Remote Sensing of Environment 102 (2006) 99 – 115

www.elsevier.com/locate/rse

Land surface temperature derived from airborne hyperspectral scanner

thermal infrared data
José A. Sobrino a,⁎, Juan C. Jiménez-Muñoz a , Pablo J. Zarco-Tejada b ,
Guadalupe Sepulcre-Cantó b , Eduardo de Miguel c
a
Global Change Unit, Department of Thermodynamics, Faculty of Physics, University of Valencia, Valencia, Spain
b
Instituto de Agricultura Sostenible (IAS), Consejo Superior de Investigaciones Científicas (CSIC), Córdoba, Spain
c
Instituto Nacional de Técnica Aeroespacial (INTA), Dpto. de Observación de la Tierra, Teledetección y Atmósfera, Madrid, Spain
Received 24 October 2005; received in revised form 30 January 2006; accepted 1 February 2006

Abstract

The AHS (airborne hyperspectral scanner) instrument has 80 spectral bands covering the visible and near infrared (VNIR), short wave infrared
(SWIR), mid-infrared (MIR), and thermal infrared (TIR) spectral range. The instrument is operated by Instituto Nacional de Técnica Aerospacial
(INTA) and it has been involved in several field campaigns since 2004. The main goal of this paper is to analyze the feasibility of retrieving land
surface temperature from the 10 AHS thermal infrared bands, from 71 to 80, located in the region between 8 and 13 μm. For this purpose, three
different methods have been considered: (i) the single-channel method, which uses only one thermal band; (ii) the split-window method, which
uses a combination of two thermal bands; and (iii) the TES (temperature and emissivity separation) method, which needs at least four thermal
bands. The calibration of the AHS thermal bands and the algorithms have been tested with in situ measurements collected in the framework of the
SPARC (Spectra Barrax Campaign) and EAGLE (Exploitation of AnGular effects in Land surfacE observations from satellites) field campaigns,
which took place simultaneously in the agricultural area of Barrax (Albacete, Spain), and also in the framework of the AGRISPECTRA field
campaign, carried out over an olive orchard in Córdoba (Spain), in July 2004. AHS flights were conducted at two different altitudes, 975m and
2745m above ground level. The results show that AHS bands 71 (8.18 μm), 72 (8.66 μm), and 73 (9.15 μm) were affected by a calibration
problem. Taking into account that AHS bands 74 (9.60μm) and 80 (12.93μm) are located in an absorption region, bands from 75 to 79 have been
finally selected for land surface temperature retrieval. The single-channel method has been applied to AHS band 75 (10.07 μm), which shows the
highest atmospheric transmissivity, whereas the split-window method has been applied to the combination between bands 75 and 79 (12.35 μm),
which provides the best results according to simulated data. All the AHS thermal bands ranging from 75 to 79 have been used in the TES method.
The tests conducted on the different algorithms used in this study show that single-channel and split-window methods provided similar results,
with root mean square errors (RMSE) between 1.6 and 1.9 K. The TES method slightly improved the results, with a RMSE of 1.4K.
© 2006 Elsevier Inc. All rights reserved.

Keywords: Airborne hyperspectral scanner; Visible and near infrared; Spectral band

1. Introduction Earth remote sensing instruments in 1999, which included TIR

sensors, thermal data are poised to become a major source of
Although remote sensing is recognized as a powerful tool in quantitative and qualitative information on land surface
the collection, analysis, and modeling of environmental data, processes and for their characterization, analysis, and modelling
less attention has been given to the use of thermal infrared (TIR) (Quattrochi & Luvall, 2004). There are two fundamental
remote sensing. With the launch of the NASA Terra suite of reasons why TIR data contribute to an improved understanding
of land surface processes: (i) through measurement of surface
⁎ Corresponding author. Dr Moliner 50-46100 Burjassot, Valencia, Spain. temperatures as related to specific landscape and biophysical
Tel./fax: +34 96 354 31 15. components and (ii) through relating surface temperatures with
E-mail address: [email protected] (J.A. Sobrino). energy fluxes for specific landscape phenomena or processes
0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2006.02.001
100 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

(Quattrochi & Luvall, 1999). Different thematic areas where ground-based measurements; and in Section 4, we describe the
TIR remote sensing data have been applied to the analysis of AHS sensor. The results obtained are shown in Section 5, which
landscape attributes or land surface processes could be found, as includes the analysis of the calibration of the AHS thermal
for example: (i) landscape characterization, (ii) thermal intertia bands and the tests and validation conducted on the algorithms.
and landscape analysis, (iii) estimation of energy fluxes, (iv) Finally, we present the conclusions drawn in this study.
evaporation/evapotranspiration/soil moisture, (v) quantification
of energy balance or energy flux, and (vi) forest energy 2. Methods for LST retrieval
exchange.
In all these thematic areas and other environmental studies, Different methods or techniques have been proposed in the
land surface temperature (LST) is a key parameter which can be last years in order to retrieved LST from thermal infrared data.
retrieved from TIR data. Hence, except for solar irradiance A review of methods can be found in Sobrino et al. (2002), Dash
components, most of the fluxes at the surface/atmosphere et al. (2002), and Kerr et al. (2004). Basically, these methods
interface can only be parameterized through the use of surface could be classified as (i) single-channel methods, which used
temperature. LST can play either a direct role, such as when only one thermal band; (ii) two-channel or split-window
estimating long wave fluxes, or indirectly as when estimating methods, which use a combination between two thermal
latent and sensible heat fluxes (Kerr et al., 2004). LST can be bands; and (iii) two-angle methods, which use one thermal
also used as an input data in water and energy balance studies, band and two view angles. There are also other methods, which
which is an important issue in environmental studies in order to use more than two thermal bands or based on other techniques
achieve a better understanding on the exchange of heat and (see for example Becker & Li, 1990a; Sun & Pinker, 2003; Wan
moisture between the land surface and lower atmosphere, also & Li, 1997). The availability of sensors with multispectral
leading to a better understanding on the water and carbon capabilities in the thermal infrared region has also favoured the
cycles. Moreover, many other applications rely on the development of methods for LST retrievals, which use several
knowledge of LST, such as geology, hydrology, vegetation thermal bands, as the temperature and emissivity separation
monitoring, global circulation models, and evapotranspiration, (TES) method, developed by Gillespie et al. (1998), which also
among others. provides surface emissivities jointly with the temperature. In
Different approaches have been published in the last years in order to retrieve LST from AHS data, we considered the single-
order to retrieve LST from satellite data. However, most of these channel, two-channel, and TES methods, which are widely used
approaches have been developed for low spectral resolution by the scientific community and described below.
sensors, with only one or two thermal bands, as the Thematic
Mapper (TM) onboard the LANDSAT platform, the Advanced 2.1. Theoretical background
Very High Resolution Radiometer (AVHRR) onboard the
NOAA series, the Along Track Scanning Radiometer (ATSR) Methods for LST retrieval based on the radiative transfer
onboard ERS-1 and ERS-2 platforms, and the most recent equation, which can be written in the thermal infrared region for
Advanced ATSR (AATSR) onboard the ENVISAT platform. a certain sensor band i as:
Among these methods, we highlight the single-channel and
two-channel or split-window algorithms, which will be Li ðTi Þ ¼ LLLR
i si þ Lz
i ð1Þ
described below. Different spectral and spatial resolution
where Li (Ti) is the radiance measured by the sensor (Ti is the at-
sensors are currently available, with several thermal bands as
sensor brightness temperature), τi is the atmospheric transmis-
the Advanced Spaceborne Thermal Emission and Reflection
sivity, and Li↑ is the up-welling path radiance. The term LLLR is
Radiometer (ASTER), and the Moderate Resolution Imaging
the land-leaving radiance (LLR) or radiance measured at
Spectrometer (MODIS) onboard Terra (or also AQUA for
ground-level, which is given by:
MODIS) satellite, or the SEVIRI (Spinning Enhanced Visible
and Infrared Imager) onboard the MSG-1 (Meteosat Second FiA
Generation). Airborne sensors available are the digital airborne LLLR ¼ ei Bi ðTs Þ þ ð1  ei Þ ð2Þ
i
p
imaging spectrometer (DAIS), the airborne hyperspectral
scanner (AHS), or the airborne reflective/emissive spectrometer where εi is the surface emissivity, Bi (Ts) is the Planck radiance
(ARES), among others. Despite the methods developed for low at surface temperature Ts, and Fi↓ is the down-welling sky
resolution sensors that can be adapted to high resolution data, irradiance. In Eq. (2), the assumption of Lambertian behaviour
new methods have been developed in order to retrieve LST for the surface has been considered in order to express the
from multispectral thermal data, as the TES method (Gillespie reflection term as (1 − ε)π− 1F↓. The magnitudes involved in
et al., 1998). Eqs. (1) and (2) are band averaged values using the spectral
The purpose of this paper is to analyze the feasibility of response functions.
retrieving LST from AHS thermal infrared data as well as the
accuracy obtained depending on the number of thermal bands 2.2. Single-channel method
used on the LST retrieval. The paper is organized as follows: in
Section 2, we describe the methods used in this study for LST Single-channel methods retrieve LST only from one thermal
retrieval; in Section 3, we present the field campaigns and band. Different single-channel algorithms can be found in the
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 101

literature (see for example Jiménez-Muñoz & Sobrino, 2003; 2.4. TES method
Qin et al., 2001). In this paper, we propose a single-channel
algorithm based on the solving for the Planck radiance The TES (temperature and emissivity separation) method
according to Eqs. (1) and (2): estimates land surface emissivity εi and temperature Ts from
multispectral thermal data. The TES algorithm is described in
Li ðTi Þ  Lz ð1  ei ÞFiA =p detail in Gillespie et al. (1998), so a detailed description of the
Bi ðTs Þ ¼ i
 ð3Þ
e i si ei method will not be given here. It is based on the radiative
transfer equation (see Eqs. (1) and (2)) applied to each thermal
LST can be finally retrieved by inversion of the Planck's law:
band and it is composed by three modules (NEM, RATIO, and
" !#1 MMD). The method uses atmospherically corrected data (land-
c2 c1
Ts ¼ ln 5 þ 1 ð4Þ leaving radiance and down-welling atmospheric irradiances)
ki ki B i and a semi-empirical relation determined from laboratory
spectra, between the minimum emissivity (εmin) and spectral
where Bi is given by Eq. (3), and c1 and c2 are the constants in
contrast (maximum–minimum difference, MMD). The semi-
the Planck's function (c1 = 1.19104 × 108 W μm4 m− 2 sr− 1 and
empirical relation between εmin and MMD has been originally
c2 = 14387.7 μm K). In this method, the surface emissivity εi is
calculated for the ASTER sensor onboard the Terra platform,
assumed to be known. Assuming that atmospheric water vapour
with five thermal bands in the region between 8 and 12μm.
is the primary factor controlling the magnitude of atmospheric
Using 86 laboratory spectra, Gillespie et al. (1998) found the
transmittance, the hemispheric down-welling radiance and up-
following simple power law:
welling atmospheric radiance for a given band, different
approaches can be proposed in order to related these emin ¼ 0:994  0:687MMD0:737 ð6Þ
atmospheric parameters with the atmospheric water vapour,
which is a more easily accessible parameter. The results with a correlation coefficient squared of R2 = 0.983 and 95% of
obtained will be shown in Section 3. This procedure has been the samples falling within ± 0.02 emissivity units. Once the
applied to high resolution data by Jiménez-Muñoz and Sobrino surface emissivities have been recovered, LST is obtained by
(2005), providing good results (root mean square errors inversion of the Planck's law according to Eq. (4) and using the
RMSE < 1.8 K) for atmospheres with low water vapour content thermal band for which the emissivity is maximum, in order to
and using the thermal band for which the atmospheric minimize the effect of the sky irradiance in the reflection term.
transmissivity is higher.
3. Field data
2.3. Two-channel algorithms
Field campaigns are of importance in order to validate or
The basis of the two-channel technique (or split-window test the algorithms developed for retrieving a certain bio-
when it is applied in the region 10–12.5 μm) is that the geophysical parameter from satellite or aircraft data. The
atmospheric attenuation suffered by the surface emitted aircraft images and the in situ data used in order to conduct the
radiance is proportional to the difference between the at-sensor study shown in this paper were acquired in the framework of
radiances measured simultaneously in two different thermal three field campaigns related to three different projects. The
channels (McMillin, 1975). Many papers have used this SPARC (Spectra Barrax Campaign, http://gpds.uv.se/sparc)
technique to extract sea surface temperature (Deschamps & and EAGLE (Exploitation of AnGular effects in Land surfacE
Phulpin, 1980; McClain et al., 1985; Sobrino et al., 1993a, etc.) observations from satellites, http://www.uv.es/ucg/eagle) field
and land surface temperature (Becker & Li, 1990b; Prata, 1993; campaigns took place simultaneously in the Barrax test site
Price, 1984; Sobrino et al., 1991, 1994, etc.). In this paper, the (Albacete, Spain), in 2004. Another field campaign was
two-channel algorithm proposed by Sobrino and Raissouni conducted in 2004 in an olive orchard located in Córdoba
(2000) has been used, which takes into account the emissivity (Spain), in the framework of the AGRISPECTRA (Estimation
and water vapour effects: of Leaf and Crop Biophysical Variables in Olive and Vineyard
Canopies through Hyperspectral Remote Sensing Methods for
Ts ¼ Ti þ a1 ðTi  Tj Þ þ a2 ðTi  Tj Þ2 þ a0 Integration with Precision Agriculture) project. The Barrax test
þ ða3 þ a4 wÞð1  eÞ þ ða5 þ a6 wÞDe ð5Þ site is an agricultural area situated in the West of the province
of Albacete, 28 km from the capital town (39°3′N, 2°6′W)
where Ts is the surface temperature (in K), Ti and Tj are the at- (Moreno et al., 2001). Fig. 1 shows the study area of Barrax
sensor brightness temperatures of the different thermal DAIS from an AHS image acquired on 15 July 2004 at 12:43 UTC,
channels (in K), ε = (εi + εj) / 2 and Δε = (εi − εj) are the mean in which the plots where field measurements were carried out
effective emissivity and the emissivity difference, w is the total are also displayed. The Córdoba site is a 4 ha irrigated olive
atmospheric water vapour (in g/cm2), and, finally, ak (k = 0 to 6) orchard (Olea europaea L. cv. ‘Arbequino’) located in
are the numerical coefficients of the two-channel algorithm. southern Spain (37.8°N, 4.8°W). A full description of this
These coefficients can be obtained by means of a simulation experimental area be found in Sepulcre-Cantó et al. (in press).
procedure, which will be described in Section 3. Fig. 2 shows the study area from an AHS image acquired on
102 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

Table 1
Main technical specifications for the field radiometers (FOV: field of view)
Model Bands μm Range °C Accuracy °C FOV°C
CIMEL 312-1 8–13 − 80 to 60 0.1 10
8.2–9.2
10.3–11.3
11.5–12.5
CIMEL 312-2 8–13 − 80 to 60 0.1 10
11–11.7
10.3–11
8.9–9.3
8.5–8.9
8.1–8.5
EVEREST 3000 8–14 − 40 to 100 0.5 4
RAYTEK ST8 8–14 − 30 to 100 0.5 8
RAYTEK MID 8–14 − 40 to 600 0.5 20

Fig. 1. The study area of Barrax. The image corresponds to AHS band 75 temperatures: (i) by means of transects, i.e., measuring while
(10.07μm) raw data and it was acquired on 15 July 2004 at 12:43 UTC, with a walking along the field, and (ii) by means of fixed
pixel size of 7m. measurements by placing the radiometers on poles with an
altitude of around 2 m.
25 July 2004 at 9:30 UTC, in which the olive orchard is A total of 10 infrared sensors Apogee model IRTS-P were
highlighted. placed on poles with an altitude of 6 m over 10 olive trees in the
In the SPARC/EAGLE field campaign, temperature Cordoba site (AGRISPECTRA project) in order to monitor
measurements were made using different broadband and crown temperature continuously as function of a tree water
multiband field thermal radiometers. Models EVEREST 3000, status gradient obtained through drip irrigation method.
RAYTEK ST8, and RAYTEK MID correspond to broadband Previous to the field installation, the IRT sensors were calibrated
radiometers, whereas models CIMEL CE 312-1 and CIMEL in the laboratory and under natural sun conditions to study the
CE 312-2 ASTER are multiband radiometers (they also have IRT response to the diurnal temperature variation. Temperature
a broad band). In addition, two blackbodies (EVEREST 1000 over the course of the day varied between 25 °C and 40 °C,
and GALAI 204-P) were used for calibration purposes. Table enabling a comparison between the IRT-estimated temperature
1 summarizes the main technical characteristics of the thermal and a thermocouple type K (chromel–alumel) in contact with
radiometers. Field measurements were made over two plots of the water target used for calibration. The observed errors agreed
bare soil (referred as ‘bare soil 1’ and ‘bare soil 2’) and two with the sensitivity of the instrument (Apogee, www.apogee-
plots of corn (referred as ‘corn 1’ and ‘corn 2’), one plot of inst.com) yielding a deviation of ± 0.4 °C between the 5°C and
water and one plot of grass (see Fig. 1). Two different 40°C range. The 52° field-of-view (FOV) of the IRTS-P sensors
techniques were considered in order to measure surface placed on the top of each olive tree at 1 m distance from the
crown enabled the measurement of an integrated canopy
temperature for each single tree crown.

4. The AHS sensor

4.1. Technical characteristics

The airborne hyperspectral scanner (AHS) (developed by

SensyTech Inc., currently ArgonST, USA) is operated by the
Spanish Institute of Aeronaoutics (INTA) and it was placed
onboard the aircraft CASA 212-200 Paternina. The AHS sensor
is based on the integration of many advanced technologies
developed by SenSyTech under R&D contracts over the past
few years. While the combination of these components is
offered here for the first time, each of the individual items has
been delivered and field-tested in operational use. The AHS
incorporates advanced components to ensure high performance
while maintaining the ruggedness to provide operational
Fig. 2. The study area of Córdoba. The image has been obtained from the at- reliability in a survey aircraft. The main AHS technical
sensor radiance in AHS band 75 (10.07μm) and it was acquired on 25 July 2004 specifications and the arrangement of the spectral bands are
at 9:30 UTC, with a pixel size of 2.5 m. 80 bands in four ports (VIS, NIR, SWIR, MWIR, and LWIR);
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 103

Table 2 4.3. AHS data processing and atmospheric correction

AHS thermal bands
Band FWHM (μm) Effective wavelength (μm) The processing of the AHS imagery included the raw data to
71 7.95–8.42 8.18 radiance transformation and the atmospheric correction. AHS
72 8.45–8.84 8.66 images were processed to at-sensor radiance for the thermal
73 8.94–9.35 9.15 infrared bands (from 71 to 80) using calibration coefficients
74 9.38–9.81 9.60
determined during flight from two blackbody sources which are
75 9.85–10.27 10.07
76 10.31–10.86 10.59 viewed for every mirror scan. The first (cool) blackbody has
77 10.89–11.45 11.18 been set to 15°C, whereas the second (warm) blackbody had
78 11.49–12.05 11.78 been set to 55 °C. This procedure is performed for every thermal
79 12.09–12.57 12.35 band on every scanline. The AHS onboard blackbodies are
80 12.65–13.14 12.93
copper plates covered with a black paint with emissivity greater
than 0.99 through the 3 to 14 μm spectral range (note that this
FOV: 90° (± 45°), IFOV: 2.5 mrad, GFOV: 2–6 m at 140 kt specification permits a bias < 0.5°C in the measured tempera-
cruise speed; scan speed: 6.25, 12.5, 18.75, 25, 31.25, 35rps; 12 ture), and they subtend 4° when viewed by the radiometer. The
bits digitised; 750 samples per line; black body thermal performance of the thermal control is not specified by the
references (set to 15 °C and 55 °C). The arrangement of the manufacturer and we have observed some temporal instabilities.
AHS thermal bands from 71 to 80 is given in Table 2. So, we use the actual line-by-line blackbody data in the
calibration instead of the nominal or average one. As the
4.2. Imagery acquired temporal variations show a low frequency, we have not
considered the possible delay in the response of the measuring
During the SPARC/EAGLE campaign, the AHS flights thermistors as a source of error. Circuitry to monitor the
took place in 2 days, 15 July 2004 (flight 1) and 18 July 2004 blackbodies temperature (YSI Thermilinear thermistors net-
(flight 2). Flight 1 started at 10:43 UTC and finished at 12:45 work) is designed to yield better than 0.5° accuracy (including
UTC. It was composed by 7AHS images acquired at low the sensors). The non-linearity effects are limited to ± 0.15 °C.
altitude (975 m above ground level—AGL, pixel size 2.5 m) The atmospheric correction of AHS images was performed
and 2 AHS images acquired at high altitude (2745 m AGL, using the radiative transfer equation given by Eq. (1), from
pixel size 7 m). Flight 2 started at 10:30 UTC and finished at which the land-leaving radiance (LiLLR) can be obtained from
12:20 UTC, also composed by seven images at low altitude the at-sensor radiance after compensation of atmospheric effects
and two images at high altitude (same characteristics as flight according to:
1, but different line flights).
In the AGRISPECTRA campaign, three AHS images were Li ðTi Þ  Lz
LLLR ¼ i
: ð7Þ
carried out on 25 July 2004 at different times: 7:30 GMT (flight i
si
1), 9:30 GMT (flight 2), and 12:30 GMT (flight 3). The flights
had similar characteristics to the low flights conducted in The atmospheric parameters τi, Fi↓, and Li↑ involved in the
SPARC/EAGLE, with an altitude of 980m AGL and a pixel size atmospheric correction were estimated using the MODTRAN 4
of 2.5m. radiative transfer code (Berk et al., 1999) and the in situ

1.4
Low flight (975 m)
High flight (2745 m)
1.2 Satellite altitude (700 km)
Atmospheric Transmissivity

71 72 73 74 75 76 77 78 79 80
1.0

0.8

0.6

0.4

0.2

0.0
7 8 9 10 11 12 13 14 15
Wavelength (μm)

Fig. 3. Location of the AHS thermal bands. The plot also shows the atmospheric transmissivity spectrum overlayed with the location of the AHS thermal bands for the
low (975m AGL) and high (2745m AGL) flights carried out on 15 July 2004 and also for a typical satellite altitude (700km).
104 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

radiosoundings launched almost simultaneously with the AHS Eq. (2), since at ground-level is assumed that τ ≈ 1 and L↑ ≈ 0.
overpass. The band values were finally obtained using the filter Surface emissivity values (ε) can be measured in situ, for
functions of the AHS thermal bands. These filter functions are example using the box method (Nerry et al., 1990), or extracted
shown in Fig. 3 compared to the atmospheric transmissivity for from spectral libraries, as for example the ASTER spectral
the low flight (975 m AGL), the high flight (2745 m AGL), and a library (http://speclib.jpl.nasa.gov), whereas F ↓ /π can be
typical satellite altitude (700 km). Bands 71 and 80, located approximately measured by pointing with the radiometer to
around 8 and 13 μm, respectively, show the highest atmospheric the sky with a near-nadir view or, more accurately, by pointing
absorption, whereas bands 75 to 79 are located in the to the sky with a view angle of 53° measured from the zenith
atmospheric window 10–12.5 μm, with band 75 showing the (Kondratyev, 1969). Therefore, from Eq. (2), it is possible to
highest atmospheric transmissivity. Band 74 is located in the find the value of Ts from ground-based measurements and by
region of the ozone absorption, but in the AHS flights this inversion of the Planck's law. The term ‘radiometric temper-
absorption is not observed because the maximum absorption ature’ (Trad) is used when Planck's law is inverted using the
of the ozone is usually located at atmospheric altitudes higher quantity LLLR, so the difference between the land surface
than 10 km. temperature (Ts) and the radiometric temperature (Trad) is due to
the emissivity (ε) and atmospheric effects (F↓), both coupled.
5. Results and analysis In order to illustrate the differences between radiometric and
land surface temperatures and also the problems related with the
5.1. LST estimation from ground-based measurements thermal heterogeneity of the natural surfaces, Fig. 4 shows an
example of the Ts and Trad values obtained with the Raytek MID
The thermal radiometers described in Section 3 and used at radiometer fixed on the mast between 13:00 and 13:15 LT (local
ground-level measure the land-leaving radiance (LLLR) given by time, GMT = LT − 2) the 15th of July in 2004 over bare soil (Fig.

(a) 52
Trad
51
Ts
50

49
Temperature (°C)

48

47

46

45

44

43

42
12:59 13:00 13:01 13:02 13:03 13:04 13:05 13:06 13:07 13:08 13:09 13:10 13:11 13:12 13:13 13:14 13:15 13:16
Local Time
(b)
32.0

31.5

31.0
Temperature (°C)

30.5

30.0

29.5

29.0

Trad
28.5
Ts

28.0
12:59 13:00 13:01 13:02 13:03 13:04 13:05 13:06 13:07 13:08 13:09 13:10 13:11 13:12 13:13 13:14 13:15 13:16
Local Time

Fig. 4. Radiometric temperature (Trad) and surface temperature (Ts) measured with the Raytek MID radiometer located on fixed masts over (a) bare soil and (b) green
grass.
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 105

4a) and grass (Fig. 4b) in the Barrax site. The difference figure shows a significant variability in the measurements
between Ts and Trad is higher for bare soil (around 2K) than for carried out by transects, as has been stated before. The
green grass (around 1 K), due to the higher emissivity for green differences between the fixed measurements and the transect
grass. Surfaces with high surface emissivity minimize the effect measurements are also lower for green grass, which is a more
of the emissivity and atmospheric correction. The graphs homogeneous surface. The variability shown in the radiometric
included in Fig. 4 also show a higher heterogeneity for bare temperatures can be also due to turbulence induced fluctuations
soil than for green grass. Hence, the difference between the of wind speed at the surface (Balick et al., 2003). These
maximum and minimum value for bare soil is 5 K, whereas for fluctuations can be larger than 1K, so accuracies worse than 1 K
green grass is 1.2K. Moreover, differences between two conse- are always expected in the validation procedures. These
cutive measurements (in steps of 1 min) are higher than 1.5 K for examples illustrate the difficulties involved in the field
bare soil but lower than 0.5K for green grass. The heterogeneity measurements and the validation of satellite products, mainly
of the surfaces can be also shown by calculating the mean value over heterogeneous surfaces.
and the standard deviation of the measurements. Values of Ts =
(321.0 ± 1.5) K for bare soil and Ts = (304.1 ± 0.4) K for green 5.2. Analysis of the AHS thermal data: calibration and optimal
grass have been obtained, showing a higher standard deviation bands
for bare soil. Another example is presented in Fig. 5, which
shows a comparison between the values measured with the Eqs. (1) and (2) can be used in order to reproduce the
Raytek MID radiometer located on fixed masts and the values radiance measured by the sensor (Li) or the at-sensor brightness
measured with the Raytek ST8 radiometer by making transects temperature (Ti) (by inversion of the Planck's law) if the
around the same sample for bare soil and green grass. This atmospheric parameters (τi, Fi↓, and Li↑), the emissivity spectrum

(a)
50
RAYTEK MID (Fixed)
RAYTEK ST8 (Transect)
Radiometric Temperature (°C)

48

46

44

42

40

38
12:51 12:54 12:57 13:00 13:03 13:06 13:09 13:12
Local Time
(b)
35

34
Radiometric Temperature (°C)

33

32

31

30

29

28

27
RAYTEK MID (Fixed)
26
RAYTEK ST8 (Transect)

25
13:24 13:26 13:27 13:29 13:30 13:32 13:33 13:35 13:36 13:37 13:39
Local Time

Fig. 5. Comparison between the radiometric temperature measured with the Raytek MID radiometer located on fixed masts and the radiometric temperature measured
with the Raytek ST8 radiometer by making transects over (a) bare soil and (b) green grass.
106 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

(εi), and the LST (Ts) are known. Values of at-sensor brightness emissivity and reflection terms (1 − ε)π− 1F↓ involved in Eq. (2)
temperatures have been compared with the extracted from the have a minimal contribution. The bare soil plot provides a good
AHS images (by selecting boxes of m × n pixels located around shape for the spectrum; however, high differences (> 4K) have
the measurement sites) in order to check the calibration of the been found, except for the bare soil observed in the AHS image
AHS sensor. The atmospheric parameters have been extracted acquired on 18 July at 11:03 UTC (low flight, Fig. 8). An
from the radiosoundings launched in situ and the MODTRAN 4 explanation of the different behaviour of this bare soil and the
code. The emissivity spectra for water and grass have been others could be found in the different radiometers used in the
obtained from the ASTER spectral library, whereas for the corn measurements. This last bare soil plot was measured with the
plot a constant value of 0.99 have been assumed due to its high CIMEL radiometer, which is the most accurate thermal
coverage. One sample of soil was collected at the field and sent radiometer. Problems on the ground-based measurements have
to the Jet Propulsion Laboratory (JPL) in order to measure its been also found in the corn plots for the low flight carried out on
emissivity spectrum. Fig. 6 shows all the emissivity spectra for 18 July at 11:03 UTC (see Fig. 8). Differences when comparing
the AHS thermal bands considered in this study. The analysis is with ground-based measurements higher than 4K have been
only shown for the SPARC/EAGLE campaign, since in the obtained. The corn plot for the AHS image acquired at 10:50
AGRISPECTRA campaign the atmosphere was not character- UTC (also included in Fig. 8) shows ground-based brightness
ized. From all the AHS images and ground-based measurements temperatures substantially different to those measured at 11:03
available, only the plots measured simultaneously with the UTC. Taking into account that the corn plot was in a fully
sensor overpass have been selected for checking the calibration covered stage and well irrigated, temperatures between the two
of the AHS thermal bands, leading to a total amount of 19 plots consecutive flights should be similar. Significant differences
acquired over different surfaces (bare soil, grass, corn, and have been found also in some green grass plots. Despite these
water). The results obtained are shown in Fig. 7 for the high plots have been labelled as “green”, a visual inspection at the
flight (975 m AGL) and in Fig. 8 for the low flight. All the field site shows a mixed contribution composed by green and
graphs shown in these figures have been scaled in steps of 1 K also senescent grass, leading to a less homogeneous plot.
(Y-axis, horizontal lines) for a better comparison of the Despite these problems, the AHS thermal bands provides good
differences obtained over the different plots and flights. results in comparison with ground-based measurements for the
According to the results obtained, AHS band 73 (9.15 μm) bands located in the split-window region from 10 to 13μm
seems to be affected by a calibration problem, because the (AHS bands from 75 to 79).
values obtained with this band do not follow the expected
tendency in the shape of the brightness temperatures versus the 5.3. Simulated data for single-channel and split-window
wavelength. AHS band 71 (8.183 μm) also shows some algorithms
calibration problems in the low flights (see Fig. 8), whereas
band 72 (8.659 μm) seems to show a random behaviour. In order to obtain operative algorithms for retrieving LST
Looking to the different plots considered in this analysis, the with single-channel or split-window methods, different atmo-
results obtained over water provide, in general, the better spheric conditions have been simulated. For this purpose the
accordance with the ground-based measurements. It should be MODTRAN 4 radiative transfer code (executed in the thermal
noted that the water behaviour is similar to a blackbody, so the radiance mode) and a set of 54 radiosoundings extracted from

1.00

0.99

0.98
Emissivity

0.97

0.96
WATER
BARE SOIL
0.95
GRASS
CORN

0.94
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

Fig. 6. Emissivity spectra for different plots. Emissivities for water and grass have been extracted from the ASTER spectral library, whereas the bare soil sample was
collected at the field and measured in the Jet Propulsion Laboratory (JPL). A constant value equal to 0.99 has been considered for the corn plot, due to its greenness and
high cover.
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 107

15 July 2004 - 12:43 UTC

320
ground-based

At-sensor Brightness Temperature (K)

BARE SOIL 1 AHS
315

310

305 GRASS

300

295 WATER

290
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (mm)

18 July 2004 - 12:17 UTC

315
ground-based
At-sensor Brightness Temperature (K)

AHS

310
GRASS

305

300

WATER
295

290
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

Fig. 7. Comparison between the ground-based and the AHS at-sensor brightness temperature for the high flights (2745m above ground level) over different surfaces.

the TIGR (TOVS Initial Guess Retrieval) database (Scott & Lz

75 ¼ 0:2746w þ 3:5473w  0:2231
2

Chedin, 1981) and described in Sobrino et al. (1993b) have ðR2 ¼ 0:981; r ¼ 0:3 Wd m2 d Am1 d sr1 Þ ð10Þ
been used. In this way, spectral values of the atmospheric
parameters τ, F↓, and L↑ have been obtained. These values have and for the high flight (2745 m AGL):
been averaged according to the AHS filter functions in order to
obtain the effective values for the thermal bands from 75 to 79 s75 ¼ 0:0143w2  0:0678w þ 1:0006
(see Eq. (3)). Fig. 3 showed that the highest τ is obtained for ðR2 ¼ 0:997; r ¼ 0:007Þ ð11Þ
AHS band 75. For this reason, band 75 (10.07 μm) has been
chosen as the optimal band for applying the single-channel Lz
75 ¼ 0:1531w þ 0:4900w  0:0513
2

algorithm given by Eq. (3). The dependence of this algorithm ðR2 ¼ 0:995; r ¼ 0:07 Wd m2 d Am1 d sr1 Þ ð12Þ
with the atmospheric parameters have been avoided by finding
relations with w. Taking into account the set of the 54 LA75 ¼ 0:0431w2 þ 1:8217w þ 0:1503
radiosoundings mentioned before, the following results have ðR2 ¼ 0:976; r ¼ 0:3 Wd m2 d Am1 d sr1 Þ ð13Þ
been obtained for the low flight (975 m AGL):
where L↓ ≡ F↓/π, R2 is the correlation coefficient squared, and σ
s75 ¼ 0:0505w2  0:0691w þ 1:0010
the standard error of estimation.
ðR2 ¼ 0:998; r ¼ 0:003Þ ð8Þ In order to select the best combination between the bands i
Lz ¼ 0:5638w þ 0:4498w  0:0164
2 and j involved in the split-window algorithm (see Eq. (16)),
75
the numerical coefficients ak have been obtained for all the
ðR ¼ 0:998; r ¼ 0:03 Wd m2 d Am1 d sr1 Þ
2
ð9Þ combinations between AHS bands from 75 to 79, and then a
108 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

15 July 2004 - 11:57 UTC

325
ground-based
At-sensor Brightness Temperature (K) AHS
320 BARE SOIL 1

315

310 GRASS

305

300
WATER

295

290
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

18 July 2004 - 10:30 UTC

297
ground-based
At-sensor Brightness Temperature (K)

WATER
AHS

296

295

294

293

292
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

18 July 2004 - 10:50 UTC

308
WATER: ground-based
At-sensor Brightness Temperature (K)

306 WATER: AHS

CORN 1: ground-based

304 CORN 1: AHS

CORN 2: ground-based
CORN 2: AHS
302

300

298

296

294

292
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

Fig. 8. Comparison between the ground-based and the AHS at-sensor brightness temperature for the low flights (975m above ground level) over different surfaces.
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 109

18 July 2004 - 11:03 UTC

330
ground-based

At-sensor Brightness Temperature (K)

325 BARE SOIL 2 AHS

320

315
GRASS
310

305

300 WATER

295

290
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)
18 July 2004 - 11:03 UTC
306
At-sensor Brightness Temperature (K)

CORN 1: ground-based
CORN 1: AHS
CORN 2: ground-based
304 CORN 2: AHS

302

300

298
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)
18 July 2004 - 11:28 UTC
315
ground-based
At-sensor Brightness Temperature (K)

GRASS AHS

310

305

300
WATER

295

290
8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0
Wavelength (μm)

Fig. 8 (continued).

sensitivity analysis according to Sobrino et al. (2004) (see Eqs. At-sensor brightness temperatures for the bands i and j
(10) to (14) in the cited paper) has been carried out in order to involved in Eq. (5) are simulated using the radiative transfer
estimate the error on the LST retrieved with these algorithms. equation (see Eqs. (1) and (2)), in which atmospheric
110 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

Table 3 5.4. Simulated data for the TES method: relation εmin–MMD
Emissivity spectra extracted from the ASTER spectral library and used in the and testing
simulation procedure
Class Subclass Samples The MMD module included in the TES method uses an
Rocks Igneous 100 empirical relationship between the minimum emissivity (εmin)
Metamorphics 76 and the spectral contrast (MMD) in order to recover the
Sedimentary 68
surface emissivities (Gillespie et al., 1998). This relationship,
Soils Alfisol 9
Aridisol 14 initially obtained for the ASTER sensor, needs to be
Entisol 10 recalculated when other sensors are used. The emissivity
Inceptisol 7 spectra (299 samples) described in the previous section have
Mollisol 9 been employed to find the relation between εmin and MMD
Vegetation Green grass 1
when AHS bands 75 to 79 are considered. Fig. 9 shows the
Dry grass 1
Conifers 1 plot of εmin versus MMD for the 299 emissivity spectra of
Decideous 1 natural surfaces. The following relation has been finally
Water–snow–ice Water 1 obtained for AHS data:
Ice 1
Total: 299
emin ¼ 0:986  1:350MMD1:019 ð16Þ

parameters are obtained again with the MODTRAN 4 code with a correlation coefficient squared of R2 = 0.93 and a
and the set of 54 radiosoundings. In this simulation, surface standard error of estimation of σ = 0.019. It should be noted
emissivity values are also needed. A total amount of 299 that the MMD values involved in Eq. (16) have been
emissivity spectra extracted from the ASTER spectral library obtained from AHS bands 75 to 79, located in the 10–
(http://speclib.jpl.nasa.gov) and averaged according to the 12μm region. In theory, Eq. (16) works better when bands
AHS filter functions have been used in this simulation. All the located in the 8–10 μm region are also included. AHS
emissivity spectra correspond to natural surfaces, as is shown bands located in this region (71 to 74) have not been
in Table 3. Using 54 radiosoundings and 299 emissivity included due to the calibration problems found (Section
spectra, we obtain 16146 data simulating different atmospheric 5.2). Despite some problems could arise when using only
conditions over different natural surfaces. With this amount of AHS bands 75 to 79 over high MMD surfaces (rocks),
simulated data, statistical fits are applied in order to recover accurate results are expected over low MMD surfaces as
the numerical coefficients involved in the split-window agricultural areas. AHS images with no calibration problems
algorithm. The simulated data is also used in the sensitivity will require a new computation for the relationship between
analysis (Eqs. (10) to (14) in Sobrino et al., 2004), in order to εmin and MMD.
obtain the theoretical error of the algorithm over different The TES method has been tested using the simulated data
conditions. The mean value of the error and the standard described in the previous section for different flight altitudes:
deviation are used in order to compute the root mean square 700km, 975m, and 2745m. The LST retrieved with the method
error (RMSE), assumed to be the final theoretical error of the has been compared with the LST included in the simulated data
algorithm. The results obtained are shown in Table 4 for the and extracted from the radiosoundings. In all the cases, the TES
different flight altitudes and also for a typical satellite altitude. method provided a RMSE < 1.1 K.
Regarding to the low and high flights, the best combination is
obtained with AHS bands 75 (10.07 μm) and 79 (12.35 μm).
When a satellite altitude is considered, the best combination is Table 4
obtained with AHS bands 76 (10.59 μm) and 79 (12.35 μm), Error on the land surface temperature retrieved with the split-window algorithm
which agrees with typical split-window combinations used in given in Eq. (5) for different combinations between AHS thermal bands and for
other thermal sensors as NOAA-AVHRR or ENVISAT- different flight altitudes
AATSR. In all the cases (low, high, and satellite altitude), Band ‘i’ Band ‘j’ Low flight High flight Satellite altitude
the error is around 1 K. The split-window algorithms with the (975m AGL) (2745m AGL) (700km AGL)
numerical coefficients for the low and high flight are given in Error (K) Error (K) Error (K)
Eqs. (14) and (15), respectively: 75 76 1.3 1.8 1.6
75 77 1.2 1.3 1.6
Ts ¼ T75 þ 0:485ðT75  T79 Þ þ 0:0068ðT75  T79 Þ2 75 78 1.2 1.1 1.7
þ 0:0798 þ ð47:15  10:80W Þð1  eÞ 75 79 1.1 1.0 1.6
76 77 2.2 2.6 3.0
þ ð49:05 þ 21:53wÞDe ð14Þ 76 78 1.7 1.6 1.8
76 79 1.4 1.1 1.2
Ts ¼ T75 þ 0:734ðT75  T79 Þ þ 0:0096ðT75  T79 Þ2 77 78 2.4 2.6 3.0
77 79 1.7 1.5 1.6
þ 0:1198 þ ð47:46  5:20W Þð1  eÞ 78 79 2.4 2.3 2.3
þ ð61:82 þ 14:97wÞDe ð15Þ The minimum error in each case is highlighted.
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 111

1.0
299 emissivity spectra
2
R = 0.93
0.9 σ = 0.019

0.8
εmin
0.7

0.6

0.5
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
MMD

Fig. 9. Empirical relationship between minimum emissivity (εmin) and spectral contrast (MMD) obtained from the emissivity spectra described in Table 3 (R2 is the
correlation coefficient squared and σ is the standard error of estimation).

5.5. Algorithms validation with ground truth data between the LST retrieved with the algorithm and the measured
in situ. In both cases, the Bias is positive, which indicates that
The algorithms described in the previous sections have been the algorithm overestimates the LST. This overestimation is
validated using the field measurements carried out in the more clearly shown in the results achieved in the AGRISPEC-
framework of the different field campaigns. Single-channel and TRA campaign, with a Bias of 1.7K. We attribute this to the
two-channel methods have been tested in the framework of the poor characterization of the atmosphere during this experiment,
SPARC/EAGLE and AGRISPECTRA field campaigns. In the in which a water vapour value measured 250 km far from the
SPARC/EAGLE campaign, a total amount of 17 plots field site has been chosen. The results obtained with the single-
composed by water, grass, bare soil, and corn have been used, channel algorithm proposed in this paper agree with those
whereas in the AGRISPECTRA campaign a total amount of 30 obtained by Jiménez-Muñoz and Sobrino (2005) also from high
plots have been used (10 trees per flight). In order to apply the resolution data over the Barrax test site, with RMSE < 1.8K.
TES method, an accurate estimation of the atmospheric Results obtained with the split-window algorithm are shown
parameters is needed, since it uses atmospherically corrected in Fig. 11, in which RMSE values of 1.9 K and 1.6K have been
data. In the SPARC/EAGLE campaign, different radiosound- obtained for the SPARC/EAGLE and AGRISPECTRA cam-
ings were launched almost simultaneously with the sensor paigns, respectively. In this last case, the Bias is lower than the
overpass. However, in the AGRISPECTRA campaign, radio- obtained with the single-channel method, which indicates that
sounding data or atmospheric characterizations were not the split-window is less sensitive to an accurate knowledge of
available, so TES method has been only tested in the framework the water vapour content. Additional results obtained with the
of the SPARC/EAGLE campaign. The values of atmospheric split-window algorithm in the AGRISPECTRA campaign can
water vapour content needed in order to apply the single- be found in Sepulcre-Cantó et al. (in press). Despite similar
channel and split-window algorithms are given in Table 5.
These values have been obtained using the radiosoundings
launched in situ in the framework of the SPARC/EAGLE Table 5
Atmospheric water vapour content (w) at different altitudes
campaigns and the MODTRAN 4 radiative transfer code. In the
AGRISPECTRA campaign, no radisoundings were launched, Campaign Date Altitude (km AGL) w (g/cm2)
as has been commented before, so in this case we only have the SPARC/EAGLE 15-July-2004 0.975 0.71
value measured with a sunphotometer in the “El Arenosillo” SPARC/EAGLE 15-July-2004 0.975 1.49
SPARC/EAGLE 15-July-2004 100 1.66
site, located around 250km far from the field and which is
SPARC/EAGLE 18-July-2004 2.745 0.79
included in the AERONET (Aerosol Robotic Network). Details SPARC/EAGLE 18-July-2004 2.745 1.58
can be found at http://aeronet.gsfc.nasa.gov. The total water SPARC/EAGLE 18-July-2004 100 1.74
vapour content extracted from the AERONET data has been AGRISPECTRA 25-July-2004 0.980 0.90
scaled to the altitude of the flight (980 m AGL) using the AGRISPECTRA 25-July-2004 100 2.54
MODTRAN 4 code and assuming the atmospheric profile Values for the SPARC/EAGLE campaign have been obtained using the
included in the midlatitude summer atmosphere. radiosoundings launched in situ and the MODTRAN 4 code. Water vapour for
The test of the single-channel method is shown in Fig. 10a the AGRISPECTRA campaign has been obtained from the AERONET network
and measured with a sunphotometer in the “El Arenosillo” site, around 250km
for the SPARC/EAGLE and in Fig. 10b for AGRISPECTRA, far from the field. Total content has been rescaled using the MODTRAN 4 and
with RMSE values of 1.6 K and 1.9 K, respectively. The Bias assuming a midlatitude summer atmosphere in order to obtain the value for an
has been calculated as the mean value for the difference altitude of 980m.
112 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

(a)
330
BIAS = 1.1 K
325 St. Dev. = 1.1 K
RMSE = 1.6 K

LST single-channel (K)

320

315

310

305

300

295 Low Flight (975m AGL)

High Flight (2745m AGL)
290
290 295 300 305 310 315 320 325 330
LST in situ (K)
(b)
320
BIAS = 1.7 K
St. Dev. = 0.8 K
RMSE = 1.9 K

315
LST single-channel (K)

310

305
Flight 1 (7:30 GMT)
Flight 2 (9:30 GMT)
Flight 3 (12:30 GMT)

300
300 305 310 315 320
LST in situ (K)

Fig. 10. Comparison between the land surface temperature retrieved with the single-channel method and the one measured in situ in the framework of the (a) SPARC/
EAGLE and (b) AGRISPECTRA field campaigns.

results have been obtained with split-window and single- retrieval. Differences between temperatures obtained from at-
channel algorithms, the split-window technique provides better sensor data and the ones obtained after compensation of
results over a world-wide scale, overall in wet atmospheres atmospheric and emissivity effects range between 2 and 5K.
(w > 3 g/cm2), whereas single-channel methods only provide The lowest differences (< 2K) are obtained over vegetated areas
good results for low atmospheric absorption (w < 2 g/cm2), as and water (dark tones in the image), in which the emissivity is
has been stated by Sobrino and Jiménez-Muñoz (2005). near to 1 and the differences are only due to the atmospheric
Fig. 12 shows the results obtained with the TES method in effect. This example shows the importance of the atmospheric
the framework of the SPARC/EAGLE campaign. This method and emissivity correction in order to obtain accurate values
provides the best estimation of the LST, with a RMSE of 1.4 K. of LST.
The Bias is similar to the previous algorithms, around 1 K. This
result suggests that from sensors providing multispectral 6. Summary and conclusions
thermal data (with at least four thermal bands) it is possible to
retrieve the LST with slightly better accuracy. However, the Land surface temperature (LST) is a key parameter in many
main constraint is that the atmospheric correction needs to be environmental studies, as energy balance, evapotranspiration,
well controlled. global circulation models, and vegetation monitoring, among
Finally, Fig. 13 shows the AHS image (acquired on 15 July others. In the last years, different airborne sensors with several
2004 at 12:43 UTC) for the at-sensor brightness temperature thermal bands have been developed and made available for
obtained from band 75, the LST image obtained with the TES imagery acquisitions. The data extracted from these sensors
method and the difference image. This difference shows the provide the opportunity of retrieving LST with a very-high
impact of the emissivity and atmospheric effects on the LST spatial resolution, which is important, for example, in the
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 113

(a)
330
BIAS = 1.2 K
325 St. Dev. = 1.5 K
RMSE = 1.9 K

320

LST split-window (K)

315

310

305

300

295 Low flight (975m AGL)

High Flight (2745m AGL)
290
290 295 300 305 310 315 320 325 330
LST in situ (K)
(b)
320
BIAS = 1.3 K
St. Dev. = 0.9 K
RMSE = 1.6 K

315
LST split-window (K)

310

305
Flight 1 (7:30 GMT)
Flight 2 (9:30 GMT)
Flight 3 (12:30 GMT)
300
300 305 310 315 320
LST in situ (K)

Fig. 11. Comparison between the land surface temperature retrieved with the split-window method and the one measured in situ in the framework of the (a) SPARC/
EAGLE and (b) AGRISPECTRA field campaigns.

330
BIAS = 1.0 K
325 St. Dev. = 1.0 K
RMSE = 1.4 K

320
LST TES (K)

315

310

305

300

295 Low Flight (975m AGL)

High Flight (2745m AGL)
290
290 295 300 305 310 315 320 325 330
LST in situ (K)

Fig. 12. Comparison between the land surface temperature retrieved with the TES method and the one measured in situ in the framework of the SPARC/EAGLE field
campaign.
114 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115

atmospheric absorption during the flights, with total atmospher-

ic water vapour contents lower than 1.8 g/cm2 for the high
flights (2745 m above ground level) and lower than 1g/cm2 for
the low flights (945 m above ground level). With higher water
vapour contents and, therefore, with higher atmospheric
absorption, the split-window algorithms are expected to provide
better results, whereas single-channel algorithms become
almost unusable (Sobrino & Jiménez-Muñoz, 2005). The TES
method, developed originally for the ASTER sensor by
Gillespie et al. (1998), was applied to AHS bands from 75 to
79, and it was also tested with simulated data, with a
RMSE < 1.1K. The TES method provided the best result in
the validation with ground truth data, with a RMSE < 1.4 K. It
should be noted that this method also provides land surface
emissivities, which are assumed to be known a priori in the
single-channel and split-window methods, despite that the
emissivity spectra obtained with the TES have not been tested or
validated. According to the results obtained in this study, the use
298 318 300 324 2 5
of multispectral thermal data leads to a more accurate values of
(a) (b) (c) surface temperatures, albeit slightly, and it also solves the
problem of the emissivity uncertainty. However, an accurate
Fig. 13. (a) At-sensor brightness temperature obtained from the AHS band 75 atmospheric correction should be performed to the sensor data
(T75), (b) land surface temperature (LST) obtained with the TES method and (c)
difference image (ΔT = LST − T75). The AHS image was acquired on 15 July
in order to obtain the good results pointed out.
2005 over the Barrax test site (see Fig. 1) at 12:43 UTC and the spatial resolution
is 7m. Temperature values are given in K. Acknowledgements

context of precision agriculture and vegetation stress monitor- We thank the European Union (EAGLE, project SST3-CT-
ing. The airborne hyperspectral scanner (AHS) sensor, with 10 2003-502057), the Ministerio de Ciencia y Tecnología
thermal bands (from 71 to 80) covering the range between 8 and (DATASAT, project ESP2005-07724-C05-04), and the Europe-
13 μm, is an example of the new generation of airborne sensors an Space Agency (SPARC, project 18307/04/NL/FF) for the
with multispectral thermal infrared capabilities. financial support. Funding support provided for the Agrispectra
The feasibility of retrieving LST from the 10 AHS thermal project AGL2002-04407-C03 from the Spanish MCyT is
bands has been analyzed in the framework of the SPARC and acknowledged. We also thank Don Sabol (University of
EAGLE field campaigns, which took place simultaneously in Washington) and Cindy Grove (Jet Propulsion Laboratory) for
the Barrax (Albacete, Spain) test site, and in the framework of laboratory measurements of soil spectra.
the AGRISPECTRA project, which took place in an orchard
crop field in Córdoba (Spain), all of them in July 2004. For this References
purpose, the single-channel, split-window, and TES (tempera-
Balick, L. K., Jeffery, C. A., & Henderson, B. G. (2003). Turbulence-induced
ture and emissivity separation) methods have been adapted to
spatial variation of surface temperature in high-resolution thermal IR
the AHS characteristics. Previously, the at-sensor brightness satellite imagery. Proceedings of SPIE, vol. 4879 (pp. 221–230). Crete,
temperatures extracted from the AHS thermal bands were Greece, SPIE—the International Society for Optical Engineering.
compared with in situ measurements carried out over different Becker, F., & Li, Z. -L. (1990a). Temperature independent spectral indices in
plots (bare soil, water, grass, and corn) in order to assess the thermal infrared bands. Remote Sensing of Environment, 32, 17−33.
optimal selection of bands for LST retrieval. This comparison Becker, F., & Li, Z. -L. (1990b). Towards a local split window method over land
surfaces. International Journal of Remote Sensing, 11, 369−394.
has shown that the AHS sensor had a calibration problem with Berk, A., Anderson, G. P., Acharya, P. K., Chetwynd, J. H., Bernstein, L. S.,
bands 71, 72, and 73. Bands from 75 to 79, located in the Shettle, E. P., et al. (1999). MODTRAN4 user's manual. Hanscom AFB,
atmospheric window between 10 and 12μm, seem to agree MA: Air Force Research Laboratory.
better with in situ data, so these bands were finally selected. The Dash, P., Göttsche, F. -M., Olesen, F. -S., & Fischer, H. (2002). Land surface
single-channel method, which only uses one thermal band, was temperature and emissivity estimation from passive sensor data: Theory and
practice—current trends. International Journal of Remote Sensing, 23(13),
applied to AHS band 75 because this band showed the highest 2563−2594.
atmospheric transmissivity. The split-window algorithm, which Deschamps, P. Y., & Phulpin, T. (1980). Atmospheric correction of infrared
used a combination between two thermal bands, was applied to measurements of sea surface temperature using channels at 3.7, 11 and
AHS bands 75 and 79, because this combination showed the 12 μm. Boundary-Layer Meteorology, 18, 131−143.
lowest error on the LST. Both algorithms were validated using Gillespie, A., Rokugawa, S., Matsunaga, T., Cothern, J. S., Hook, S., & Kahle,
A. B. (1998). A temperature and emissivity separation algorithm for
ground truth data, showing similar results, with root mean advanced spaceborne thermal emission and reflection radiometer (ASTER)
square errors (RMSE) lower than 2 K. The good results obtained images. IEEE Transactions on Geoscience and Remote Sensing, 36,
with the single-channel method are attributed to the low 1113−1126.
 J.A. Sobrino et al. / Remote Sensing of Environment 102 (2006) 99–115 115

Jiménez-Muñoz, J. C., & Sobrino, J. A. (2003). A generalized single-channel Sepulcre-Cantó, G., Zarco-Tejada, P.J., Jiménez-Muñoz, J.C., Sobrino, J.A., de
method for retrieving land surface temperature from remote sensing data. Miguel, E., Villalobos, F.J., in press. Within-field thermal variability
Journal of Geophysical Research, 108(D22). doi:10.1029/2003JD003480. detection as function of water stress in Olea europaea L. orchards with high-
Jiménez-Muñoz, J. C., & Sobrino, J. A. (2005). Atmospheric water vapour spatial remote sensing imagery. Agricultural and Forest Meteorology
content retrieval from visible and thermal data in the framework of the (available online).
DAISEX campaigns. International Journal of Remote Sensing, 26(15), Sobrino, J. A., Caselles, V., & Coll, C. (1993). Theoretical split window
3163−3180. algorithms for determining the actual surface temperature. Il Nuovo
Kerr, Y. H., Lagouarde, J. P., Nerry, F., & Ottlé, C. (2004). Land surface Cimento, 16, 219−236.
temperature retrieval techniques and applications: Case of AVHRR. In D. A. Sobrino, J. A., Coll, C., & Caselles, V. (1991). Atmospheric correction for land
Quattrochi, & J. C. Luvall (Eds.), Thermal remote sensing in land surface surface temperature using NOAA-11 AVHRR Channels 4 and 5. Remote
processes (pp. 33−109). Florida, USA: CRC Press. Sensing of Environment, 38, 19−34.
Kondratyev, K. Y. (1969). Radiation in the atmosphere. New York: Academic Sobrino, J. A., & Jiménez-Muñoz, J. C. (2005). Land surface temperature
Press. retrieval from thermal infrared data: An assessment in the context of the
McClain, E. P., Pichel, W. G., & Walton, C. C. (1985). Comparative Surface Processes and Ecosystem Changes Through Response Analysis
performance of AVHRR-based multichannel sea surface temperatures. (SPECTRA) mission. Journal of Geophysical Research, 110(D16103).
Journal of Geophysical Research, 90(11), 587−601. doi:10.1029/2004JD005588.
McMillin, L. M. (1975). Estimation of sea surface temperature from two infrared Sobrino, J. A., Jiménez-Muñoz, J. C., El-Kharraz, J., Gómez, M., Romaguera,
window measurements with different absorption. Journal of Geophysical M., & Sòria, G. (2004). Single-channel and two-channel methods for land
Research, 80, 5113−5117. surface temperature retrieval from DAIS data and its application to the
Moreno, J., Calera, A., Caselles, V., Cisneros, J. M., Martínez-Lozano, J. A., Barrax test site. International Journal of Remote Sensing, 25(1), 215−230.
Meliá, J., et al. (2001). The measurement programme at Barrax. DAISEX Sobrino, J. A., Li, Z. -L., Soria, G., & Jiménez, J. C. (2002). Land surface
final results workshop, SP-499 (pp. 43−51). Noordwijk (Holland): ESA temperature and emissivity retrieval from remote sensing data. Recent
Publications Division. Research Developments on Geophysics, 4, 21−44.
Nerry, F., Labed, J., & Stoll, M. P. (1990). Spectral properties of lands surfaces Sobrino, J. A., Li, Z. -L., & Stoll, M. P. (1993). Impact of the atmospheric
in the thermal infrared band: Part II. Field method for spectrally averaged transmittance and total water vapor content in the algorithms for estimating
emissivity measurements. Journal of Geophysical Research, 95, satellite sea surface temperatures. IEEE Transactions on Geoscience and
7027−7044. Remote Sensing, 31(5), 946−952.
Prata, A. J. (1993). Land surface temperatures derived from the AVHRR and Sobrino, J. A., Li, Z. -L., Stoll, M. P., & Becker, F. (1994). Improvements in the
ATSR: 1. Theory. Journal of Geophysical Research, 89(16), 689−702. split-window technique for land surface temperature determination. IEEE
Price, J. C. (1984). Land surface temperature measurements from the split Transactions on Geoscience and Remote Sensing, 32, 243−253.
window channels of the NOAA 7 AVHRR. Journal of Geophysical Sobrino, J. A., & Raissouni, N. (2000). Toward remote sensing methods for land
Research, 89, 7231−7237. cover dynamic monitoring. Application to Morocco. International Journal
Qin, Z., Karnieli, A., & Berliner, P. (2001). A mono-window algorithm for of Remote Sensing, 21, 353−366.
retrieving land surface temperature from Landsat TM data and its application Sun, D., & Pinker, R. T. (2003). Estimation of land surface temperature from a
to the Israel–Egypt border region. International Journal of Remote Sensing, Geostationary Operational Environmental Satellite (GOES-8). Journal of
22(18), 3719−3746. Geophysical Research, 108(D11). doi:10.1029/2002JD002422.
Quattrochi, D. A., & Luvall, J. C. (1999). Thermal infrared remote sensing data Wan, Z., & Li, Z. -L. (1997). A physics-based algorithm for retrieving land
for analysis of landscape ecological processes: Methods and applications. surface emissivity and temperature from EOS/MODIS data. IEEE Transac-
Landscape Ecology, 14(6), 577−598. tions on Geoscience and Remote Sensing, 35(4), 980−996.
Quattrochi, D. A., & Luvall, J. C. (2004). Thermal remote sensing in land
surface processes (pp. 440). Boca Raton, FL, USA: CRC Press.
Scott, N. A., & Chedin, A. (1981). A fast line by line method for atmospheric
absorption computations: The authomatized atmospheric absorption atlas.
Journal of Meteorology, 20, 802−812.