JP 1.20
Global Hyperspectral Resolution Surface IR Emissivity Spectra
Derived from AIRS Measurements
Jinlong Li, Jun Li, Elisabeth Weisz, Eva Borbas
Cooperative Institute for Meteorological Satellite Studies (CIMSS), University of Wisconsin - Madison
Figure 4: The upper shows the global surface emissivities at 8.21µm derived by
the composite of clear sky single field of view retrievals for January 2004, the
lower is the ecosystem land cover map through slightly re-group IGBP
ecosystem categories.
3. Retrieval experiments
1. Introduction
The hyperspectral resolution surface IR emissivity spectra are very important
for retrieving products such as dust properties, land surface temperature and
cloud-top properties, as well as hyperspectral radiance assimilation overland.
The emissivity uncertainty has a significant impact on the retrieval of
boundary layer temperature and moisture, especially over the desert regions
where surface IR emissivity has large variations both spectrally and spatially.
In this research, a physically based algorithm (Li et al. 2007) has first been
developed to retrieve hyperspectral IR emissivity spectrum simultaneously
with the temperature and moisture profiles from AIRS. Besed on that, a
global hyperspectral emissivity spectra product has been further derived.
Simulation over desert:
Regression retrieval: T, W, O3 profiles, Ts, Emisssivity
Three types of physical retrieval
1) Using constant emissivities of 0.98 and fixed in iterations.
2) Using regression emissivities and fixed in iterations.
3) Using regression emissivities and updated in iterations.
Comparison with MODIS operational collection 4 product
Emissivity spectral cross-sections along the latitudes shown in Fig. 4
United States
North African
2. Retrieval algorithm
In general, atmospheric measurement equation is written as
y = F ( x) + e
y = ( R1 , R2 ,..., Rn )T ;
x = (t ( p ); w( p ); o( p ); t s ; ε 1 ,..., ε n ; )
T
Retrieval process is based on the regularization and discrepancy principle
developed by J. Li and H.L. Huang (1999). Because surface emissivity is
channel related, there are too many parameters to be retrieved if
including all channels’ emissivities !!! However, retrieving emissivity
spectrum is still possible if the eigenvector expansion is used:
l
x = ∑ aiϕi = aφ ;
i
J ( x ) = J ( aφ )
Tskin RMS (K)
F: forward model operator;
e: measurement and model error
y: measured radiance;
x: retrieved parameters
⎧φ: eigenvector matrix;
⎨
⎩a: eigenvector coefficients
Reg
0.624
Rtv
0.540
Fixed emis
0.822
Emis=0.98
9.544
Figure 2: The RMSE of simulated retrievals of surface emissivity, surface
skin temperature, temperature and moisture profiles for three
configurations described along with the first guess (from regression)
results over desert region. The true surface emissivities are shown in the
upper right panel.
Figure 7: (top left) The CIMSS AIRS convolved 8-day (01–08 January 2008)
emissivity retrieval map at 8.55 µm MODIS spectral band, (bottom left) the
operational MODIS 8-day composite emissivity map at 8.55 mm spectral band
(collection 4), (top right) the difference image between AIRS and MODIS, and
(bottom right) the histogram of the emissivity differences.
China
Australia
Comparison with AIRS operational version 5 L2 standard product
Initial guess effects (granule 002 of 6 January 2004)
= ( ym − yc (aφ ))T E −1 ( ym − yc ( aφ )) + (( a − a0 )φ )T γS −1 ((a − a0 )φ )
ym: satellite measurement
yc: forward model calculated radiance
a0: initial guess
E: measurement error covariance matrix
S: background error covariance matrix
γ: dynamical factor to balance measurement and background contribution
Figure 5: The selected emissivity spectral cross-sections along the latitudes
shown in Figure 5 for January of 2004. The spectral and spatial variations of
derived emissivity contain useful information on the ecosystem and land
surface type properties. They reflect the dramatic differences in International
Geosphere-Biosphere Program (IGBP) classification.
By minimizing cost function and using quasi-nonlinear Gauss-Newton
iteration
Define : c = a − a0
~
~
~ ~
~
ci +1 = (γS −1 + K T E −1 K ) −1 K T E −1[( ym − yc (ci + a0 )) + Kci ]
~
~
K = Kφ ; S = S φ ;
K is the jacob matrix :
k ij = ∂Fi ( x ) ∂x j
After transformation, retrieval variables have been reduced from a
couple thousands to a few tens for AIRS application.
5. Comparison with different products
Figure 3: The AIRS 9.3 um surface emissivity retrieval images overlaying
on AIRS brightness temperature (K) image (black/white) for Granule 002
on 06 January 2004 with (top left) 0.98 (constant spatially and spectrally)
as first guess and (middle left) regression as first guess, respectively. (top
right) The difference image and (middle right) the histogram of differences.
(bottom left) The IGBP ecosystem land type map, and (bottom right) the
location of the AIRS granule over central African.
Brightness temperature residuals from different emissivity products
Figure 8: The CIMSS AIRS 8-day (01–08 January 2008) emissivity retrieval maps
(top) vs the operational AIRS version 5 L2 standard products (bottom) at 12.02 µm
(left) and 3.82 µm (right) spectral bands.
6. Summary
4. Global hyperspectral surface IR
emissivity spectra product
A physically based algorithm has been developed to retrieve hyperspectral IR
emissivity spectrum simultaneously along with the temperature and moisture profiles.
The algorithm has been further applied to one month of AIRS radiance
measurements. A global hyperpectral surface IR emissivity maps, with the full AIRS
spectral coverage, are then derived by the composite of clear sky single field of view
retrievals. The product shows very promising results. The spatial and spectral
variations of derived emissivity contain useful information on the ecosystem and land
surface type properties represented by IGBP classifications. This derived product will
be very useful in the remote sensing and climate research community.
Global surface emissivity (Jan. 2004) and IGBP ecosystem maps
Re-group
from IGBP
category:
Figure 1: To make the computation more efficient, an analytical Jacobian
calculation was used in the retrieval. A calculated surface emissivity
Jacobian along with the brightness temperature and the first 10 eigenvectors
of surface emissivity for AIRS spectrum are shown in the figure.
Forests:
Evergreen
needle forests
Evergreen
broad forests;
Deciduous
needle forests;
Deciduous
broad forests;
mixed forests;
Shrubs:
Opened
shrubs; Closed
shrubs;
Savanna:
Woody
savanna;
Savanna;
Cropland:
Cropland;
Crop mosaic;
Snow/Ice:
Snow; Ice;
Tundra;
Desert:
Desert/Barren;
Ecosystem land cover
Fig. 4
ACKNOWLEDGEMENT: This program is supported by NOAA GOES-R project at CIMSS.
Global AIRS emissivity map – CIMSS research product
(January 2004)
Figure 6: The calculated AIRS brightness temperature residuals by using
different emissivity products for AIRS observation at 00:03 UTC of 15 January
2005 around the location (latitude=25.0749, longitude=26.0577). (a) MODIS
based UW best fit emissivity database (cyan for collection 4 and black for
collection 5); (b) MODIS based UW hyperspectral emissivity database (blue
for collection 4 and magenta for collection 5); (c) AIRS L2 standard product
(red); (d) AIRS based UW hyperspectral emissivity research product (green).
(Borbas et al., 16th ITSC, 2008)
REFERENCES
Li, J. and J. Li, 2008: Derivation of global hyperspectral resolution surface emissivity spectrafrom
advanced infrared sounder radiance measurements, Geophys. Res. Lett., 35, L15807,
doi:10.1029/2008GL034559.
Li, J., J. Li, E. Weisz, and D.K. Zhou, 2007: Physical retrieval of surface emissivity spectrum from
hyperspectral infrared radiances, Geophys. Res. Lett., 34, L16812,
doi:10.1029/2007GL030543.
Li, J., and H.-L. Huang, 1999: Retrieval of atmospheric profiles from satellite sounder measurements
by use of the discrepancy principle, Appl. Optics, Vol. 38, No. 6, 916-923.
Li, J., 1994: Temperature and water vapor weighting functions from radiative transfer equation with
surface emissivity and solar reflectivity, Adv. Atmos. Sci., 11, 421– 426
Contact: Jinlong.Li@ssec.wisc.edu; Jun.Li@ssec.wisc.edu;