s5p cloud products
DESCRIPTION
S5P cloud products. Sebastián Gimeno García, Ronny Lutz, Diego Loyola German S5P Verification Meeting 1 Bremen, 28-29 November 2013. www.DLR.de • Chart 1. > Vortrag > Autor • Dokumentname > Datum. Outline. General overview OCRA adaptation to S5P Cloud model – OCRA/ROCINN-CAL,-CRB - PowerPoint PPT PresentationTRANSCRIPT
S5P cloud products
Sebastián Gimeno García, Ronny Lutz, Diego Loyola
German S5P Verification Meeting 1Bremen, 28-29 November 2013
www.DLR.de • Chart 1 > Vortrag > Autor • Dokumentname > Datum
Chart 2
Outline
General overview
OCRA adaptation to S5P
Cloud model – OCRA/ROCINN-CAL,-CRB
ROCINN CRB: CA variable vs. CA fixed
Cloud inhomogeneity effects
Conclusions
Outlook
Chart 3
S5P cloud information primarily needed for accurate trace gas retrieval
Influence of clouds on gas retrieval (1D):
albedo effect shielding effect Multiple scattering effect
Overview
+ others: e.g. multiple cloud layering system, …
Chart 4
S5P cloud information primarily needed for accurate trace gas retrieval
Influence of clouds on gas retrieval (3D):
neighbouring pixel effect in-pixel inhomogeneity effects
Overview
+ others: e.g. effect of scene variability on spectral calibration, …
Chart 5
OCRA adaptation to S5P – Input Data
OCRA for GOME,
SCIAMACHY and GOME-2
uses the PMD UVN data with
a resolution of ~10x40km2
OCRA for TROPOMI will use
the UVN radiance data with
a resolution of 7x7km2
The initial S5P cloud-free
composites will be based on
OMI data with a resolution of
13x24km2 at nadir
Chart 6
OCRA adaptation – OMI cloud-free composite
UV cloud-free for July
VIS cloud-free for July
UV cloud-free for January
VIS cloud-free for January
Monthly composite of cloud-free reflectances in UV-2 and VIS OMI channels
Chart 7
OCRA adaptation – OMI cloud fraction results
CF comparison:
OCRA OMTO3 OMDOAO3
Global pattern good represented by all products
Scan angle dependency
Comparison with OMI official cloud products:
OMCLDO2 OMCLDRRongoing …
Chart 8
OCRA adaptation – OMI cloud fraction results (2)
Clear correlation between all CF products
OCRA shows slope in mean differences
OMDOAO3 delivers larger CFs than the other two products
Chart 9
Cloud fraction (CF) is retrieved using a RGB color space approach → OCRA
Cloud parameters (CTH, COT) are retrieved in the Oxygen A-band using regularization theory → ROCINN
CRB: Clouds are treated as Reflecting Boundary (Lambertian equivalent reflectors)
CAL: Clouds are treated As homogeneous Layers
Photon cloud penetration is allowed
Multiple scattering is accounted for
Modeled radiance contains information below the cloud layer
Retrieved CTH expected to be closer to the geometrical CTH
Cloud Model – OCRA/ROCINN-CAL/CRB
Chart 10
Cloud Model – OCRA/ROCINN-CAL/CRB (2)
Intra-cloud correction
Loyola et al., JGR 2011
Surface
Lambertian Cloud
CAL: Cloud As scattering Layer | CRB: Cloud as Reflecting Boundary
Chart 11
Cloud Model – OCRA/ROCINN-CAL/CRB (3)
Comment from a reviewer of the S5P Cloud ATBD:
„To treat clouds as simple reflectors … is far to simple and might work for large pixels averaging over more than 2000 Km , but is very likely not working for the interpretation of much finer spatial resolution TROPOMI measurements.“
Chart 12
Cloud Model – OCRA/ROCINN-CAL/CRB (4)
100000 independent spectra were simulated using the ROCINN CAL forward model (VLIDORT) covering the whole ROCINN CAL state space (1% noise added):
SH in [0, 2] km SA in [0, 1] CTH in [0, 15] km COT in [0, 125] CGT in [0.5, 14.5] km SZA in [0, 85] ° VZA in [0, 75] ° CF in [0, 1]
CRB retrievals of CAL spectra: effects due to different cloud models
Relative difference: CTHCFXX
XXX
ref
ref ,;)(
*100:)(
Chart 13
Cloud Model – OCRA/ROCINN-CAL/CRB (5)
- CRB retrieved cloud “top” height is systematically smaller than the
geometrical cloud top height. - Discrepancy increases as cloud optical depth decreases.
Global Mean Lambertian Model
Chart 14
ROCINN-CRB: CA variable versus CA fixed
100000 independent spectra were simulated using the ROCINN CRB forward model (VLIDORT) covering the whole ROCINN CRB state space (1% noise added):
SH in [0, 2] km SA in [0, 1] CH in [0, 15] km CA in [0, 1] SZA in [0, 85] ° VZA in [0, 75] ° CF in [0, 1]
Cloud albedo (CA) was set to 0.8 in the cloud property retrieval
Results show the impact of fixing CA to 0.8 in CRB in comparison with a variable CA (not CRB vs. CAL!)
Relative difference: CTHCFXX
XXX
ref
ref ,;)(
*100:)(
Chart 15
ROCINN-CRB: CA variable versus CA fixed (2)
ROCINN CRB with fixed CA (=0.8):
underestimates CF if actual CA is lower than 0.8 overestimates CF if actual CA is higher than 0.8 overestimates CTH if actual CA is lower than 0.8 underestimates CTH if actual CA is higher than 0.8 the larger the SA, the larger the CTH underestimation
CF rel. diff. vs.cloud albedo
CTH rel. diff. vs. cloud albedo
CTH rel. diff. vs. surface albedo
Chart 16
MoCaRT (Monte Carlo Radiative Transfer) Model reflectivities
Cloud inhomogeneity effects
Chart 17
Conclusions
OCRA CF algorithm has been adapted for S5P/TROPOMI
preliminary results for OMI look very promising
OCRA algorithm is computationally very efficient
good agreement with existing algorithms (OMTO3, OMDOAO3): OCRA CFs correlate with both
ROCINN CRB (LER) evaluation:
ROCINN CRB underestimates CTH (as expected) CTH discrepancies increase with decreasing CA/COT
Setting CA to a fixed value (CA_ref=0.8) leads to a complex two-regime (below and above CA_ref) dependency of {CTH, CF} on cloud albedo (cloud optical thickness) and surface albedo
Chart 18
Outlook
Comparisons of OCRA with official OMI cloud products (OMCLDO2, OMCLDRR) ongoing
Case studies with synthetic spectra
OCRA ROCINN-CRB/CAL 3D effects
Chart 19
Thank you for your attention!
Chart 20
Information theory analysis
Degree of freedom of the signal (DFS) ~ 2
Only two independent parameters can be retrieved in the O2 A-band
CTH and COT are retrieved with ROCINN in the O2 A-band
OCRA/ROCINN --- CAL
Chart 21
ROCINN CRB verification
100000 independent spectra were simulated using the ROCINN CRB forward model (VLIDORT) covering the whole ROCINN CRB state space (1% noise added):
SH in [0, 2] km SA in [0, 1] CH in [0, 15] km CA in [0, 1] SZA in [0, 85] ° VZA in [0, 75] ° CF in [0, 1]
Test retrieval performance with respect to {CF, CTH}
Relative difference: CTHCFXX
XXX
ref
ref ,;)(
*100:)(
Chart 22
ROCINN_CRB --- CTH, CA --- verification (1)
The relative differences between the reference CF‘s and CTH‘s and corresponding retrieved values,
X_rel := 100 * (X_out – X_ref) / X_ref,
show good overall perfonmance of the algorithm
Median of the distributions close to zero Most differences within few percent
Chart 23
Very good overall CF retrieval performance
Almost perfect correlation between reference and retrieved CFs
Relative differences show higher spread for large SZA (small cosines: CSZA)
CF retrieval does not show dependency on cloud (CA) and surface albedo (SA)
ROCINN_CRB --- CTH, CA --- verification (2)
CF_out vs. CF_ref CF_rel vs. CSZA
CF_rel vs. CA CF_rel vs. SA
Chart 24
Good overall CTH retrieval performance
CTH slightly understimated and higher spread of CTH_rel for large SZA
CTH relative differences show higher spread for small „cloud albedo fractions“ CAF=CA*CF
CTH retrieval does not show dependency on surface albedo (SA)
ROCINN_CRB --- CTH, CA --- verification (2)
CTH_out vs. CF_ref CTH_rel vs. CSZA
CTH_rel vs. CAF CTH_rel vs. SA
Chart 25
CA --- COT --- SZA relationship