands
4 FPK 48.3080° N 105.1018° W 634 m Grasslands
5 GCM 34.2547° N 89.8729° W 98 m Cropland/natural vegetation mosaic
6 PSU 40.7203° N 77.9310° W 376 m Cropland/natural vegetation mosaic
7 SXF 43.7343° N 96.6233° W 473 m Croplands
8 EVO 38.5403° N 8.00328° W 227 m Woody savannas
9 KAL 22.9328° S 17.9922° E 1380 m Open shrublands
10 GBB 23.5510° S 15.0514° E 421 m Barren/sparsely vegetated
11 GBZ 38.9150° N 100.3042° E 1567 m Barren/sparsely vegetated
12 HZZ 38.7652° N 100.3186° E 175 m Barren/sparsely vegetated
13 HMZ 38.7781° N 100.6967° E 1625 m Barren/sparsely vegetated

a
BND: Bondville, TBL: Table Mountain, DRA: Desert Rock, FPK: Fort Peck, GCM: Goodwin Creek, PSU: Pennsylvania State University, SXF: Sioux Falls, EVO:
Evora, KAL: Farm Heimat, GBB: Gobabeb, GBZ: Gobi, HZZ: Huazhaizi, and HMZ: Huangmo.

18
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

EVO KAL GBB

Fig. 2. The three KIT LST validation sites.

desiccated grass. Due to the hyperarid desert climate, the site is spa-
GBZ HZZ HMZ tially and temporally highly stable and, therefore, ideal for the long-
term validation of satellite products (Göttsche et al., 2013, 2016). Long-
term average air temperature at GBB is 21.1 °C, whereas average annual
precipitation is < 100 mm and highly variable. Two KT-15.85 IIP in-
frared radiometers were deployed at the GBB site. One radiometer, with
a field of view (FOV) of 8.5°, is mounted at a height of 25 m and ob-
serves an area of approximately 14 m2. The other radiometer views the
sky at a 53° zenith angle. All station measurements are collected once
per minute. The surface emissivity of the gravel plain is considered
constant (Masiello et al., 2018; Göttsche et al., 2018) and is estimated
to be 0.94 for KT15.85 IIP radiometers (Göttsche and Hulley, 2012).

3.3. HiWATER sites

The Heihe Watershed Allied Telemetry Experimental Research


Fig. 3. The three HiWATER sites used in this study. (HiWATER) is a comprehensive eco-hydrological experiment conducted
in the Heihe River Basin, the second largest inland river basin in the
arid regions of northwest China (37.7°–42.7° N, 97.1°–102.0° E). It was
The climate at the station is warm temperate with hot, dry summers,
designed from an interdisciplinary perspective to address problems that
annual temperature averages between 15 °C and 16 °C, and an average
include heterogeneity, scaling, uncertainty, and closing of the water
annual precipitation of 669 mm. The in situ measurements are collected
cycle at the watershed scale (Li et al., 2013a). We used ground-based
every minute by three KT-15.85 IIP infrared radiometers, observing the
measurements from three sites, i.e., GBZ, HZZ, and HMZ. Fig. 3 shows
sunlit background, a tree crown, and the sky at a 53° zenith angle,
the three HiWATER sites and Table 1 provides some site-specific details.
which is used to estimate the atmospheric downwelling radiance. The
The three sites are covered by bare soil and low desert grass. The
KT-15.85 IIP measures thermal infrared radiance in the 9.6–11.5 μm
GBZ site is equipped with a Kipp & Zonen CNR1 net radiometer, which
domain and obtains brightness temperatures with an absolute accuracy
measures upwelling and downwelling hemispherical broad-band ra-
of ± 0.3 K (Theocharous et al., 2010). Surface emissivity was estimated
diances (5–50 μm) from a height of 6 m. The HZZ and HMZ sites are
using fixed end-member fractions (tree = 32%, grass/ground = 68%),
each equipped with two Apogee SI-111 infrared radiometers (8–14 μm;
i.e., the temporal dependence of the cover fractions was ignored
44° full view angle). One radiometer each observes the surface at nadir
(Ermida et al., 2014). The KT-15.85 IIP emissivity was set to a static
from a height of 4 m with a footprint of 8 m2 at the HMZ site and one
value of 0.974, which is a typical value for vegetation and close to the
from a height of 2.65 m with a footprint of 3.6 m2 at the HZZ site. The
corresponding LSA SAF emissivity for SEVIRI channel 9 at the EVO site.
other radiometer at each site views the sky at an effective angle of
KAL LST validation station is located approximately 100 km south-
approximately 55° from zenith to measure atmospheric downwelling
east from Windhuk on a plateau of the Kalahari semi-desert in Namibia.
radiance. The SI-111 and CNR1 measurements are collected once per
The station has a size of approximately 50 km2. It is characterized by
minute. Emissivity of 0.948 for the Kipp & Zonen CNR1 instrument at
hot and arid climate, which exhibits a natural seasonality. The station is
GBZ followed the approach described in Section 3.1 for the SURFRAD
located in a typical Kalahari land scape and a wide area around the
sites via Eq. (3). Surface emissivities at the HZZ and HMZ sites were
mast is mainly covered by patchy, desiccated grass dotted with bushes
determined using the vegetation cover method. The end-member
and isolated camel thorn trees. The in situ measurements are collected
emissivities of bare soil and desert grass were measured using an ABB
every minute by three KT-15.85 IIP infrared radiometers, observing
BOMEM MR304 spectroradiometer with a spectral resolution of
grass/bare ground, bush/tree crown, and the sky at a 53° zenith angle.
1 cm−1. Nine and five emissivity samples were measured for bare soil
The measured component radiances are mixed according to their re-
and desert grass, respectively. The emissivity spectra were retrieved
spective cover fractions, which were determined from high-resolution
using the Iterative Spectrally Smooth Temperature and Emissivity Se-
satellite imagery. The KT-15.85 IIP's emissivities of the two observed
paration (ISSTES) algorithm (Ingram and Muse, 2001). The Fractional
end-members are set to the value retrieved operationally by LSA SAF
Vegetation Cover (FVC) of the sites was obtained with a photographic
for MSG/SEVIRI channel 9: over the course of the year this value varies
method at nadir view. The SI-111 channel emissivity was obtained by
little between about 0.973–0.984, which is in good agreement with
convolving the emissivity spectra with the spectral response function of
literature values for vegetation (Göttsche et al., 2016).
the SI-111 infrared radiometers: for the HZZ and HMZ sites an emis-
GBB LST validation station is located on the large gravel plains
sivity value of 0.97 was obtained (Li et al., 2014a).
(several thousand km2) of the Namib Desert in Namibia, which are
covered by a highly homogeneous mixture of gravel, sand, and sparse

19
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

4. Methodologies further than three standard deviations from the mean is regarded as an
outlier (Pearson, 2002). The probability that a point is wrongly re-
4.1. In situ LST estimation moved as an outlier is approximately 0.3%. However, the “3σ-edit rule”
usually fails in practice because outliers lead to biased estimates of the
For the KT-15.85 IIP infrared radiometers at the KIT sites and the SI- mean and standard deviation. A robust method for outlier detection is
111 infrared radiometers at the HiWATER sites, in situ LST was esti- the “3σ-Hampel identifer” (Davies and Gather, 1993). In this method,
mated from directional surface-leaving radiances and downwelling ‘sky’ the mean is replaced by the median, whereas the standard deviation is
radiances: estimated as:
Rg, λ − (1 − ελ ) Rd, λ ⎤ S = 1.4826 × median {|x i − x m|} (6)
LSTg = Bλ−1 ⎡

⎣ ελ ⎥
⎦ (4) where xm is the median of the data sequence {xi}, and S is the standard
deviation of the data sequence {xi}. The constant 1.4826 is chosen to
where LSTg is the in situ LST (K), B is the Planck function, Rg,λ is the
obtain an unbiased estimate of the standard deviation for Gaussian data
radiance measured by the surface-observing radiometer (W/m2/sr/μm),
(Pearson, 2002).
ελ is the spectral emissivity at wavelength λ, and Rd,λ is the down-
In this study, the data sequence {xi} is formed by the differences
welling radiance (W/m2/sr/μm); for a radiometer that views the sky at
between the MODIS and in situ LST. The data points with LST differ-
the ‘representative’ zenith angle of 53°, the measured Rd,λ is equivalent
ences less than xm-3S or larger than xm + 3S are regarded as outliers.
to the hemispherical downwelling irradiance divided by π (Kondratyev,
The “3σ-Hampel identifer” only removes a relatively small fraction of
1969; Garcia-Santos et al., 2013).
outliers (mainly undetected clouds), typically < 10%.
For the pyrgeometers at the SURFRAD sites and the CNR1 net
radiometers at the HiWATER sites, in situ LST were estimated from the
measured upwelling and downwelling broadband hemispherical ra- 5. Results and analyses
diances using Stefan-Boltzmann's law:
5.1. Evaluation of spatial representativeness of ground-based LST
Rg − (1 − εb ) Rd ⎤1/4 observations
LSTg = ⎡

⎣ σεb ⎥
⎦ (5)
In order to validate satellite LST products accurately, it has to be
where LSTg is the in situ LST (K), Rg is the upwelling broadband ensured that the ground-based (point) measurements are representative
hemispherical radiance (W/m2), Rd is the downwelling broadband on the satellite pixel scale (Guillevic et al., 2018). The easiest way to
hemispherical radiance (W/m2), σ is the Stefan-Boltzmann constant ensure this is to perform the LST validation over a large and relatively
(5.67 × 10−8 W/m2/K4), and εb is the surface broadband emissivity. homogeneous area (Göttsche et al., 2016). Prior to the validation of the
MODIS LST product, the ASTER LST product at a spatial resolution of
4.2. Temperature-based validation method 90 m was used to evaluate the spatial homogeneity of the validation
sites. A subset of 11 × 11 ASTER pixels corresponding to the MODIS
Three methods were widely used to validate LST products derived pixel centered on each station was used to calculate the spatial standard
from satellite measurements, i.e., temperature-based validation method deviation (STD) of LST; all available clear-sky ASTER granules down-
(Wan et al., 2002, 2004b; Coll et al., 2009; Göttsche et al., 2016; Xu loaded from the EARTHDATA website were included in the statistics.
et al., 2014; Ouyang et al., 2017; Martin and Göttsche, 2016), radiance- Clear-sky ASTER granules during daytime were identified by visual
based validation method (Wan and Li, 2008; Wan, 2008, 2014; Coll inspection of ASTER RGB false-color images while at nighttime the
et al., 2009; Niclòs et al., 2011; Hulley and Hook, 2012; Gomis-Cebolla MODIS cloud mask product was used.
et al., 2018; Duan et al., 2018), and inter-comparison method (Guillevic Fig. 4 shows boxplots of the STD of 11 × 11 ASTER LST subsets
et al., 2014; Frey et al., 2012; Trigo et al., 2008; Ermida et al., 2014; during daytime (a) and nighttime (b): the STDs during daytime are
Duan and Li, 2015, Duan et al., 2017b). The three methods are com- considerably larger than at night. The results show that LST exhibits
plementary and provide different levels of information about the ac- higher spatial variability during daytime than at night; here, sites with
curacy of the satellite-derived LST products. A comprehensive review of a spatial STD < 1.5 K and a median STD < 1 K are considered homo-
LST validation methods and best practices are provided by Guillevic geneous. Therefore, sites S3 (DRA), K2 (KAL), K3 (GBB), H1 (GBZ), H2
et al. (2018). (HZZ), and H3 (HMZ) were selected as homogeneous ‘daytime’ sites,
The temperature-based validation method involves a direct com- whereas all sites but S2 (TBL) were selected as homogeneous ‘night-
parison of ground-based LST observations and satellite-derived LST time’ sites.
product (Coll et al., 2005; Hook et al. 2007; Wang et al., 2008; Wang
and Liang, 2009; Guillevic et al., 2012, 2014; Göttsche et al., 2013; 5.2. Validation of daytime MODIS LST
Ermida et al., 2014; Li et al., 2014a; Krishnan et al., 2015). It allows the
determination of the accuracy in satellite-derived LST product. In situ measurements at the DRA, KAL, GBB, GBZ, HZZ, and HMZ
LST values of MOD11_L2 and MYD11_L2 products were extracted sites were used to validate daytime MODIS LST. Fig. 5 shows C5 and C6
for the pixel closest to each site based on longitude and latitude: to MODIS LST versus in situ LST for the six sites; although these are re-
minimize the effect of cloud contamination on the validation results, latively homogeneous at the ASTER pixel scale (see Fig. 4), RMSE larger
only high-quality data (i.e., QC = 0) were evaluated. Furthermore, only than 2 K is found for both LST products over all sites except for C6
in situ LST with the quality flag corresponding to good data were used. MODIS LST over the GBB site. The results indicate that the temperature-
MODIS and in situ LST were temporally matched with each other to based method is challenging to use for validating daytime LST and re-
better than 1 min. quires in situ LST that are representative of the satellite pixel, such as
the GBB site. For the GBB site, the RMSE for C6 (1.7 K) is approximately
4.3. Robust outlier removal two times less than that for C5 (3.8 K). The improved results are due to
the adjusted emissivities of the C6 MODIS GSW algorithm over bare soil
To obtain robust statistics of LST validation, it is necessary to re- surfaces. However, when comparing these results to the literature
move outliers due to cloud contamination (Göttsche et al., 2013). A (Göttsche et al., 2016; Ermida et al., 2014, 2017; Guillevic et al., 2013,
popular method for outlier detection is the “3σ-edit rule”. It assumes 2014) and bearing in mind the high degree of homogeneity of the
that a data sequence is approximately normally distributed and a point ‘daytime’ sites (see Fig. 4), the negative biases are likely due to viewing

20
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 4. Spatial STD of 11 × 11 ASTER LST subsets corresponding to the MODIS pixel centered on each site for (a) daytime and (b) nighttime. S1: BND, S2: TBL, S3:
DRA, S4: FPK, S5: GCM, S6: PSU, S7: SXF, K1: EVO, K2: KAL, K3: GBB, H1: GBZ, H2: HZZ, and H3: HMZ. There are no clear-sky ASTER granules at SXF during
nighttime. In the boxplots, the upper and lower whiskers mark the maximum and minimum, respectively, while the lines in the middle of the boxes indicate the
median, and the bottom and top of a box show the first and third quartiles. Plus signs are outliers.

Fig. 5. Scatterplots of C5 and C6 MODIS LST versus in situ LST during daytime at (a) DRA, (b) KAL, (c) GBB, (d) GBZ, (e) HZZ, and (f) HMZ.

21
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 6. C6 MODIS LST versus in situ LST during daytime at (a) BND and (b) DRA. Only data points are shown for which the STD of the 11 × 11 ASTER LST subsets
centered on the MODIS pixel is < 1 K.

Fig. 7. C5 and C6 MODIS LST versus in situ LST during nighttime at (a) BND, (b) DRA, (c) FPK, (d) GCM, (e) PSU, (f) SXF, (g) EVO, (h) KAL, (i) GBB, (j) GBZ, (k) HZZ,
and (l) HMZ.

22
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 7. (continued)

Table 2 the usually warmer soil background (Ermida et al., 2014).


Bias and RMSE of the difference between C6 Terra/Aqua-MODIS LST and in situ It should be noted that whether in situ LST at selected sites is spa-
LST during nighttime. tially representative at satellite pixel scale also depends on satellite
Site ID Terra-MODIS Aqua-MODIS overpass time and date, e.g., due to seasonal vegetation changes and
agricultural activities. The spatial representativeness of in situ mea-
Num Bias (K) RMSE (K) Num Bias (K) RMSE (K) surements for the coarse satellite pixel scale can be evaluated with a
fine LST image acquired at the same observation time. In this study,
BND 67 −0.94 1.37 58 −1.04 1.70
DRA 286 −3.45 3.53 290 −3.74 3.85 only in situ measurements were used to further evaluate the accuracy of
FPK 113 −1.14 1.61 155 −0.54 1.27 the MODIS LST product, for which the STD of the 11 × 11 simultaneous
GCM 132 1.36 2.00 97 1.40 1.88 ASTER LST centered on the MODIS pixel was < 1 K. Only at BND and
PSU 63 0.48 1.66 66 0.51 1.61 DRA the number of data points is > 5. Fig. 6 shows C6 MODIS LST
SXF 131 −0.26 1.53 25 −0.95 1.48
EVO 119 0.35 1.40 85 0.73 1.43
versus in situ LST during daytime at BND and DRA. For both sites, the
KAL 178 −0.11 0.76 84 −0.01 0.74 C6 MODIS LST product is in good agreement with in situ LST, with an
GBB 224 −0.92 1.49 109 −0.43 1.34 absolute bias < 0.2 K and an RMSE < 1.3 K. The results indicate that
GBZ 217 −1.45 1.75 224 −1.62 1.86 when spatial thermal homogeneity is high on the ASTER pixel scale
HZZ 172 −0.95 1.33 166 −0.75 1.17
(about 100 m), the in situ measurements obtained during daytime are
HMZ 204 −1.52 1.78 205 −1.00 1.24
also representative of MODIS LST.

geometry differences between satellite and in-situ sensors, which at 5.3. Validation of nighttime MODIS LST
daytime over sparse vegetation canopies cause satellites to view a larger
fraction of vegetation (at near air temperature) and a smaller fraction of In situ measurements at all sites except for the TBL site were used to

23
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 8. Monthly mean bias (MODIS LST – in situ LST) during nighttime for the years 2013 and 2014 at (a) BND, (b) DRA, (c) FPK, (d) GCM, (e) PSU, (f) SXF, (g) EVO,
(h) KAL, (i) GBB, (j) GBZ, (k) HZZ, and (l) HMZ.

validate the MODIS LST product during nighttime. Fig. 7 shows scat- summarized in Table 2. Only small differences (< 0.5 K) between the
terplots of C5 and C6 MODIS LST versus in situ LST during nighttime for RMSE for C6 Terra-MODIS and Aqua-MODIS LST are observed over the
the twelve sites. Except for the GBB site, there are only small differences investigated sites.
(< 0.4 K) between the RMSE for C5 and C6. For the GBB site, the RMSE To analyze the bias variability at the twelve sites, the monthly mean
for C6 (1.5 K) is significantly less than that for C5 (3.6 K). The accuracy bias (MODIS LST – in situ LST) during nighttime for the years 2013 and
of the C5/C6 MODIS LST product during nighttime is better than 2 K at 2014 are shown in Fig. 8. Except for the DRA, GBB, GBZ, HZZ, and HMZ
all sites with the exception of the C5 product at GBB and the C5/C6 sites, there are no significant discrepancy between the monthly mean
product at DRA. Furthermore, there is an obvious overestimation of the bias for C5 and C6 during nighttime. The bias for most months at FPK,
C5/C6 MODIS LST product at GCM, which is covered by a hetero- PSU, SXF, EVO, and KAL is between −1 K and 1 K. As shown in Fig. 8,
geneous mixture of grassland and woods. When looking at the photo of the overall absolute bias at the five sites is < 1 K. Smaller absolute bias
the GCM site (Fig. 1), “a possible explanation could be that the mixture can be found during summer months at BND and GCM, which could be
of grassland and woods observed by the satellite stays warmer during attributed to the changes in vegetation phenology. The bias decreases to
nighttime than the ‘pure’ grass observed by the in situ sensor, because a minimum during the summer as the vegetation fully mature in July
tree crown temperature is usually closer to air temperature while open and August, resulting in more homogeneous temperature distribution
(short grass) areas experience stronger radiative cooling” (Martin and around the two sites. Larger differences between the monthly mean bias
Göttsche, 2016). for C5 and C6 during nighttime are observed at DRA, GBB, GBZ, HZZ,
We further compare the accuracy of the C6 Terra-MODIS and Aqua- and HMZ.
MODIS LST product. The bias and RMSE of the difference between C6 To further analyze the factors resulting in the discrepancy between
Terra/Aqua-MODIS LST and in situ LST during nighttime are the monthly mean bias for C5 and C6 during nighttime at DRA, GBB,

24
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 8. (continued)

GBZ, HZZ, and HMZ, we compared the emissivity values in MODIS GBB site, the C6 emissivity values in MODIS bands 31 and 32 are closer
bands 31 and 32 and their mean and difference for the ASTER GED and to those in the ASTER GED product. Both the emissivity mean values for
C5 and C6 MODIS emissivity products at the five sites. The comparison C5 and C6 are nearly equal to those for the ASTER GED product,
results are shown in Fig. 9. The ASTER GED product was converted into whereas the emissivity difference value for C6 is more close to that for
emissivity values corresponding to MODIS bands 31 and 32 in terms of the ASTER GED product. Wan (2014) reported that the GSW algorithm
MODIS and ASTER spectral response function and emissivity spectra is more sensitive to the change in the emissivity difference than the
from the ASTER spectral library. The surface types of the five sites are change in the emissivity mean. Therefore, the accuracy of the C6
barren/sparsely vegetated or open shrublands. The ASTER emissivity MODIS LST product is better than that of the C5 MODIS LST product at
product has higher retrieval accuracy over barren/sparsely vegetated or GBB. For the GBZ, HZZ, and HMZ sites, the C5 emissivity values in
open shrublands (Hulley et al., 2015). Therefore, the ASTER GED MODIS bands 31 and 32 and their mean are closer to those in the
product is used as reference values to evaluate the performance of the ASTER GED product. Furthermore, the emissivity difference is nearly
C5 and C6 MODIS emissivity products. equal for C5 and C6 at these three sites. Therefore, at GBZ, HZZ, and
By comparing the C5 and C6 MODIS and ASTER GED emissivity HMZ the accuracy of the C5 MODIS LST product is better than that of
products, the five sites can be divided into three groups: the first for the C6 MODIS LST product. The results can mainly be contributed to
DRA, the second for GBB, and the third for GBZ, HZZ, and HMZ. For the the misclassification of surface types as grasslands in the C5.1 MODIS
DRA site, the discrepancies between the ASTER GED product and the C5 land cover type product at these three sites, which is used to estimate
and C6 MODIS emissivity products are small in terms of emissivity the C6 MODIS emissivity in the GSW algorithm.
values in MODIS bands 31 and 32 and their mean and difference. The We further investigate the reasons for large discrepancies between
results indicate that surface emissivity is not the main factor resulting in the C5 and C6 MODIS LST and in situ LST at DRA. Fig. 10 shows the
the discrepancy between the C5 and C6 MODIS LST at DRA. For the scatterplot of the C6 MODIS LST versus in situ LST during nighttime. It

25
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Fig. 9. Comparison of emissivity values in MODIS bands 31 and 32 and their mean and difference for the ASTER GED and C5 and C6 MODIS emissivity products at
DRA, GBB, GBZ, HZZ, and HMZ. (a) Emissivity values in MODIS band 31, (b) emissivity values in MODIS band 32, (c) emissivity mean, and (d) emissivity difference
between MODIS bands 31 and 32. The ASTER GED product was converted into emissivity values corresponding to MODIS bands 31 and 32 by means of MODIS and
ASTER spectral response function and emissivity spectra from the ASTER spectral library.

Fig. 11. C6 MODIS LST versus in situ LST at the sites of Wan et al. (2002,
Fig. 10. C6 MODIS LST versus in situ LST during nighttime at DRA. The data 2004a, 2004b) and Coll et al. (2016).
points represent the cases for which the STD of the 11 × 11 ASTER LST subsets
centered on the MODIS pixel is < 1 K.
6. Discussion

is obvious that a large systematic bias (−3.44 K) between the C6 6.1. Issues of the MODIS LST product
MODIS LST and in situ LST can be found. Similar systematic biases
(−3.85 K for AATSR LST and −2.05 K for VIIRS LST) were also re- According to the results and analyses above, the largest uncertainty
ported in Ghent et al. (2017) and Guillevic et al. (2014) when compared in the MODIS GSW algorithm is the determination of surface emissivity.
AATSR and VIIRS LST with in situ LST during nighttime at DRA. The The MODIS GSW algorithm uses the classification-based emissivity
results may be due to the hemispherical in situ sensor observing more method. The key point of this method is to assign surface emissivities
bushes (which is at air temperature; see Fig. 1) than the directional based on land cover classification (Snyder et al., 1998; Li et al., 2013c).
satellite sensor. The smaller bias at daytime may be due to an acci- Two main aspects of the classification-based emissivity method cause
dentally representative mixture of sunlit soil, shadow, and bush in the uncertainty in surface emissivity: (1) the accuracy of the emissivity
FOV of the in situ sensor. estimation depends on the accuracy of the land cover type product.
Misclassifications in the land cover type product would lead to emis-
sivity errors. (2) Emissivities used in the MODIS GSW algorithm are
derived from fixed values for a limited number of land cover types,

26
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

which do not fully represent the natural variation in emissivity, espe- spatially considerably more homogeneous LST.
cially over bare soil surfaces. The incorporation of a dynamic emissivity (4) Except for bare soil surfaces, the RMSE difference between C5 and
product, e.g., as generated by the temperature and emissivity separa- C6 MODIS LST is < 0.4 K.
tion (TES) algorithm (Hulley and Hook, 2011; Malakar and Hulley, (5) The emissivity adjustments of the C6 MODIS GSW algorithm suc-
2016; Islam et al., 2017), into the GSW algorithm would be one way to cessfully reduced C6 MODIS LST errors over some bare soil sur-
improve the accuracy of the MODIS LST product. faces, e.g. the GBB site.
(6) The largest uncertainty in the MODIS GSW algorithm is the de-
6.2. Challenges of temperature-based validation termination of surface emissivity. The classification-based emis-
sivity method is not able to reliably characterize the spectral
As noted by Yu et al. (2012), there are many challenges in the emissivity variation over bare soil surfaces. One way to improve the
temperature-based validation of satellite-derived LST product. Because GSW algorithm would be to incorporate dynamic LSE retrieved
of large spatial variations in LST, in situ LST measurements do not with a physics-based algorithm, e.g., the TES algorithm.
necessarily represent LST measurements at the satellite pixel scale, (7) Over all surfaces but bare soil the performance of the C6 MODIS
especially during the day. At night, air temperature and LST are similar GSW algorithm is nearly the same as that of the C5 algorithm.
to each other and the LST is more spatially homogeneous; therefore, in Therefore, we conclude that users do not need to switch to the C6
situ LST observations at night are more representative of LST at the MODIS LST product and can continue to use the C5 MODIS LST
satellite pixel scale. Consequently, Wang et al. (2008) validated MODIS product. However, for studies over bare soil surfaces (e.g., deserts),
LST product using nighttime data only. users are recommended to use the C6 MODIS LST product, which
To obtain spatially representative and high-quality in situ LST, outperforms the corresponding C5 product over most desert sites
previous studies conducted field campaigns only over large homo- (Duan et al., 2017b, 2018; Wan, 2014).
geneous sites, such as lakes, snow, grasslands, silt playas, and cropland
fields (Wan et al., 2002, 2004a, 2004b; Coll et al., 2005, 2009). Fig. 11 Acknowledgments
shows a scatterplot between the C6 MODIS LST and in situ LST com-
piled for the sites of Wan et al. (2002, 2004a, 2004b) and Coll et al. This work was supported by the National Key R&D Program of
(2016). The RMSE of the LST differences is < 1 K, which indicates that China, 2018YFB0504800 (2018YFB0504804), by the National Natural
for well-chosen and dedicated sites, temperate-based validation is a Science Foundation of China under Grant 41871275, and by the Youth
suitable means for corroborating satellite-derived LST product. How- Innovation Promotion Association CAS, China under Grant 2016333.
ever, quantitative assessments of satellite-derived LST product require
dedicated and high-quality in situ LST observations over sites that are References
homogeneous at the satellite pixel scale. Therefore, it is necessary to
ensure spatial homogeneity of the land cover on the scale of the field Anderson, M.C., Norman, J.M., Kustas, W.P., Houborg, R., Starks, P.J., Agam, N., 2008. A
radiometer as well as over the satellite pixel (Göttsche et al., 2013). thermal-based remote sensing technique for routine mapping of land-surface carbon,
water and energy fluxes from field to regional scales. Remote Sens. Environ. 112,
TIR field radiometers usually collect ground-based measurements 4227–4241.
near nadir at wider view angles, whereas satellite sensors with wide Augustine, J.A., DeLuisi, J.J., Long, C.N., 2000. SURFRAD—A national surface radiation
swaths, e.g., MODIS, AVHRR, and VIIRS, may observe a specific site budget network for atmospheric research. Bull. Am. Meteorol. Soc. 81, 2341–2357.
Becker, F., Li, Z.-L., 1990. Towards a local split window method over land surfaces. Int. J.
considerably off-nadir at narrow view angles. Such observation differ- Remote Sens. 11, 369–393.
ences can cause discrepancies between in situ LST and satellite-derived Cheng, J., Liang, S., Yao, Y., Zhang, X., 2013. Estimating the optimal broadband emis-
LST. Furthermore, a larger VZA results in a longer atmospheric path sivity spectral range for calculating surface longwave net radiation. IEEE Geosci.
Remote Sens. Lett. 10, 401–405.
length, which causes a significant decrease in atmospheric transmit- Coll, C., Caselles, V., Galve, J.M., Valor, E., Niclòs, R., Sánchez, J.M., Rivas, R., 2005.
tance and may introduce significant retrieval errors in the GSW algo- Ground measurements for the validation of land surface temperatures derived from
rithm, particularly under high LST conditions. Additionally, a larger AATSR and MODIS data. Remote Sens. Environ. 97, 288–300.
Coll, C., Wan, Z., Galve, J.M., 2009. Temperature-based and radiance-based validations of
VZA leads to larger differences in the target areas actually observed by
the V5 MODIS land surface temperature product. J. Geophys. Res. 114, D20102.
TIR field radiometers and satellite sensors, especially over spatially Coll, C., García-Santos, V., Niclòs, R., Caselles, V., 2016. Test of the MODIS land surface
heterogeneous validation sites. temperature and emissivity separation algorithm with ground measurements over a
rice paddy. IEEE Trans. Geosci. Remote Sens. 54, 3061–3069.
Davies, L., Gather, U., 1993. The identification of multiple outliers. J. Am. Stat. Assoc. 88,
7. Conclusions 782–792.
Duan, S.-B., Li, Z.-L., 2015. Intercomparison of operational land surface temperature
The C6 MODIS LST product was validated using in situ measure- products derived from MSG-SEVIRI and Terra Aqua-MODIS data. IEEE J. Sel. Top.
Appl. Earth Obs. Remote Sens. 8, 4163–4170.
ments collected from various sites under different atmospheric and Duan, S.-B., Li, Z.-L., Tang, B.-H., Wu, H., Tang, R., 2014. Generation of a time-consistent
surface conditions, including seven SURFRAD sites in the United States, land surface temperature product from MODIS data. Remote Sens. Environ. 140,
three KIT sites in Portugal and Namibia, and three HiWATER sites in 339–349.
Duan, S.-B., Li, Z.-L., Leng, P., 2017a. A framework for the retrieval of all-weather land
China. We draw the following conclusions: surface temperature at a high spatial resolution from polar-orbiting thermal infrared
and passive microwave data. Remote Sens. Environ. 195, 107–117.
(1) It is necessary to evaluate the spatial representativeness of in situ Duan, S.-B., Li, Z.-L., Cheng, J., Leng, P., 2017b. Cross-satellite comparison of operational
land surface temperature products derived from MODIS and ASTER data over bare
LST measurements prior to the validation of satellite-derived LST
soil surfaces. ISPRS J. Photogramm. Remote Sens. 126, 1–10.
product. Therefore, it is considerably easier to perform tempera- Duan, S.-B., Li, Z.-L., Wu, H., Leng, P., Gao, M., Wang, C., 2018. Radiance-based vali-
ture-based validation over large and relatively homogeneous sites. dation of land surface temperature products derived from Collection 6 MODIS
thermal infrared data. Int. J. Appl. Earth Obs. Geoinf. 70, 84–92.
(2) In situ LST measurements during daytime in most cases cannot be
Ermida, S.L., Trigo, I.F., DaCamara, C.C., Göttsche, F.-M., Olesen, F.-S., Hulley, G., 2014.
used to validate satellite-derived LST product due to high spatial Validation of remotely sensed surface temperature over an oak woodland land-
thermal heterogeneity. However, if the site has high spatial thermal scape—the problem of viewing and illumination geometries. Remote Sens. Environ.
homogeneity corresponding to the crossing time and the pixel scale 148, 16–27.
Ermida, S.L., Dacamara, C.C., Trigo, I.F., Pires, A.C., Ghent, D., Remedios, J., 2017.
of the satellite being validated, in situ measurements during day- Modelling directional effects on remotely sensed land surface temperature. Remote
time can also be used for LST validation. Sens. Environ. 190, 56–69.
(3) In situ LST measurements during nighttime are the most reliable for Frey, C.M., Kuenzer, C., Dech, S., 2012. Quantitative comparison of the operational
NOAA-AVHRR LST product of DLR and the MODIS LST product. Int. J. Remote Sens.
validating satellite-derived LST product, which arises from similar 32, 7165–7183.
LST and air temperature (i.e., absence of surface overheating) and

27
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

Garcia-Santos, V., Valor, E., Caselles, V., Mira, M., Galve, J.M., Coll, C., 2013. Evaluation Experimental Design. Bulletin of the. vol. 94. American Meteorological Society, pp.
of different methods to retrieve the hemispherical downwelling irradiance in the 1145–1160.
thermal infrared region for field measurements. IEEE Trans. Geosci. Remote Sens. 51, Li, Z.-L., Tang, B.-H., Wu, H., Ren, H., Yan, G., Wan, Z., Trigo, I.F., Sobrino, J.A., 2013b.
2155–2165. Satellite-derived land surface temperature: current status and perspectives. Remote
Ghent, D.J., Corlett, G.K., Göttsche, F.-M., Remedios, J.J., 2017. Global land surface Sens. Environ. 131, 14–37.
temperature from the along-track scanning radiometers. J. Geophys. Res. Atmos. 122. Li, Z.-L., Wu, H., Wang, N., Qiu, S., Sobrino, J.A., Wan, Z., Tang, B.-H., Yan, G., 2013c.
https://doi.org/10.1002/2017JD027161. Land surface emissivity retrieval from satellite data. Int. J. Remote Sens. 34,
Gomis-Cebolla, J., Jiménez-Muñoz, J.C., Sobrino, J.A., 2018. LST retrieval algorithm 3084–3127.
adapted to the Amazon evergreen forests using MODIS data. Remote Sens. Environ. Li, H., Sun, D., Yu, Y., Wang, H., Liu, Y., Liu, Q., Du, Y., Wang, H., Cao, B., 2014a.
204, 401–411. Evaluation of the VIIRS and MODIS LST product in an arid area of Northwest China.
Göttsche, F.-M., Hulley, G.C., 2012. Validation of six satellite-retrieved land surface Remote Sens. Environ. 142, 111–121.
emissivity product over two land cover types in a hyper-arid region. Remote Sens. Li, S., Yu, Y., Sun, D., Tarpley, D., Zhan, X., Chiu, L., 2014b. Evaluation of 10 year AQUA/
Environ. 124, 149–158. MODIS land surface temperature with SURFRAD observations. Int. J. Remote Sens.
Göttsche, F.-M., Olesen, F.-S., Bork-Unkelbach, A., 2013. Validation of land surface 35, 830–856.
temperature derived from MSG/SEVIRI with in-situ measurements at Gobabeb, Liu, Y., Yu, Y., Yu, P., Göttsche, F.-M., Trigo, I.F., 2015. Quality assessment of S-NPP
Namibia. Int. J. Remote Sens. 34, 3069–3083. VIIRS land surface temperature product. Remote Sens. 7, 12215–12241.
Göttsche, F.-M., Olesen, F.-S., Trigo, I., Bork-Unkelbach, A., Martin, M.A., 2016. Long Lu, J., Tang, R., Tang, H., Li, Z.-L., 2013. Derivation of daily evaporative fraction based on
term validation of land surface temperature retrieved from MSG/SEVIRI with con- temporal variations in surface temperature, air temperature, and net radiation.
tinuous in-situ measurements in Africa. Remote Sens. 8, 410. Remote Sens. 5, 5369–5396.
Göttsche, F.-M., Olesen, F., Poutier, L., Langlois, S., Wimmer, W., Santos, V.G., Coll, C., Malakar, N.K., Hulley, G.C., 2016. A water vapor scaling model for improved land surface
Niclos, R., Arbelo, M., Monchau, J.-P., 2018. Report from the field inter-comparison temperature and emissivity separation of MODIS thermal infrared data. Remote Sens.
experiment (FICE) for land surface temperature. http://www.frm4sts.org/project- Environ. 182, 252–264.
documents/, Accessed date: 13 September 2018. Martin, M., Göttsche, F.-M., 2016. Satellite LST validation report. http://www.
Guillevic, P.C., Privette, J.L., Coudert, B., Palecki, M.A., Demarty, J., Ottle, C., Augustine, globtemperature.info/index.php/public-documentation/deliverables-1, Accessed
J.A., 2012. Land surface temperature product validation using NOAA's surface cli- date: 13 September 2018.
mate observation networks–scaling methodology for the Visible Infrared Imager Masiello, G., Serio, C., Venafra, S., Liuzzi, G., Poutier, L., Göttsche, F.-M., 2018. Physical
Radiometer Suite (VIIRS). Remote Sens. Environ. 124, 282–298. retrieval of land surface emissivity spectra from hyper-spectral infrared observations
Guillevic, P.C., Bork-Unkelbach, A., Göttsche, F.-M., Hulley, G., Gastellu-Etchegorry, J.- and validation with in situ measurements. Remote Sens. 10, 976.
P., Olesen, F.S., Privette, J.L., 2013. Directional viewing effects on satellite land Niclòs, R., Galve, J.M., Valiente, J.A., Estrela, M.J., Coll, C., 2011. Accuracy assessment of
surface temperature products over sparse vegetation canopies—a multisensory ana- land surface temperature retrievals from MSG2-SEVIRI data. Remote Sens. Environ.
lysis. IEEE Geosci. Remote Sens. Lett. 10, 1464–1468. 115, 2126–2140.
Guillevic, P.C., Biard, J.C., Hulley, G.C., Privette, J.L., Hook, S.J., Olioso, A., Göttsche, F.- OuYang, X., Chen, D., Duan, S.-B., Lei, Y., Dou, Y., Hu, G., 2017. Validation and analysis
M., Radocinski, R., Román, M.O., Yu, Y., Csiszar, I., 2014. Validation of land surface of long-term AATSR land surface temperature product in the Heihe River Basin,
temperature product derived from the Visible Infrared Imaging Radiometers Suite China. Remote Sens. 9, 152.
(VIIRS) using ground-based and heritage satellite measurements. Remote Sens. Pearson, R.K., 2002. Outliers in process modeling and identification. IEEE Trans. Control
Environ. 154, 19–37. Syst. Technol. 10, 55–63.
Guillevic, P., Göttsche, F., Nickeson, J., Hulley, G., Ghent, D., Yu, Y., Trigo, I., Hook, S., Sandholt, I., Rasmussen, K., Andersen, J., 2002. A simple interpretation of the surface
Sobrino, J.A., Remedios, J., Román, M., Camacho, F., 2018. Land surface temperature temperature/vegetation index space for assessment of surface moisture status.
product validation best practice protocol. https://lpvs.gsfc.nasa.gov/PDF/CEOS_LST_ Remote Sens. Environ. 79, 213–224.
PROTOCOL_Feb2018_v1.1.0_light.pdf, Accessed date: 13 September 2018. Snyder, W.C., Wan, Z., Zhang, Y., Feng, Y.-Z., 1998. Classification-based emissivity for
Hansen, J., Ruedy, R., Sato, M., Lo, K., 2010. Global surface temperature change. Rev. land surface temperature measurement from space. Int. J. Remote Sens. 19,
Geophys. 48, RG4004. 2753–2774.
Hook, S.J., Vaughan, R.G., Tonooka, H., Schladow, S.G., 2007. Absolute radiometric in- Sobrino, J.A., Jiménez-Muñoz, J.C., 2014. Minimum configuration of thermal infrared
flight validation of mid infrared and thermal infrared data from ASTER and MODIS bands for land surface temperature and emissivity estimation in the context of po-
on the Terra spacecraft using the Lake Tahoe, CA/NV, USA, Automated Validation tential future missions. Remote Sens. Environ. 148, 158–167.
Site. IEEE Trans. Geosci. Remote Sens. 45, 1798–1807. Sobrino, J.A., Del Frate, F., Drusch, M., Jiménez-Muñoz, J.C., Manunta, P., Regan, A.,
Hulley, G.C., Hook, S.J., 2009. Intercomparison of versions 4, 4.1 and 5 of the MODIS 2016. Review of thermal infrared applications and requirements for future high-re-
land surface temperature and emissivity product and validation with laboratory solution sensors. IEEE Trans. Geosci. Remote Sens. 54, 2963–2972.
measurements of sand samples from the Namib desert, Namibia. Remote Sens. Theocharous, E., Usadi, E. and Fox, N. (2010). CEOS comparison of IR brightness tem-
Environ. 113, 1313–1318. perature measurements in support of satellite validation. Part I: Laboratory and ocean
Hulley, G.C., Hook, S.J., 2011. Generating consistent land surface temperature and surface temperature comparison of radiation thermometers. NPL REPORT OP 3.
emissivity product between ASTER and MODIS data for earth science research. IEEE National Physical Laboratory, Teddington, UK.
Trans. Geosci. Remote Sens. 49, 1304–1315. Trigo, I.F., Monteiro, I.T., Olesen, F., Kabsch, E., 2008. An assessment of remotely sensed
Hulley, G.C., Hook, S.J., 2012. A radiance-based method for estimating uncertainties in land surface temperature. J. Geophys. Res. 113, D17108.
the Atmospheric Infrared Sounder (AIRS) land surface temperature product. J. Voogt, J., Oke, T., 2003. Thermal remote sensing of urban climates. Remote Sens.
Geophys. Res. Lett. 117, D20117. https://doi.org/10.1029/2012JD19102. Environ. 86, 370–384.
Hulley, G.C., Hook, S.J., Abbott, E., Malakar, N., Islam, T., Abrams, M., 2015. The ASTER Wan, Z., 2008. New refinements and validation of the MODIS land-surface temperature/
Global Emissivity Dataset (ASTER GED): mapping Earth's emissivity at 100 meter emissivity product. Remote Sens. Environ. 112, 59–74.
spatial scale. Geophys. Res. Lett. 42, 7966–7976. Wan, Z., 2014. New refinements and validation of the collection-6 MODIS land-surface
Ingram, P.M., Muse, A.H., 2001. Sensitivity of iterative spectrally smooth temperature/ temperature/emissivity product. Remote Sens. Environ. 140, 36–45.
emissivity separation to algorithmic assumptions and measurement noise. IEEE Wan, Z., Dozier, J., 1996. A generalized split-window algorithm for retrieving land-sur-
Trans. Geosci. Remote Sens. 39, 2158–2167. face temperature from space. IEEE Trans. Geosci. Remote Sens. 34, 892–905.
Islam, T., Hulley, G.C., Malakar, N.K., Radocinski, R.G., Guillevic, P.C., Hook, S.J., 2017. Wan, Z., Li, Z.-L., 1997. A physics-based algorithm for retrieving land-surface emissivity
A physics-based algorithm for the simultaneous retrieval of land surface temperature and temperature from EOS/MODIS data. IEEE Trans. Geosci. Remote Sens. 35,
and emissivity from VIIRS thermal infrared data. IEEE Trans. Geosci. Remote Sens. 980–996.
55, 563–576. Wan, Z., Li, Z.-L., 2008. Radiance-based validation of the V5 MODIS land-surface tem-
Kabsch, E., Olesen, F.-S., Prata, F., 2008. Initial results of the land surface temperature perature product. Int. J. Remote Sens. 29, 5373–5395.
(LST) validation with the Evora, Portugal ground-truth station measurements. Int. J. Wan, Z., Zhang, Y., Zhang, Q., Li, Z.-L., 2002. Validation of the land-surface temperature
Remote Sens. 29, 5329–5345. product retrieved from Terra Moderate Resolution Imaging Spectroradiometer data.
Kogan, F.N., 2001. Operational space technology for global vegetation assessment. Bull. Remote Sens. Environ. 83, 163–180.
Am. Meteorol. Soc. 82, 1949–1964. Wan, Z., Wang, P., Li, X., 2004a. Using MODIS land surface temperature and normalized
Kondratyev, K.Y., 1969. Radiation in the Atmosphere. Academic Press, New York, USA. difference vegetation index product for monitoring drought in the southern Great
Krishnan, P., Kochendorfer, J., Dumas, E.J., Guillevic, P.C., Baker, C.B., Meyers, T.P., Plains, USA. Int. J. Remote Sens. 25, 61–72.
Martos, B., 2015. Comparison of in-situ, aircraft, and satellite land surface tem- Wan, Z., Zhang, Y., Zhang, Q., Li, Z.-L., 2004b. Quality assessment and validation of the
perature measurements over a NOAA climate reference network site. Remote Sens. MODIS global land surface temperature. Int. J. Remote Sens. 25, 261–274.
Environ. 165, 249–264. Wang, K., Liang, S., 2009. Evaluation of ASTER and MODIS land surface temperature and
Lambin, E.F., Ehrlich, D., 1997. Land-cover changes in sub-Saharan Africa (1982–1991): emissivity product using long-term surface longwave radiation observations at
application of a change index based on remotely-sensed surface temperature and SURFRAD sites. Remote Sens. Environ. 113, 1556–1565.
vegetation indices at a continental scale. Remote Sens. Environ. 61, 181–200. Wang, W., Liang, S., Meyers, T., 2008. Validating MODIS land surface temperature pro-
Leng, P., Song, X.N., Li, Z.-L., Ma, J.W., Zhou, F.C., Li, S., 2014. Bare surface soil moisture duct using long-term nighttime ground measurements. Remote Sens. Environ. 112,
retrieval from the synergistic use of the optical and thermal infrared data. Int. J. 623–635.
Remote Sens. 35, 988–1003. Weng, Q., 2009. Thermal infrared remote sensing for urban climate and environmental
Li, Z.-L., Becker, F., 1993. Feasibility of land surface temperature and emissivity de- studies: methods, applications, and trends. ISPRS J. Photogramm. Remote Sens. 64,
termination from AVHRR data. Remote Sens. Environ. 43, 67–85. 335–344.
Li, X., Cheng, G., Liu, S., Xiao, Q., Ma, M., Jin, R., et al., 2013a. Heihe Watershed Allied Xu, H., Yu, Y., Tarpley, D., Göttsche, F.-M., Olesen, F.-S., 2014. Evaluation of GOES-R
Telemetry Experimental Research (HiWATER): Scientific Objectives and land surface temperature algorithm using SEVIRI satellite retrievals with in situ

28
S.-B. Duan, et al. Remote Sensing of Environment 225 (2019) 16–29

measurements. IEEE Trans. Geosci. Remote Sens. 52, 3812–3822. of China. Remote Sens. 6, 11494–11517.
Yu, Y., Tarpley, D., Privette, J.L., Flynn, L.E., Xu, H., Chen, M., Vinnikov, K.Y., Sun, D., Zhao, W., Li, A., Zhao, T., 2017a. Potential of estimating surface soil moisture with the
Tian, Y., 2012. Validation of GOES-R satellite land surface temperature algorithm triangle-based empirical relationship model. IEEE Trans. Geosci. Remote Sens. 55,
using SURFRAD ground measurements and statistical estimations of error properties. 6494–6504.
IEEE Trans. Geosci. Remote Sens. 50, 704–713. Zhao, W., Li, A., Jin, H., Zhang, Z., Bian, J., Yin, G., 2017b. Performance evaluation of the
Yu, W., Ma, M., Wang, X., Geng, L., Tan, J., Shi, J., 2014. Evaluation of MODIS LST triangle-based empirical soil moisture relationship models based on Landsat-5 TM
product using longwave radiation ground measurements in the northern arid region data and in situ measurements. IEEE Trans. Geosci. Remote Sens. 55, 2632–2645.

29

Vous aimerez peut-être aussi