AMTAtmospheric Measurement TechniquesAMTAtmos. Meas. Tech.1867-8548Copernicus GmbHGöttingen, Germany10.5194/amt-8-2981-2015Kalman filter physical retrieval of surface emissivity and temperature
from SEVIRI infrared channels: a validation and intercomparison studyMasielloG.SerioC.carmine.serio@unibas.itVenafraS.LiuzziG.GöttscheF.https://orcid.org/0000-0001-5836-5430F. TrigoI.https://orcid.org/0000-0001-8640-9170WattsP.Scuola di Ingegneria, Università della Basilicata,
Potenza, ItalyKarlsruhe Institute of Technology (KIT),
IMK-ASF, Karlsruhe, GermanyInstituto Portugues do Mar e da Atmosfera IP, Land SAF,
Lisbon, PortugalEuropean Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), Darmstadt, GermanyC. Serio (carmine.serio@unibas.it)29July2015872981299724March201523April201510July201513July2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://amt.copernicus.org/articles/8/2981/2015/amt-8-2981-2015.htmlThe full text article is available as a PDF file from https://amt.copernicus.org/articles/8/2981/2015/amt-8-2981-2015.pdf
A Kalman filter-based approach for the physical retrieval of surface
temperature and emissivity from SEVIRI (Spinning Enhanced Visible and
Infrared Imager) infrared observations has been developed and validated
against in situ and satellite observations. Validation for land has been
provided based on in situ observations from the two permanent stations at
Evora and Gobabeb operated by Karlsruhe Institute of Technology (KIT) within
the framework of EUMETSAT's Satellite Application Facility on Land Surface
Analysis (LSA SAF). Sea surface retrievals have been intercompared on a broad
spatial scale with equivalent satellite products (MODIS, Moderate
Resolution Imaging Spectroradiometer, and AVHRR, Advanced Very High
Resolution Radiometer) and ECMWF (European Centre for Medium-Range Weather
Forecasts) analyses. For surface temperature, the Kalman filter yields a root
mean square accuracy of ≈±1.5∘C for the two land sites
considered and ≈±1.0∘C for the sea. Comparisons with
polar satellite instruments over the sea surface show nearly zero temperature
bias. Over the land surface the retrieved emissivity follows the seasonal
vegetation cycle and permits identification of desert sand regions using the
SEVIRI channel at 8.7 µm due to the strong quartz reststrahlen bands
around 8–9 µm. Considering the two validation stations, we have found
that emissivity retrieved in SEVIRI channel 10.8 µm over the gravel
plains of the Namibian desert is in excellent agreement with in situ
observations. Over Evora, the seasonal variation of emissivity with
vegetation is successfully retrieved and yields emissivity values for green
and dry vegetation that are in good agreement with spectral library data. The
algorithm has been applied to the SEVIRI full disk, and emissivity maps on
that global scale have been physically retrieved for the first time.
Introduction
In the authors exploited the high
temporal resolution of data acquisition by geostationary satellites
and their capability to resolve the diurnal cycle
to develop a Kalman filter (KF) approach e.g.
for the simultaneous retrieval of surface temperature, Ts, and emissivity, ϵ.
The case of SEVIRI (Spinning Enhanced Visible and
Infrared Imager) Meteosat-9 high-rate level 1.5 image data was examined.
It was shown that the KF approach results in an
algorithm which does not need to increase the dimensionality of the
data space, e.g. because of time accumulation of observations,
while preserving the temporal resolution of the geostationary instrument (15 min for
SEVIRI). The reliability and quality of the approach has been further
demonstrated in by applying the methodology
to study the diurnal emissivity dynamics in bare versus biocrusted
sand dunes in a coastal desert region.
The present study mainly focuses on the KF implementation and
comparison of its results with in situ data, and other similar
satellite products. As previously mentioned, the KF approach follows the basic methodology
developed by . The implementation used in this study deals
with surface parameters alone, namely (Ts, ϵ). For this retrieval problem we apply a strictly
temporal only method – that is, we do not consider spatial constraints.
Despite this simplification, the present KF approach is a new-concept algorithm in the broad
research area of temperature–emissivity (Ts,ϵ) retrieval from satellite, which to date
relies on statistical retrieval algorithms e.g. and static
physical schemes e.g.. A distinctive aspect of our approach is that it
is a dynamically, physically based approach, which makes it unique at this present time.
An in-depth assessment of the expected retrieval performance for land and sea surface and its
dependence on the tuning parameters and settings of the present KF implementation has been performed
and presented in and , which the reader is referred to for
further details. In addition, KF emissivity products have been intercompared with IASI
(Infrared Atmospheric Sounding Interferometer; ) retrievals in and .
The present study aims at complementing the results presented in
and assessing the capability of the time dimension KF
approach to provide accurate retrievals at the SEVIRI full-disk scale and in the case of
long time periods, which can include
large data voids because of, for example, clouds.
Towards this objective, we have set up a study to validate the KF approach on
a broad spatio-temporal scale, from individual SEVIRI pixels to the SEVIRI
full disk, and from days to the whole year. Validation for land has been
performed based on 1-year in situ observations from the two permanent
stations near Evora (Portugal) and Gobabeb (Namibia) operated by Karlsruhe
Institute of Technology (KIT). For sea surface, retrievals have been
intercompared with other satellite products, namely the Moderate Resolution
Imaging Spectroradiometer (MODIS) and the Advanced Very High Resolution
Radiometer (AVHRR). Finally, ECMWF (European Centre for Medium-Range Weather
Forecasts) analyses for sea surface temperature have been compared to
retrieved SEVIRI products on a timescale spanning from hours to years.
The paper is organized as follows. Section is devoted to the presentation of SEVIRI and ancillary data. The
basic methodology for the KF is presented in Sect. . The
results from the validation exercise are reported in Sect.
and conclusions are drawn in Sect. .
Data
For the various case studies, which we will describe below, SEVIRI
observations (Meteosat-9 high-rate SEVIRI level 1.5 image data) have been
used. For cloud masking we have used the operational Meteosat Second
Generation (MSG) products cloud mask. The SEVIRI imager on board Meteosat-9
allows for a complete image scan (full Earth scan) once every 15 min period
with a spatial resolution of 3 km for 12 channels (8 in the infrared), over
the full disk covering Europe, Africa and part of South America. Infrared
channels range from 3.9 to 12 µm, and their definition in terms of
channel number is given in Table . In order to identify and work with clear-sky radiances, the SEVIRI operational cloud mask has been used in the analysis. SEVIRI radiances and associated cloud mask
were downloaded from the EUMETSAT Data Centre through the Unified
Meteorological Archive Facility (UMARF).
For the present study the SEVIRI full disk is defined in such a way
to include viewing zenith angles (VZA) below or equal to 70∘.
At larger angles, emissivity tends quickly to zero, whereas the
SEVIRI pixel size increases. The forward model we use in the
retrieval scheme assumes a plane parallel atmosphere and could be
unsuitable for large zenith angles. Also, the dependence of land emissivity on the
viewing angle could be an issue at VZA beyond the
limit of 70∘. It also needs to be stressed that the limit of
70∘ is reasonable in order to work in the plane parallel
approximation e.g..
The area of the SEVIRI disk included within the interval VZA =±70∘ is shown in Fig. . This region contains
some 9×106 SEVIRI pixels.
SEVIRI full disk according to the rule VZA (viewing
zenith angle) ≤ 70∘. The two dots over land give the location of KIT validation stations. The dot over
the Mediterranean sea is the location of the SEVIRI pixel used to compare the
retrieved Ts against AVHRR data.
[t]
Evora (Portugal) and Gobabeb (Namibia) validation stations. The geographic location is shown
on the left, and the landscape around the
validation stations is shown on the right.
Case study definition
For the purpose of validation, to assess the reliability and stability of the scheme and,
moreover, its capability to run on the global scale, we have defined a series of case studies, which are now
presented.
Single SEVIRI pixels spanning an entire year
Individual SEVIRI pixels have been considered which correspond to two
validation stations: (1) Evora station in Portugal (38.55∘ N, 8.01∘ E) and (2) Gobabeb station in Namibia (23.55∘ S and
15.18∘ E). The geographical location of the two stations is shown in
Fig. .
We have nearly continuous records available of in situ measurements of surface temperature for both stations.
The year 2010 has been used for the intercomparison exercise.
The two stations are operated by KIT and are part of the EUMETSAT's Satellite Application Facility on Land Surface
Analysis (LSA SAF). Evora station is in the
temperate Mediterranean climate, with a land cover of cork–oak trees and grass.
Gobabeb station is in the arid Namibian
desert climate and is located on a flat and homogeneous gravel plain .
The core instruments of the two stations are self-calibrating, chopped
radiometers (Heitronics KT15.85 IIP, 9.6–11.5 µm) which measure the
radiation from the relevant components, e.g. grass, soil, tree, shadow, and
sky once per minute (see , and ). In the case of Evora station,
where local temperature measurements of sunlit/shaded ground and tree canopy
may present significant differences, the pixel “in situ” temperature is
reconstructed using fixed land cover fractions obtained by classifying high-resolution satellite data (IKONOS), which then served to weigh the
radiometric measurements of the endmembers (tree 32 %, background 68 %). We
also note that the newest approach (not available for the year 2010) uses
dynamic cover fractions from a geometric model e.g..
Southern Italy region (boxed area) considered for the case study over
sea surface. The cyan box is the SEVIRI pixel used to compare the
retrieved Ts against AVHRR data.
As mentioned above, the in situ observations of surface temperature refers to the whole year of
2010. The data record is nearly continuous, although there can be data voids due to weather conditions or
instrument maintenance.
For the two stations we also have available the SEVIRI LSA SAF surface temperature derived
with the algorithm developed by and .
Regional case study
This case study has been set up to check possible spatial biases and the
stability of the scheme when processing time series of radiance data points
of long time period extent, e.g. 1 year. SEVIRI infrared radiances are
recorded and available every 15 min. However, cloudy radiances, if detected,
can be skipped during the KF run, and therefore the actual processed SEVIRI
radiance record could be made of data points which do not correspond to
equally spaced times. In principle, cloud fields can produce large data
voids, which could be detrimental to KF stability because the radiance data
record is lacking time contiguity and continuity.
With this in mind, an additional case study has been set up, for which we
have acquired SEVIRI data for the whole year of 2013. The target area include
a relatively large region (both ocean and land) of southern Italy (see Fig. ).
For the purpose of comparison, MODIS sea skin temperature and AVHRR sea
surface temperature retrievals were acquired for the same period and
location. The level 2 MODIS data products are produced and distributed by
NASA Goddard Space Flight Center's Ocean Data Processing System (ODPS). The
data are available from the website
http://oceandata.sci.gsfc.nasa.gov. Both the AQUA and TERRA satellites have
been used. The MODIS product used in the present study is MOD28 Sea Surface
Temperature 5-Min L2 Swath 1 km.
The AVHRR data are not direct satellite observations. In fact, these data are
better referred to as AVHRR OI (optimal interpolation) SST (sea surface
temperature) analysis , because they are the results of an
optimal interpolation scheme which combines AVHRR, buoy and ship data to form
daily averages. The data are available at the website
ftp://eclipse.ncdc.noaa.gov/pub/OI-daily-v2/NetCDF/ and they are
provided on a regular grid of 0.25∘×0.25∘.
For the whole target area shown in Fig. , we have also acquired
ECMWF analysis products for the sea skin temperature at the canonical hours
0:00, 6:00, 12:00 and 18:00 UTC. ECMWF model data points are provided on a
0.125∘×0.125∘ regular grid. They have been space-collocated
just by overlapping the SEVIRI spatial grid to that of ECMWF (see, for
example, . Once the space collocation has been performed, the ECMWF
Ts corresponding to a given SEVIRI observation, recorded at time t, is
obtained through a simple linear interpolation using the ECMWF Ts values
at the canonical hours.
Full-disk case study
For the purpose of checking the feasibility of the scheme to run at the global scale
we have defined a full-disk case study for the month of November 2007.
The area of the full disk covered corresponds to VZA ≤70∘
(see Fig. ) and consists of 9 046 159 pixels, of which
3 581 915 are over land and 5 464 244 over sea surface.
Ancillary data: emissivity
The application of the KF approach relies on proper a priori information
about emissivity. This is needed to properly build up the background, state
vector and related covariance matrix, which are used within the KF retrieval
approach.
For land surface, emissivity is derived from the University of
Wisconsin Baseline Fit Global Infrared Land Surface Emissivity
Database (UW/BFEMIS database, e.g. http://cimss.ssec.wisc.edu/iremis/)
. The database
has a spatial resolution of 0.05∘ and a time step of 1
month, which is enough to include the expected seasonality of
surface emissivity. UW/BFEMIS covers the years 2003–2013; therefore
it is also capable of providing time and spectral
cross-correlation among channel emissivities.
The UW/BFEMIS database has been re-mapped to the SEVIRI channels and spatial
grid mesh and then used to define a background for the channel emissivity
(state vector and its covariance) which depends on time (monthly resolution)
and geographic location (SEVIRI pixel resolution). The re-mapping involves a
high-spectral-resolution algorithm which is first applied to the 10-hinge-point UW/BFEMIS emissivities to generate the emissivity spectrum, which, in
turn, is convolved with he SEVIRI spectral response. Details of this
procedure can be found in . In passing, we note that
the algorithm to transform UW/BFEMIS emissivities to high spectral resolution
was first proposed and developed by .
For sea surface, the emissivity is defined and derived according to
Masuda's emissivity model . We have developed a
look-up table with sea surface emissivity over the spectral range
500 to 3000 cm-1 and a spectral resolution of 0.25 cm-1.
The emissivity has been calculated for view angles
(vertical zenith angle) ranging from 0 to 89∘ (step
size of 1∘) and wind speed from 0 to 15 m s-1 (step size 1 m s-1). For a given VZA, the emissivity state vector is calculated for
an average wind speed of 5 m s-1, whereas the values corresponding to
the other wind speeds are used to derive the background
covariance. The high-spectral-resolution emissivity is convolved
with the SEVIRI instrumental spectral response function (ISRF) to yield
the SEVIRI channel emissivities.
Ancillary data: ECMWF analysis
The ECMWF analysis, at the four canonical hours, for the atmosphere and
surface is used to initialize the KF and to provide information
for the atmospheric state vector; temperature profile, T; water vapour
mixing ratio profile, Q; and ozone mixing ratio profile, O. The surface
and atmospheric parameters are directly downloaded from the ECMWF MARS
(Meteorological Archival and Retrieval System) archive and consist of surface
temperature and pressure, profiles of temperature, water vapour and ozone.
The analysis is available on a horizontal grid mesh of 0.125∘× 0.125∘.
The atmospheric profiles are obtained on either 91 pressure levels
(until 25 July 2013) or 137 pressure levels (from 26 June 2013). ECMWF
analysis is linearly interpolated to the SEVIRI time–space grid mesh before
it is used to initialize the KF. The procedure is the same as that
used for the surface temperature (see end of
Sect. ).
Methodology: implementation of the Kalman filter for (Ts,ϵ)
The basic KF implementation performs a simultaneous mathematical inversion of
the radiative transfer equation for (Ts,ϵ). The retrieval
algorithm has been developed in such a way as to provide a suitable prototype
for use at a satellite data processing centre for a range of applications
involving remote sensing of the surface. In fact, the system can ingest
SEVIRI data and ancillary MARS/ECMWF analyses in their native format. For the
sake of clarity, here we limit ourselves to showing the KF basic equations which
apply to the (Ts,ϵ) retrieval problem.
The retrieved state vector, v, is made up of m
(m= 8) SEVIRI infrared channel emissivities and the surface
temperature,
v=(e1,e2,…,em,Ts)T,
where the superscript T stands for transpose and e is the logit-transformed emissivity
e=logϵ1-ϵ.
The logit transform ensures that the we deal with emissivity correctly
constrained in its physical variability range of 0–1. The transform has been
successfully used with emissivity retrieval for IASI and
SEVIRI .
In principle the scheme can be applied to the m=8 infrared
channels of the SEVIRI imager, which are listed in Table .
However, effective results are expected for the three
atmospheric window channels (7, 9 and 10) which are less sensitive to
atmospheric parameters, namely temperature, water vapour and ozone.
For the shortwave window channel at 3.9 µm, retrieval is only
recommended at night-time to avoid solar contamination.
In the following, to simplify the exposition we will assume that the times
are indexed by integers, t=1,2,…, although handling unequally
spaced times does not add any fundamental difficulty. With this in mind, the
non-linear KF estimate or analysis, v^t of
vt, at a generic time t is given by e.g.x^t=xta+KtTSε-1Kt+Sa-1KtTSε-1yt-Ktxta,
where we have posed
x^t=v^t-vt;fgyt=Rt-F(vt;fg)xta=vta-vt;fg
and where the subscripts a and fg indicate background and first-guess parameters, respectively, at time t. R is the radiance vector
and F is the forward model.
Again, with reference to Eqs. () and (), Kt is the
Jacobian at time t,
Kt=∂F∂vv=vt;fg.
The a posteriori covariance of the estimate Eq. () is given by
S^t=KtTSε-1Kt+Sa-1.
The analysis is propagated forward to time t+1 to obtain the
forecast according to
v^t+1f=Mv^t,
which has covariance given by
S^t+1f=MS^tMT+Sη.
The dynamical model operator, M, is assumed to be
the identity matrix . In this way, the retrieval scheme
assumes a persistence model for both temperature and emissivity.
To complete the description of the scheme, the covariance matrices
Sε, Sa, and Sη are the observational covariance matrix, the background
covariance matrix and the stochastic term covariance matrix.
KF operates sequentially, which means that the updated analysis at a
given time t+1 is based on the new observations at time t+1 and the state
vector estimate at the previous time t, which in our case (because M is the identity) is identified with the forecast (see Eq. ). Update and forecast steps are repeated as time advances and
new observations arrive.
It can be shown e.g. that the analysis (Eq. ) is the minimizer
of the linearized quadratic form or cost function S given by
S=minx12yt-KtxtTSε-1yt-Ktxt+12xt-xtaTSa-1xt-xta,
which states that the KF Eqs. () and ()
are the same as the equations of optimal estimation .
Optimal Estimation can be regarded as a particular case of KF.
e.g..
In passing, we also note here that the retrieval scheme above is non-linear
because at each time t we have to linearize the forward model and iterate
the solution until the cost function (see Eq. ) is reduced
below a given threshold e.g.. Under linearity, the
value of twice the quadratic S (Eq. ) at the minimum is
distributed as a χ2 variable with m degrees of freedom
. A χ2 threshold, χth2 at the 3σ
confidence interval, is given by χth2=m+32m: therefore the
iterative procedure is stopped when
χ2=2×S≤χth2.
Linearization of the forward model is done around a first-guess state vector,
which can depend on time and should not be confused with the background or
analysis. Because of the linearization and iteration steps the scheme is non-linear. Non-linear KF schemes are normally referred to as extended KF.
However, for the sake of brevity, we will continue to refer to the scheme
above as KF.
Yearly SEVIRI 12 µm channel data record for the validation station of Evora: (a) original radiance data record and (b) after
removing cloudy radiances.
KF parameter settings
An important aspect of a given KF implementation is the setting of the many
parameters involved within the retrieval scheme. The settings of the main
parameters is here summarized for the benefit of the reader. An in-depth
assessment and analysis of the sensitivity of the retrieval to these parameters can
be found in , whereas an analysis of the sensitivity to the state
operator M and implications by setting it to the identity
operator are analysed and discussed in and .
To begin with, we clarify that Sε is a static parameter
and is set equal to the SEVIRI radiometric noise.
The covariance operator Sa has to be initialized
at time t=0. At time t=0, the matrix Sa does not consider
cross-correlation between emissivity and surface temperature – that is, it is
of the form
Saϵ;0,00,σTs;02,
where Saϵ;0 is the uncertainty associated with the
emissivity vector and σTs;02 is the uncertainty associated with
the surface temperature. For emissivity, as already mentioned, Saϵ;0 is obtained from the UW/BFEMIS database (land) or the
Masuda's emissivity model (sea surface). The initial value
σTs;02=1 K2 is used for the surface temperature of either land
or
ocean. According to , these uncertainties do not need to be
accurately prescribed at t=0 because at a later time
Sa evolves according to Eq. – that is, it is
identified with the forecast covariance. We note that
at later time Sa may also include cross-correlation between Ts and ϵ.
Another important parameter is the matrix Sη, because
it trades off between a retrieval dominated by either
the dynamical model (a persistence equation in our implementation) or the observations.
Sη is the covariance of the stochastic term, and in our
implementation it is a static parameter (not evolved with time). It takes the
form
Sη=f-2Saϵ;0,00,σTs;η2,
where f and σTs;η2 are two tuning parameters. For the work
shown here, we use for land f=5 and σTs;η2=1 K2, whereas
for sea surface we have f=5 and σTs;η2=0.1 K2.
The forward model
One of the key aspects of the KF scheme is the use of a physical forward
model which solves the radiative transfer equation in the form needed for
the present application. The forward model implemented with the baseline
version of KF is the so-called σ-SEVIRI code .
Also, for potential applications to hyper-spectral sounders, σ-SEVIRI
has been designed as a monochromatic forward model, based on a look-up table
for the optical depth. The sampling along the wave number axis used to
develop and implement the look-up table has been optimized for SEVIRI
e.g.. Furthermore, the forward model can deal with
Lambertian and specular reflecting surfaces. Again, this is an important
aspect when dealing with the emissivity retrieval of land and ocean
emissivity. To date, considering the SEVIRI instrument, this capability of
running in the infrared wave number range with a Lambertian model is unique to
σ-SEVIRI.
The code σ-SEVIRI computes analytical Jacobian derivatives for any
state vector parameter. Spectral monochromatic radiance and Jacobians are
reduced to the SEVIRI spectral resolution through convolution with the SEVIRI
ISRF.
The code σ-SEVIRI is a legacy of σ-IASI
, a radiative transfer model developed for IASI.
Over the past year, the model σ-IASI has been largely validated with
aircraft and satellite high spectral infrared observations
e.g..
Results: validation and comparison to similar satellite-derived products
This section is devoted to the presentation and discussion of the case
studies we have defined in order to check the retrieval performance and stability of
the KF approach.
Before presenting the results, we remark that SEVIRI radiances are processed
at their higher rate of 15 min. However, in the case of detected cloudiness, radiances are skipped and, therefore, the time lag between two consecutive observations may become several hours or even days. This is
exemplified in Fig. , which shows the yearly data record for the
Evora station corresponding to the SEVIRI channel at 12 µm. Once the
original, cloudy data set has been screened for clouds, we are left with a
data record sampled at unequally times. It can be seen from Fig.
that in wintertime (especially in February) the time lag between two
consecutive data points can be as large as several days.
Evora station. Scatter plot of SEVIRI surface temperature retrieval and in situ
observations at Evora for 2010 (a) KF (this study retrieval) and (b) LSA SAF
retrieval. Note: R2 is the linear correlation coefficient.
Thus, in the case of large data voids we could experience a lack of continuity and
the problem of whether this lack of continuity could affect the stability
of the algorithm arises. As mentioned above, the assessment of this stability is one of the
main objective of the present paper and will be discussed in the following of
this section on the basis of the various case studies we have set up.
To begin with, we discuss the results for the two in situ validation
stations.
Evora station, year 2010. Monthly mean surface temperature difference (SEVIRI–in situ) and related SD. The low values in
February are due to few data points being available for this month.
Evora station
Figure a shows a scatter plot of the SEVIRI KF retrieval against
the in situ measurements of Ts, whereas Fig. b provides the
same comparison, but now of in situ Ts with the SEVIRI LSA SAF surface
temperature. The comparison is performed only for retrievals which reached
convergence according to the cost function criterion (see Eq. ).
The yearly root mean square (rms) difference of retrievals and situ Ts is
1.84 ∘C for KF and 1.91 ∘C for LSA SAF. The yearly bias is 1.04 ∘C for KF against 1.15 ∘C, showing that the KF is slightly
superior to LSA SAF. This is also confirmed from Fig. , which
shows the monthly mean (bias) and standard deviation (SD) of the difference
(KF–in situ) and (LSA SAF–in situ). In this figure the low values
corresponding to February are not statistically significant because of the
very few data points for this month (e.g. see Fig. ). For this
month only three SEVIRI data points were available because of persistent cloud
coverage. For the other months the number of data points is normally above
100. We stress again that the number of retrievals for each months is
ultimately determined by the cost function criterion (see Eq. ).
Apart from February, we see that the bias oscillates around 1 ∘C and
KF performs better than LSA SAF during the summer season. Also, from the
scatter plots of Fig. , we see that, compared to KF, the LSA SAF
bias is slightly larger at higher temperatures. This is better seen from Fig. which shows a short sequence in July 2010 of retrievals, in
situ observations and the corresponding differences. It is seen from Fig. that LSA SAF has larger bias than KF when the maximum
temperature is reached. Figure also exemplifies the stability
of the KF in the case of large data voids. A large data void occurs in
between the Julian days 218 and 220. Yet, the retrieved surface temperature
shows a stable behaviour and no important bias is seen at the gap end
points.
It should also be stressed that Evora station does not have an homogeneous land
type and coverage within the SEVIRI pixel. In situ observations are obtained
by merging together the radiant temperatures from the various components of
the composite scene. This inhomogeneity can explain some large temperature
differences (e.g. Fig. ). Normally the larger fluctuations
appear at sunrise (which corresponds to the minimum temperature), when the
shadowing effects may change quickly . Because of this
difficulty, the same accuracy of in situ data remains a problem and we cannot
really say which algorithm (KF or LSA SAF ) performs better for Ts.
However, when comparing KF to LSA SAF, it should be stressed that, unlike LSA
SAF, KF simultaneously retrieves emissivity along with temperature.
Evora station. (a) Example of Ts time series for a few days in July 2010. (b) Difference
(KF–in situ) and (LSA SAF–in situ).
Emissivity retrieval for the Evora validation station. The
retrieval has been smoothed with a moving-average filter with time windows of 3 h and 1 day. Top panel, 12 µm; middle panel, 10.8 µm;
bottom panel, 8.7 µm. Retrievals refer to the year 2010.
Comparing the emissivity retrieval for the Evora station at the three window channels.
The retrieval has been smoothed with a moving-average filter with a
time window of 1 day. Retrievals refer to the year 2010.
Scatterplot of SEVIRI surface temperature retrieval and in situ
observations at Gobabeb for 2010. (a) KF (this study retrieval) and (b) LSA SAF
retrieval. Note: R2 is the linear correlation coefficient.
Gobabeb station, year 2010. Monthly mean surface temperature difference (SEVIRI–in situ) and related SD.
Figure shows the time sequence of the retrieval for emissivity.
The retrieval has been smoothed with a moving-average filter with a time
window of 3 h in order to suppress spurious values due to undetected
cloudy radiances. Nevertheless, small-scale, random fluctuations are still
visible. In fact, these fluctuations can be further reduced if we smooth the
15 min emissivity retrieval with a moving-average filter with a time window
of 1 day. The result is once again shown in Fig. . It is
important to note that the emissivity retrieval is stable and is not affected
by the large data voids occurring during winter times.
To further intercompare the emissivity at different channels, the 1-day
smoothed emissivities are plotted together in Fig. . It can be
seen that the retrieval is sensitive to the vegetation seasonal cycle. In
fact, in summer the emissivity at 8.7 µm is definitely larger than that
at 10.8 and 12 µm, whereas the three are comparable in winter.
This phenomenon is in agreement with the different emissivity of green
(winter–spring) and dry (summer) grass e.g.. Green grass
has an emissivity which is almost constant over the spectral range 8–12 µm, whereas that of dry grass at 8.7 µm is significantly larger than
that at 10.8 and 12 µm. This green–dry grass emissivity contrast
has been used for land cover classification e.g..
The emissivity at 10.8 µm is ≈0.97 in May and ≈0.965 in
October. For comparison, the 10.8 µm
emissivity determined for 2009 by LSA SAF varies between 0.987 in May and 0.974
in October, while the corresponding UW/BFEMIS emissivity varies between 0.982
in May and 0.960 in October .
Gobabeb station
Gobabeb station has been selected mainly for the homogeneous
surface coverage, which should simplify the interpretation and
comparison of in situ measurements to retrievals .
As done for Evora station, first we show in Fig. a the
scatter plot of the KF estimated surface temperature against in situ
observations, whereas Fig. b compares in situ with SEVIRI LSA
SAF. Once again, the comparison is performed only for the retrievals which
reached convergence according to the cost function criterion. If we compare
with Fig. (corresponding to Evora station), the yearly rms
difference for the case of KF is 1.26 ∘C, which is slightly larger
than the corresponding LSA SAF, which is 1.20 ∘C.
Both KF and LSA SAF are in excellent agreement with the in situ observations.
The comparison between LSA SAF and KF suggests that the former performs
slightly better than the latter: the yearly bias is greater for KF (≈0.80∘C) than for LSA SAF (≈0.40∘C). The situation
is reversed for the SD, namely ≈0.97∘C (KF)
vs. ≈1.14∘C (LSA SAF). However, a closer look at the results
shows a sort of seasonal compensation for the case of LSA SAF. The bias
compensation is visible in Fig. and is more clearly seen from
Fig. , which shows the monthly mean and SD of
the difference SEVIRI–in situ for both KF and LSA SAF. It is seen that LSA
SAF has a slightly seasonal bias which becomes negative in May, whereas KF
shows a more uniform behaviour.
Figure shows the time sequence of the retrieval for
emissivity. The time resolution of the retrieval is 15 min (a moving-average
smoothing was applied with a window width of 3 h). There are small-scale
(day–night) variations in emissivity, which are most evident for the channel
at 8.7 µm because of the strong contrast introduced by quartz absorption
(reststrahlen effect). It is likely that these diurnal fluctuations are the
result of direct adsorption of water vapour from the atmosphere
e.g.. Due to the low emissivities in the so-called
reststrahlen bands of (dry) quartz, this effect is most pronounced around 9 µm during the dry season, which for Gobabeb coincides with winter.
Figure shows a clear-sky sequence of the retrieved pair
(Ts,ϵ) for 10 days of June 2010. It is seen that the emissivity
follows the temperature daily cycle with larger emissivity at night-time
(before sunrise), which is consistent with a daily cycle of soil moisture
driven by direct water vapour adsorption. The diurnal emissivity variation in
SEVIRI channel 8.7 µm has a peak-to-peak amplitude ≤0.015, a result
which is consistent with the findings shown by . In
contrast, the seasonal variation of emissivity is much smaller than for
vegetated soil, e.g. compare Fig. and ).
Gobabeb station. Emissivity retrieval. The
retrieval has been smoothed with a moving-average filter with a
time window of 3 h. Retrievals refer to the year 2010.
Gobabeb station surface temperature (a) and 8.7 µm
emissivity (b) retrieval for a clear-sky sequence of days in June 2010. The emissivity retrieval has been smoothed with a moving-average filter with a time window of 3 h.
Also, for Gobabeb, KF was not affected by data voids and the retrieval was
stable for Ts and ϵ (see e.g. Fig. ).
Finally, it is worth noting that the KF-retrieved emissivity for the 10.8 µm channel is in very good agreement with that estimated for the Gobabeb
gravel plain with satellite observations (MODIS and ASTER (Advanced
Spaceborne Thermal Emission and Reflection Radiometer)) and the in situ
box method approach . KF yields an estimation of 0.946 with
a variability (SD) of ±0.002, whereas the combination of
the various methods and estimates presented in gives the value of
0.944 ± 0.015. For the year 2009, the 10.8 µm emissivity determined
by LSA SAF is quasi-static at 0.948 , while the corresponding
UW/BFEMIS emissivity varies between 0.945 and 0.955
.
Regional case study
In this section we will show the results of the comparison exercise for the
case of the southern Italy target area shown in Fig. . The
comparison will deal with the sea surface, for which we know that both ECMWF
and satellite products are highly reliable.
Year 2013. Comparison of SEVIRI-retrieved skin temperature with ECMWF analysis: (a) histogram of the surface temperature difference ΔTs
(SEVIRI–ECMWF), (b) monthly mean and related SD of ΔTs, (c) Map of the yearly average surface temperature difference (SEVIRI–ECMWF) and related
SD (d).
Year 2013. Comparison of SEVIRI-retrieved skin temperature with MODIS: (a) histogram of the surface temperature difference ΔTs
(SEVIRI–MODIS),
(b) monthly mean and related SD of ΔTs, (c) map of the yearly average surface temperature difference (SEVIRI–MODIS) and related
SD (d).
Year 2013. Comparison of daily values of Ts derived from SEVIRI and AVHRR. Clockwise from top left, time series of daily mean of Ts,
scatter plot, and histogram of the SEVIRI–AVHRR difference.
Comparison with ECMWF products
The comparison with ECMWF model data for Ts shows a very good agreement
with absolute monthly mean differences below 0.4 ∘C and SD around 1 ∘C. Figure shows the histogram of
Ts differences for the whole year 2013. It is seen that the yearly average
difference is only -0.07 ∘C with a SD of 1.02
∘C. The statistics have been compiled with 2 754 238 data points. It
is worth noting that our findings testify the high reliability reached by the
ECMWF sea surface temperature product.
To check for a seasonal systematic error of the retrieval we have computed
the monthly averages and SDs of the difference between SEVIRI
and ECMWF. These are shown in Fig. b. It is seen that the mean
difference and SD tend to decrease in the summer season,
which is quite understandable because the frequency of cloudiness tends to
decrease in summer.
Figure c shows the spatial distribution of the yearly surface
temperature differences. It is seen that, apart from SEVIRI pixels close to
the coast, the difference is homogeneously zero everywhere. The variability
(SD) of the surface temperature difference does not show any important
spatial dependence as can be seen from Fig. d.
Finally, once again the results shown in this section exemplify the
resilience of the KF against data voids due to cloudiness.
Comparison with sea surface MODIS products
The comparison has been performed using the SEVIRI retrieval for Ts and
the time–space-collocated MODIS products. Also, for this case, the comparison
(Fig. ) suggests a very good agreement with a yearly mean
difference of -0.07 ∘C and a SD of 1.05 ∘C.
These two values have been obtained with a total of 3 230 710 data points.
Apart from the month of December 2013, the monthly mean difference is
normally ≈0.2∘C or below as can be seen from Fig. b. The anomalous case of December 2013 is likely due to cloud
contamination of MODIS overpasses: MODIS has a negative bias of ≈0.5∘C with respect to SEVIRI, which is uniform over the target area. This
behaviour is not seen for other months.
Figure c shows the spatial distribution of the yearly surface
temperature differences. Once again, it is seen that, apart from SEVIRI pixels
close to the coast, the difference is homogeneously zero everywhere. A good
spatial homogeneity is also seen for the variability (SD) of the surface
temperature difference as can be seen from Fig. d.
As for the comparison with the ECMWF fields, the comparison with MODIS also
evidences a good stability and reliability of the KF retrieval, despite
larger data voids due to clouds.
Comparison with AVHRR OI SST analysis
For the case of AVHRR, a comparison has been performed to gain further
insights into understanding whether the KF approach is capable of following
the seasonal cycle. The AVHRR data we are going to compare with the KF
algorithm are not direct AVHRR skin temperature values. Instead, AVHRR data
are assimilated within the OI SST scheme, yielding daily values of surface
temperature corrected for possible artifacts of the polar orbit
e.g., which, of course, cannot resolve the daily cycle.
Because of the geostationary orbit, SEVIRI products do not suffer from this
problem.
The comparison is shown in Fig. and refers to one single
SEVIRI pixel centred at lat–long coordinates (40.625∘ N, 13.875∘ E). This pixel
is located in the north-west corner of the target area shown in Fig. , within the Gulf of Naples. Figure shows the time
evolution, over the year, for the daily mean surface temperature obtained
from SEVIRI and AVHRR OI SST analysis. The shown SEVIRI KF Ts have been
obtained by daily averaging of the SEVIRI retrievals (recalling that the KF
approach provides retrieval on a time step of 15 min).
The comparison show excellent agreement as far as the dynamics of the
yearly cycle are concerned. However, the analysis also evidences a yearly
negative bias of SEVIRI of -0.30∘C, which can be compared to the value
of ≈-0.07∘C obtained with MODIS and the ECMWF analysis.
This can be explained because the AVHRR OI SST is a bulk temperature
estimate, whereas the KF Ts provides an estimation of the skin
temperature.
November 2007 map of Ts over the SEVIRI full disk.
Typical desert sand emissivity compared to the SEVIRI ISRF of the
three atmospheric window channels.
As said before, within the OI SST scheme, AVHRR satellite data undergo a
de-bias procedure based on ship and buoy direct measurements, which
transforms the satellite temperature from the skin to the surface of
the ocean . From Fig. we can see that the negative
bias is quite homogeneous during the year, because SEVIRI senses the cool
skin at the surface, whereas buoys measure the warmer layer just below the
surface. It is also important to stress that the bias we have found (-0.30 K) is consistent with the well-assessed result e.g. that the
difference between bulk and skin temperature is within ±1 K with a mean
difference ranging from 0.1 to 0.3 K.
SEVIRI full-disk maps
Finally, we come to the application of the KF approach to the SEVIRI full
disk as defined in Fig. . The full-disk retrieval exercise is
here mostly intended to show that the methodology can really be run at any
location and its stability does not depend on time–space. For this case
study, we have limited ourselves to consider the average mean field of
emissivity and Ts for November 2007.
The retrieval exercise considers land surface alone, since the case
of sea surface is rather straightforward, as demonstrated, for example, in
Sect. .
Figure shows the map for the case of surface temperature. It
is seen that the map recovers the correct latitudinal gradient and shows the
expected increase of temperature over the Sahara and the Arabian Desert. If we
consider the retrieval at the border of the ±70∘ viewing angle
circle, we cannot see any important problems for the eastern part of the
disk. The expected warmer area of the Arabian Peninsula is correctly
retrieved. However, over the western part of South America, the temperature
seems too low, which could be an effect of the SEVIRI point spread function
at this large angle. However, at this stage we do not have enough evidence to
conclude that the full-disk area should be confined to smaller viewing
angles.
As exemplified in Fig. , the SEVIRI channel at 8.7 µm peaks
in the reststrahlen band of quartz particles, and therefore it is extremely
sensitive to the presence of desert sand in the SEVIRI scene. The channel can
be used to map the desert area over the globe. In fact, in Fig. we can nicely see the sand seas characterizing the Sahara and
the Arabian Desert.
November 2007 map of channel emissivity at 8.7 µm over the SEVIRI full disk.
November 2007 map of channel emissivity at 10.8 µm over the SEVIRI full disk.
For completeness, Figs. and also show the
emissivity maps at 10.8 and 12 µm, respectively. It is seen that the
emissivity at 10.8 and 12 µm, as expected, has less contrast than
that at 8.7 µm. The emissivity at 10.8 and 12 µm are much more
dependent on vegetation growth (higher emissivity for green and lower values
for senescent vegetation) and hence shows, as expected, higher values over
regions with evergreen forests (e.g. the African rainforest) and lower
emissivities for regions corresponding to bare soil or senescent vegetation.
Lower-emissivity regions are also seen in south-eastern Africa and
western Madagascar. However, these emissivities are in agreement with the
vegetation state in November as confirmed also by normalized differential
vegetation index (NDVI) maps, which for those regions, in November, show an
NDVI below 0.1 (e.g.
http://www.ospo.noaa.gov/Products/land/vhp/vhp_images.html?product=NDVI).
To conclude this section, we will show a comparison with the UW/BFEMIS database for November 2007. Once again, we stress that UW/BFEMIS emissivities have
been re-mapped to SEVIRI using a high-spectral-resolution algorithm as
described in Sect. .
The comparison in Fig. shows that the differences are well
confined within ±0.05. At 8.7 µm the larger differences correspond
to the desert regions, e.g. the Sahara. Desert regions are those with
the higher variability at 8.7 µm, and therefore these differences are
expected. For the channel at 10.8 µm the emissivity difference is
uniformly close to zero, the same as for the channel at 12 µm. However,
at 12 µm a consistent area is seen in the subtropical region of both
Africa and South America, where the emissivity difference can reach values as
large as ≈ 0.03. For these regions, in November, AVHRR-based NDVI
maps also show a large spatial variability with NDVI, which can drop below
0.1. This difference between SEVIRI and UW/BFEMIS is currently under
investigation.
November 2007 map of channel emissivity at 12 µm over the SEVIRI full disk.
November 2007. Full-disk emissivity difference map (SEVIRI–UW/BFEMIS) for the channels at 8.7 µm (top left), 10.8 µm
(top right) and 12 µm (bottom).
Conclusions
We have developed and implemented a time dimension KF scheme which is capable
of retrieving surface temperature and emissivity from SEVIRI channels with
improved accuracy. The algorithm has been demonstrated for the SEVIRI
atmospheric window channels at 12.0, 10.8 and 8.7 µm and applied to a
series of case studies which include land, ocean, and a large variety of
climate and weather conditions.
Based on these case studies we have shown that the implementation
of the KF scheme we have developed is robust and is not driven to
instability by large and persistent data voids due to clouds
and other anomalous events.
The many case studies we have performed and described have shown
the following:
Ts for land surface, based on the Evora and Gobabeb validation
station) Ts, can be estimated with a rms error of ≈1.8∘C
for vegetated areas and ≈1.2∘C for homogeneous arid
surface. Comparison with validation stations (Evora and Gobabeb) shows a bias
of less than 1 ∘C.
Ts for sea surface can be estimated with a rms error better than 1 ∘C. Comparison
with ECMWF, MODIS and AVHRR OI SST products for Ts shows that the
bias is nearly zero.
For sea and land surface temperature the scheme has proven capable of
correctly following the daily and seasonal cycles.
For land surface emissivity, the approach has proven to be capable of
retrieving the seasonal cycle due to vegetation growth and also capable of
revealing short-timescale fluctuations of emissivity over a desert site.
The retrieval system is robust and is not affected by large data voids, and it
can rapidly recover in the case of anomalous events, which include
natural phenomena and/or bad data as well.
The scheme can be safely applied to the SEVIRI full disk, and we have derived the very first
SEVIRI full-disk emissivity maps at 12, 10.8 and 8.7 µm.
It should be stressed that, although based on a fully physical scheme which
solves and inverts the radiative transfer equation with the accuracy of a
monochromatic forward model, the software tool we have developed is very
fast. Its computational performance has been tested on a quad-core Intel
processor with a clock frequency of 2.7 GHz and 1 GB of RAM. The time
needed to process one single pixel for a single SEVIRI time slot of 15 min is
0.04 s. With a single processor the time needed to run over a regional area, such as that shown in Fig. (which is made up of 9643 SEVIRI
pixels), would be only 6 min per SEVIRI scene. Since each scene is acquired
in a time slot of 15 min, this opens the way to the very first fully physical
retrieval scheme for real-time continuous monitoring of surface parameters,
which could be used for the various purposes of tourism and agronomy, land
surveillance, and natural hazard and risk assessment analysis.
For offline applications, the scheme can also quickly process the
SEVIRI full disk for (Ts,ϵ). Considering that the scheme processes
only clear sky (about 20 % of SEVIRI pixels at the global scale), a run to
process one single month (the study has considered the month of November)
would take ≈ 10 days with 100 processor units. Therefore, a global
scale satellite data centre, such as EUMETSAT or LSA SAF, could release
monthly maps of surface emissivity and temperature in near-real time. The
emissivity maps could be used as additional input for land use/land cover
change analyses and would be beneficial for many Ts statistical retrieval
algorithms for SEVIRI, which largely rely on the availability of the channel
emissivity at 10.8 and 12 µm. This availability would improve the
exploitation of the European geostationary platforms and also lead to a
better exploitation and improved usage of other European satellite systems.
Acknowledgements
We would like to thank the three anonymous referees for their helpful comments and remarks, which greatly helped us to improve the quality of the paper. This work was partially supported through EUMETSAT
contract EUM/CO/11/4600000996/PDW and project Ritmare-Ricerca
Italiana per il Mare (CNR-MIUR). Part of Fig. and the entire Fig. were reproduced
from Google™ Earth.
Edited by: F. Prata
ReferencesAmato, U., Masiello, G., Serio, C., and Viggiano, M.: The σ-IASI code
for the calculation of infrared atmospheric radiance and its derivatives,
Environ. Modell. Softw., 17, 651–667,
doi:10.1016/S1364-8152(02)00027-0,
2002.Baldridge, A. M., Hook, S. J., Grove, C. I., and Rivera, G.: The ASTER
Spectral Library version 2.0, Remote Sens. Environ., 113, 711–715,
doi:10.1016/j.rse.2008.11.007,
2009.
Borbas, E. E. and B. C. Ruston: The RTTOV UWiremis IR land surface
emissivity module, Document NWPSAF-MO-VS-042, EUMETSAT, Darmstadt, Germany,
2010.Carissimo, A., De Feis, I., and Serio, C.: The physical retrieval methodology
for IASI: the δ-IASI code, Environ. Modell. Softw., 20, 1111–1126,
doi:10.1016/j.envsoft.2004.07.003,
2005.
Ermida, S. L., Trigo, I. F., DaCamara, C. C., Göttsche, F. M., and
Olesen, F. S.: Validation of remotely sensed surface temperature over an oak
woodland landscape – the problem of viewing and illumination geometries,
Remote Sens. Environ., 148, 16–27, 2014.Freitas, S. C., Trigo, I. F., Bioucas-Dias, J. M., and Gottsche, F.-M.:
Quantifying the uncertainty of land surface temperature retrievals from
SEVIRI/Meteosat, IEEE T. Geosci. Remote, 48, 523–534,
doi:10.1109/TGRS.2009.2027697,
2010. French, A. N., Schmugge, T. J., and
Kustas, W. P.: Discrimination of Senescent Vegetation Using Thermal
Emissivity Contrast, Remote Sens. Environ., 74, 249–254, 2000. Herman, B. M., Ben-David, A., and Thome, K. J.:
Numerical technique for solving the radiative transfer equation fro a
spherical shell atmosphere, Appl. Optics, 33, 1760–1770, 1994.
Gillespie, A. R., Matsunaga, T., Rokugawa, S., and Hook, S. J.: Temperature
and emissivity separation from Advanced Spaceborne Thermal Emission and
Reflection Radiometer (ASTER) images, IEEE T. Geosci. Remote, 36, 1113–1126,
1998.
Göttsche, F. M. and Hulley, G. C.: Validation of six satellite-retrieved
land surface emissivity products over two land cover types in a hyper-arid
region, Remote Sens. Environ., 124, 149–158, 2012.
Göttsche, F. M., Olesen, F. S., and Bork-Unkelbach, A.: Validation of
land surface temperature derived from MSG/SEVIRI with in situ measurements at
Gobabeb, Namibia, Int. J. Remote Sens., 34, 3069–3083, 2013.Grieco, G., Masiello, G., Matricardi, M., Serio, S., Summa, D., and
Cuomo, V.: Demonstration and validation of the φ-IASI inversion
scheme with NAST-I data, Q. J. Roy. Meteor. Soc., 133, 217–232,
doi:10.1002/qj.162, 2007.
Guillevic, P. C., Bork-Unkelbach, A., Göttsche, F. M., Hulley, G.,
Gastellu-Etchegorry, J. P., Olesen, F. S., and Privette, J. L.: Directional
viewing effects on satellite land surface temperature products over sparse
vegetation canopies – a multisensor analysis, IEEE T. Geosci. Remote, 10,
1464–1468, 2013.Hilton, F., Armante, R., August, T., Barnet, C.,
Bouchard, A., Camy-Peyret, C., Capelle, V., Clarisse, L., Clerbaux, C.,
Coheur, P. F., Collard, A., Crevoisier, C., Dufour, G., Edwards, D., Faijan,
F., Fourrié, N., Gambacorta, A., Goldberg, M., Guidard, V., Hurtmans, D.,
Illingworth, S., Jacquinet-Husson, N., Kerzenmacher, T., Klaes, D., Lavanant,
L., Masiello, G., Matricardi, M., McNally, A., Newman, S., Pavelin, E.,
Payan, S., Péquignot, E., Peyridieu, S., Phulpin, T., Remedios, J.,
Schlüssel, P., Serio, C., Strow, L., Stubenrauch, C., Taylor, J., Tobin,
D., Wolf, W., and Zhou, D.: Hyperspectral Earth Observation from IASI: four
years of accomplishments, B. Am. Meteorol. Soc., 93, 347–370,
10.1175/BAMS-D-11-00027.1, 2012.
Kalman, R.E: A new approach to linear filtering and prediction problems, J.
Basic Eng.-T. ASME, 82, 35–45, 1960.
Kalman, R. E. and Bucy, R. S.: New results in linear filtering and prediction
theory, J. Basic Eng.-T. ASME, 83, 95–107, 1961.Li, J., Li, Z., Jin, X., Schmit, T. J., Zhou, L., and Goldberg, M.: Land
surface emissivity from high temporal resolution geostationary infrared
imager radiances: 1. Methodology and simulation studies, J. Geophys. Res.,
116, D01304,
doi:10.1029/2010JD014637,
2011.Li, Z., Li, J., Li, Y., Zhang, Y., Schmit, T. J., Zhou, L., Goldberg, M. D.,
and Menzel, W. P.: Determining diurnal variations of land surface emissivity
from geostationary satellites, J. Geophys. Res., 117, D23302,
doi:10.1029/2012JD018279,
2012.Li, Z.-L., Tang, B.-H., Wu, H., Ren, H., Yan, G., Wan, Z., Isabel, F.
Trigo, I. F., and Sobrino, J. A.: Satellite-derived land surface temperature:
current status and perspectives, Remote Sens. Environ., 131, 14–37,
doi:10.1016/j.rse.2012.12.008,
2013.Masiello, G. and Serio, C.: Simultaneous physical retrieval of surface
emissivity spectrum and atmospheric parameters from Infrared Atmospheric
Sounder Interferometer spectral radiances, Appl. Optics, 52, 2428–2446,
doi:10.1364/AO.52.002428,
2013.Masiello, G., Serio, C., Carissimo, A., Grieco, G., and Matricardi, M.:
Application of φ-IASI to IASI: retrieval products evaluation and
radiative transfer consistency, Atmos. Chem. Phys., 9, 8771–8783,
doi:10.5194/acp-9-8771-2009,
2009.Masiello, G., Serio, C., Deleporte, T., Herbin, H., Di Girolamo, P.,
Champollion, C., Behrendt, A., Bosser, P., Bock, O., Wulfmeyer, V.,
Pommier, M., and Flamant, C.: Comparison of IASI water vapour products over
complex terrain with COPS campaign data, Meteorol. Z., 22, 471–487,
doi:0.1127/0941-2948/2013/0430,
2013a.Masiello, G., Serio, C., De Feis, I., Amoroso, M., Venafra, S., Trigo, I. F., and Watts, P.: Kalman filter physical retrieval of surface emissivity and temperature from geostationary infrared radiances, Atmos. Meas. Tech., 6, 3613–3634,
doi:10.5194/amt-6-3613-2013,
2013b.Masiello, C., Serio, C., Venafra, S., De Feis, I., Borbas, E. E.: Day-Night
variation of Sahara desert sand emissivity during the dry season from IASI
observations, J. Geophys. Res., 119, 1626–1638,
doi:10.1002/jgrd.50863, 2014.
Masuda, K., Takashima, T., and Takayama, Y.: Emissivity of pure and sea
waters for the model sea surface in the infrared window regions, Remote Sens.
Environ., 24, 313–329, 1988.
Matsunaga, T.: A temperature-emissivity separation method using an empirical
relationship between the mean, the maximum, and the minimum of the thermal
infrared emissivity spectrum, Jour. Remote Sens. Soc. Japan, 14, 230–241,
1994 (in Japanese with English abstract).
Nychka, D. W. and Anderson, J. L.: Data Assimilation. Handbook of Spatial
Statistics, edited by: Gelfand, A., Diggle, P., Guttorp, P., and Fuentes, M.,
Chapman & Hall/CRC, New York, USA. 2010.
Reynolds, R. W., Smith, T. M., Liu, C., Chelton, D. B., Casey, K. S., and
Schlax, M. G.: Daily high-resolution-blended analyses for sea surface
temperature, J. Climate, 20, 5473–5496, 2007.
Rodgers, C. D.: Inverse Methods for Atmospheric Sounding, Theory and Practice
World Scientific, Singapore, 2000.
Rozenstein, O., Agam, N., Serio, C., Masiello, G., Venafra, S., Achal, S.,
Puckrin, E., and Karnieli, A.: Diurnal emissivity dynamics in bare versus
biocrusted sand dunes, Sci. Total Environ., 506, 422–429, 2015.Schluessel, P., Emery, W. J.,
Grassl, H. and Mammen, T.: On the bulk-skin temperature difference and its
impact on satellite remote sensing of sea surface temperature, J. Geophys.
Res., 13341–13356, 10.1029/JC095iC08p13341, 1990.Seemann, S. W., Borbas, E. E., Knuteson, R. O., Stephenson, G. R., and Huang,
H. L.: Development of a Global Infrared Land Surface Emissivity Database for
Application to Clear Sky Sounding Retrievals from Multispectral Satellite
Radiance Measurements, J. Appl. Meteorol. Clim., 47, 108–123,
doi:10.1175/2007JAMC1590.1,
2007.
Serio, C., Masiello, G., Amoroso, M., Venafra, S., Amato, U., and De
Feis, I.: Study on Space-Time Constrained Parameter Estimation from
Geostationary Data, Final Progress Report, EUMETSAT contract No.
EUM/CO/11/4600000996/PDW, 2013.
Serio, C., Masiello, G., Amoroso, M., Venafra, S., Amato, U., and De
Feis, I.: Kalman Filter estimation of surface temperature and emissivity from
SEVIRI, Final Report, EUMETSAT contract No. EUM/CO/14/4600001329/PDW,
2014.Tarantola, A.: Inverse Problem Theory: Methods for Data Fitting and Model
Parameter Estimation, Elsevier, New York, USA, 1987.
Theocharous, E., Usadi, E., and Fox, N. P.:
CEOS Comparison of IR Brightness Temperature Measurements in Support of
Satellite Validation, Part I: Laboratory and Ocean Surface Temperature
Comparison of Radiation Thermometers, NPL REPORT OP3, Technical Report ISSN: 1754–2944,
National Physical Laboratory, Teddington, UK, 2010.Trigo, I. F., Monteiro, I. T., Olesen, F., and Kabsch, E.: An assessment of
remotely sensed land surface temperature, J. Geophys. Res., 113, D17108,
doi:10.1029/2008JD010035,
2008a.
Trigo, I. F., Peres, L. F., DaCamara, C. C., and Freitas, S. C.: Thermal land
surface emissivity retrieved from SEVIRI/Meteosat, IEEE T. Geosci. Remote,
46, 307–315, 2008b.Wikle, C. K. and Berliner, L. M.:
A Bayesian tutorial for data assimilation,
Physica D, 230, 1–16, 2007.
Wulfmeyer, V., Bauer, H., Di Girolamo, P., and Serio, C.: Comparison of
active and passive water vapor remote sensing from space: an analysis based
on the simulated performance of IASI and space borne differential absorption
lidar, Remote Sens. Environ., 95, 211–230, 2005.
Xu, H., Yu, Y., Tarpley, D., Göttsche, F. M., and Olesen, F. S.:
Evaluation of GOES-R land surface temperature algorithm using SEVIRI
satellite retrievals with in situ measurements, IEEE T. Geosci. Remote, 52,
3812–3821, 2014.