Requirements Consolidation of the Near-Infrared Channel of the GMES-Sentinel-5 UVNS Instrument MTR,...
-
Upload
charles-hudson -
Category
Documents
-
view
215 -
download
1
Transcript of Requirements Consolidation of the Near-Infrared Channel of the GMES-Sentinel-5 UVNS Instrument MTR,...
Requirements Consolidation of the Near-Infrared Channel of the GMES-Sentinel-5 UVNS InstrumentMTR, 1st October 2013
Task 2Scattering profile
characterisation for SWIR
Leif Vogel, Hartmut BoeschUniversity of Leicester
Study Sentinel 5 instrument concepts (A and B) w.r.t. aerosol profile characterisation for SWIR species.
Link from aerosol information in NIR to trace gas retrievals in SWIR Simulate global coverage for a single day (April) of realistic S5
sampling applying ECHAM 5 simulation supplied by Butz et al. 1) Instrument noise (based on recent input from ESA).2) Effect of vegetation fluorescence3) Error in spectral response function width (assuming 1% error)4) Spectrally uniform offset in radiance units (assuming 1% of continuum
radiance).5) The ARA requirement
SWIR S5 Products
UoL Task 2 Overview
Target gas Spectral windows
CH4 1.6μm, 2.3μm
CO 2.3μm
Additional information used for aerosol profile
O2-A 0.76μm
O2-B 0.69μm
CO2 1.6μm for proxy retrievals
Approach for Retrieval Simulations
Spectra are simulated using the forward modelling of UoL FP retrieval algorithm
two instrumental setups range of geophysical scenarios
Retrieval sensitivity tests for retrievals w.r.t. scattering profiles, retrieval applies
the same a priori trace gas profiles, temperature profile, surface albedo different setup for aerosol and cirrus a priori Bias given by difference true and retrieved XCH4
The UoL Retrieval Algorithm Measured radiance
spectra are non-linear function of atmospheric parameters
retrieval is performed iteratively by alternating calls to:
Forward Model describes physics of measurement: Multiple-scattering RT Instrument Model Solar Model
Inverse Method estimates state: Rodger’s optimal
estimation technique
XCH4, XCO and its error is computed from retrieved state after iterative retrieval has converged
Names Quantity Notes
CO and CH4 1 Multiplier to a priori profile
H2O, HDO, CO2 1 Multiplier to a priori profile
Temperature 1 Additive offset to a priori profile
Aerosols AOD, height and width
Gauss profile
Clouds AOD, height and width
Gauss profile
Surface Albedo #bands x 2para Albedo at band centre and slope
Typical State Vector
Concept A Bands NIR (685 – 773 nm)* SWIR 1 SWIR3 NIR 1 NIR 2
Wavelengths [nm] 685 - 700 750 – 773 1590 - 1675 2305 - 2385
Numbers of pixel 116 177 850 800FWHM ISF 0.39 0.39 0.25 0.25
*) Simulated retrievals do not use full range due to strongly changing surface albedo
Concept B
Bands NIR SWIR 1 SWIR3
Wavelengths [nm] 755 – 773 1590 - 1675 2305 - 2385
Numbers of pixel 450 850 870FWHM ISF 0.12 0.23 0.23
Instrumental setup
Instrumental setup
Concept A: NIR1 NIR2 SWIR1 SWIR3
Concept B: NIR SWIR1 SWIR3
Simulated scenarios
One day in April 2015 as described in Butz et al. 2010, Butz et al. 2012 applying ECHAM 5 model simulations (Stier et al 2005)
Stier et al 2005
ECHAM Desaster
Simulated scenarios
ECHAM 5 model simulations as described in Stier et al 2005, Butz et al 2010, Butz et al 2012
(18 layers x 7 aerosol types x ~2700 Observations)
Simulated scenarios
In most cases very large Aerosols particles with subsequent unrealistic low Angstroem coefficients.
•Very strong absorption in the SWIR3 band
•Erroneous relative signal to noise ratios for different wavelength channels
Simulated scenarios Alternative approach:Replacing aerosol types described in Stier et al. 2005 with similar ones described in Kahn et al. 2001 based on aerosol type and radiusCreating a joint aerosol mix per observation with weights depending on respective ECHAM composition per observationApplying original aerosol altitude profile
ECHAM Kahn et al. 2001
Mode Aerosols Base/Mixt.
Aerosols
Nucleation SU Base SU land
Aitken SU, BC, POM Mix 5a SU, acc.DU, BC, Carb
Accumulation SU, BC, POM, SS, DU
Mix 3a SU, SS, BC, Carb
Coarse SU, BC, POM, SS, DU
Mix 4a SU, acc.DU, coarse DU, Carb
Aitken BC, POM Mix 3b BC, Carb, SU, SS
Accumulation DU Base Acc. DU
Coarse DU Base Coarse DU
Simulated scenarios
Forward model:
Simulated scenarios
Dust dominated
Sulphurdominated
Aerosol properties show more realistic optical properties (Angstrom) than in first approach
But range of properties is very large which is expected to be problematic for retrieval
Concept A
Concept B
Example of produced spectra from TN1
Retrieval Setup State vector: Scaling factors for the CH4, CO, H2O, HDO, CO2 vmr profile;
Temperature factors, surface albedo + tilt per band, parameters for Gauss profile for cirrus, parameters for Gauss profile for 2 aerosol types
A priori values: Atmosphere as in simulations Aerosol extinction profile: Gaussian-shaped at height of 2 km a.g.l., width
(FWHM) of 1 km and AOD of 0.05 Cirrus extinction profile: Gaussian-shaped at height of 10 km, width
(FWHM) of 1 km and optical depth of 0.05.
2 Aerosol types: Due to the wide range of simulated aerosols which are not captured by individual Kahn mixtures, two simulated aerosols were chosen:
A) Large Angstroem Coefficients (high sulfate component)B) Small Angstroem Coefficients (high dust component)
Cirrus type: as in simulations
Aerosol + cirrus parameters differ from simulations (except cirrus type)
Quality-Filtering Retrievals
Concept A Concept B CH4 CH4
ConvergedSoundings 594 (23%) 881 (35%)
FilteredSoundings 405 (16%) 639 (25%)
Only converged retrievals are used: Number of converging iteration steps ≤ 12Number of diverging iteration steps ≤ 5
Additional post-processing quality filter: Χ2 < 1 per spectral bandCH4 error < 0.4%Retrieved AOD < 0.2Retrieved AOD+COD < 0.3 Surface albedo at O2 bands < 0.7 (removes snow and ice)
0
0
0
10
10
10
All
soun
ding
sco
nver
ged
soun
ding
sfil
tere
d so
undi
ngs
Co
nce
pt
A
Soundings
Total numberConvergedFiltered
Effect of the filter
Results: Concept A
CH4 Converged Filtered
Bias (%)0.013 +/- 0.955
0.036 +/- 0.342
Precision (%)0.112 +/- 0.055
0.103 +/-0.052
Impact of scattering error on trace gas retrieval
unfiltered filtered
CH
4 B
ias
CH
4 ra
nd.
err
.
Converged retrievals do not show obvious dependency on location
Concept A
Results: Concept ADegrees of freedom inferred from the diagonal elements
of averaging kernel matrix (Rogers, 2001)
Maximum number of DoF for aerosol and cirrus = 3 (optical depth, altitude, width) per type
Aerosol type 1 DoF ~ 2
Aerosol type 2 DoF ~1 - 2
Cirrus clouds DoF ~ 1 - 2
Mean DoF (AOD+COD): 4.63 (4.83)
Aerosols poorly retrieved
Distribution of retrieved vs. true AOD mirrors the wide range of aerosol mixtures retrieved with two opposing types and high possibly aerosol load
Good correlation between retrieved and true COD
Results: Concept ADependency on albedo
Results: Concept A
Dependency of CH4 bias on aerosol
0
0
0
10
10
10
All
soun
ding
sco
nver
ged
soun
ding
sfil
tere
d so
undi
ngs
Co
nce
pt
B
Results: Concept B
CH4 Converged Filtered
Bias (%)-0.307 +/- 1.263
-0.171 +/- 0.683
Precision (%)0.140 +/- 0.083
0.124 +/- 0.068
Impact of scattering error on trace gas retrieval
unfiltered filtered
CH
4 B
ias
CH
4 ra
nd.
err
.
Converged retrievals do not show obvious dependency on location
Concept B
Results: Concept BAerosol type 1 DoF ~ 2
Aerosol type 2 DoF ~1 – 2
In comparison to Concept A, no obvious change in DoF and distribution for both aerosol types
Cirrus clouds DoF ~ 1 - 2
Total DoF ~ 4 – 5.5
− Distribution is slightly skewed to lower values in comparison to Concept A
Good information content for retrieving aerosol and cirrus parameters
Distribution of retrieved vs. true AOD mirrors the wide range of aerosol mixtures retrieved with two opposing types, although not as extrem as Concept A
Good correlation between retrieved and true COD
Results: Concept B
Dependency of CH4 bias on aerosols
Results: Concept BDependency on albedo
Concept A Concept B CH4 CH4
ConvergedSoundings 594 (23%) 881 (35%)Bias (%) 0.013 +/- 0.955 -0.307 +/- 1.263
Precision (%) 0.112 +/- 0.055 0.140 +/- 0.083 Filtered
Soundings 405 (16%) 639 (25%)Bias (%) 0.036 +/- 0.342 -0.171 +/- 0.683
Precision (%) 0.103 +/-0.052 0.124 +/- 0.068
Comparison of concepts A & B
Results obtained for concept A and B show that
The number of converged retrievals and retrievals that passed quality filter is higher for concept B
variability of simulated aerosols vs. the two opposing types used in the retrieval and the lesser constrain by the missing O2-B band.
Mean bias and standard deviations for CH4 is larger for concept B
Both concepts have similar precision
These conclusions hold after applying the filter
Con
cept
AC
once
pt B
Main conclusions (so far):
Concept A may yield less biased results at higher precision, although the true AOD was resolved at lesser accuracy. The additional O2-B band may therefore yield important aerosol information However, in total the differences between concepts are not very big.
However, additional errors may be introduced (see RAL study)
Still to be assessed for these scenarios:1) Effect of vegetation fluorescence, from which the O2-B band is more
affected relatively 2) Error in spectral response function width (assuming 1% error)3) Spectrally uniform offset in radiance units (assuming 1% of continuum
radiance).4) The ARA requirement
Technical note will be provided ….