16/06/2015
OPERA : An Operational Atmospheric Correction for Land and Water Sindy Sterckx, Els Knaeps, Ils Reusen, Stefan Adriaensen, Liesbeth De Keukelaere (VITO ) Peter Hunter (University of Stirling) Daniel Odermatt (Odermatt-Brockmann) Claudia Giardino (CNR) Sentinel-3 for Science, 2-5 June 2015
SENTINEL-3 L1 data
Land/Sea Mask
L2 LAND processing
L2 MARINE processing
Oceans Coastal areas
Land Estuaries ? Lakes ? Rivers ?
16/06/2015 © 2014, VITO NV
2
Typical marine atmospheric correction schemes might not work
16/06/2015 © 2014, VITO NV
3
Need to account for non-zero altitude
16/06/2015 © 2014, VITO NV
4
Need to account for non-zero NIR reflectance
16/06/2015 © 2014, VITO NV
5
Knaeps et al., 2014
Need to account for non-zero NIR reflectance (even at 1020 nm) !
16/06/2015 © 2014, VITO NV
6
Need to account for non-water spectra
16/06/2015 © 2014, VITO NV
7
80 70
reflectance (%)
60 50 40 30 20 10 0 400
450
500
550
600
650
700
750
800
850
900
wavelength (nm)
Need to account for adjacency effects
16/06/2015 © 2014, VITO NV
8
950
Typical land atmospheric correction schemes might not work over water
16/06/2015 © 2014, VITO NV
9
Need to account for non-lambertian reflection
16/06/2015 © 2014, VITO NV
10
OPERA: OPERational Atmospheric correction » Sensor generic ( Sentinel-3 , Sentinel-2, Landsat, Hyperspectral missions, …) » Scene generic allowing to correct both land and water areas » including turbid, high altitude, tidal or macrophyte dominated waters » Requires to account for surface elevation variation, adjacency effects, non-Lambertian reflection of water surfaces » Focus on operationability, minimal manual interaction, processing efficiency
16/06/2015 © 2014, VITO NV
11
OPERA overview
TOA SCENE LAND/WATER MASKING LAND BASED AOT RETRIEVAL
SZA,VZA,RAA DEM
WATER VAPOR RETRIEVAL
MODTRAN5 LUT
ADJACENCY CORR. WATER: SIMEC LAND: RANGES
ATMOSPHERIC CORR.
16/06/2015 © 2014, VITO NV
L2 SCENE
12
Land-based AOT retrieval TOA SCENE
Based on Guanter et al. (2008) RSE
SZA,VZA,RAA DEM
Cloud & Water masking
Tiles creation Select Lmin(VIS)
MODTRAN5 LUT
NDVI selection Endmember inversion AOT tiles Inverse distance interpolation Resize (bilinear) Spatial smoothing AOT SCENE
16/06/2015
© 2014, VITO NV
13
Adjacency correction : SIMEC (over water pixels) » Basic assumption : Invariant shape of water reflectance in NIR, ie ‘‘similarity’’ NIR reflectance spectrum (Ruddick et al., 2006)
» No assumption on NIR albedo Sterckx et al. (2015) RSE Sterckx et al. (2011) IJRS 16/06/2015 © 2014, VITO NV
14
Adjacency correction : SIMEC (over water pixels) » Basic assumption : Invariant shape of water reflectance in NIR, ie ‘‘similarity’’ NIR reflectance spectrum (Ruddick et al., 2006)
Sterckx et al. (2015) RSE Sterckx et al. (2011) IJRS
» No assumption on NIR albedo 16/06/2015 © 2014, VITO NV
15
OPERA on MERIS FR
16/06/2015 © 2014, VITO NV
16
WO HO MO
16/06/2015 © 2014, VITO NV
17
OPERA on LANDSAT 8 (OLI)
16/06/2015 © 2014, VITO NV
18
0.04
Mantua lakes (Italy)
0.07
P1 OPERA with SIMEC
0.03
Rw
P1 OPERA without SIMEC
0.02
0.05 0.03 0.01 0
600
700
Wavelength (nm)
0.05
P2
0.04
P3 P4
800
900
P2 OPERA without SIMEC
0.02
600
700
Wavelength (nm)
800
900
P6 insitu P6 OPERA with SIMEC
0.05 0.04
P5 OPERA without SIMEC
0.03 0.02
0.01
0.01 0
0
500
600
700
Wavelength (nm)
0.05
800
P3 OPERA with SIMEC
0.03
Rw
P3 OPERA without SIMEC
0.02
500
600
700
Wavelength (nm)
0.05
P3 insitu
0.04
400
900
800
900
P7 insitu
0.04
P7 OPERA with SIMEC P7 OPERA without SIMEC
0.03
Rw
400
0.02 0.01
0.01
0
0
400
500
600
700
Wavelength (nm)
0.05
800
900
400
500
600
700
Wavelength (nm)
800
P4 insitu
0.04
P4 OPERA with SIMEC
0.03
P4 OPERA without SIMEC
Rw
Landsat 8 23 Sept 2014
500
0.06
P2 OPERA with SIMEC
0.03
400 0.07
P2 insitu
Rw
P5
500
Rw
400
P6
P5 OPERA without SIMEC
0.04
0
P1
P5 OPERA with SIMEC
0.06
0.02
0.01
P7
P5 insitu
0.08
P1 insitu
Rw
0.05
0.02 0.01
16/06/2015
0
400
VITO NV 700 500 © 2014, 600
19 800
900
900
Mantua lakes (Italy)
16/06/2015 © 2014, VITO NV
20
Belgian coast Aeronet OC (RBINS)
+/- 4.5 km
16/06/2015 © 2014, VITO NV
21
OPERA without SIMEC : Reflectance Landsat8 865 nm band OSLO
LC81970192014205LGN00
0.004
0.032 Image LC81970222014205LGN00 16/06/2015 © 2014, VITO NV
22
OPERA with SIMEC : Reflectance Landsat8 865 nm band
LC81970192014205LGN00
0.004
0.032 Image LC81970222014205LGN00 16/06/2015 © 2014, VITO NV
23
OPERA without SIMEC : Reflectance Landsat8 655 nm band
16/06/2015 © 2014, VITO NV
24
OPERA with SIMEC : Reflectance Landsat8 655 nm band
16/06/2015 © 2014, VITO NV
25
OPERA without SIMEC : Landsat8 TSM MAP based on 655nm Nechad et al, 2010 0
8 mg/l
16/06/2015 © 2014, VITO NV
26
OPERA with SIMEC : Landsat8 TSM MAP based on 655nm Nechad et al, 2010 0
8 mg/l
16/06/2015 © 2014, VITO NV
27
OPERA without SIMEC : Reflectance Landsat8 865 nm band
Image: LC81990242014075LGN00 16/06/2015 © 2014, VITO NV
28
OPERA with SIMEC : Reflectance Landsat8 865 nm band
Image: LC81990242014075LGN00 16/06/2015 © 2014, VITO NV
29
OPERA without SIMEC : Landsat8 TSM MAP based on 865nm Nechad et al, 2010
0 16/06/2015 © 2014, VITO NV
100 mg/l 30
OPERA with SIMEC : Landsat8 TSM MAP based on 865nm Nechad et al, 2010
0 16/06/2015 © 2014, VITO NV
100 mg/l 31
Conclusions » Atmospheric correction : complex » Variability of inland waters adds more complexity Remote sensing : underutilised Need for adequate atmospheric correction algorithms » OPERA : step forward » SIMEC mainly neutral or positive effect. Adjacency effect not always clearly present.
16/06/2015 © 2014, VITO NV
32
Outlook » Mountainous regions : » Stability of AOT retrieval need to be improved » Alternative AOT retrieval specific for water areas » Use of SWIR bands » BUT »SNR issues ? »Accuracy of SWIR bands ? » eg. Pahlevan et al (2014) gains for OLI: (1.034 SWIR 1, 1.21 SWIR2 ) 16/06/2015 © 2014, VITO NV
33