atm. corr. - usgs · 2018. 12. 17. · atm. corr. scene selection image resampling geo-correction...
TRANSCRIPT
Atm. Corr.Scene selection
Image ResamplingGeo-correction
SpectralResponse
AC analysisSurf. Refl.
TOA Rad. Refl.L8_OLI vs S2_MSI
OLI AC MitigationOLI – Planet
Rad. Correction
≈ 𝑨𝒕𝒎 + 𝑺𝒖𝒓𝒇𝒇( , 𝑨𝒕𝒎, 𝝆)
SZA Gases ( . . . , O3 , H2O )VZA Pressure∆AA Aerosol ( Type, AOT )
TOAR
Atm.
Surf.
SR
PlanetScope GEOTIFF SR file information{"atmospheric_correction": {"aerosol_model": "continental""aot_coverage": 0.5"aot_mean_quality": 1.0"aot_method": "fixed""aot_source": "mod09cma""aot_status": "Data Found""aot_std": 0.0021659541055997905"aot_used": 0.01755555470784505"atmospheric_correction_algorithm":"6Sv2.1""atmospheric_model": "water_vapor_and_ozone""luts_version": 3"ozone_coverage": 0.5"ozone_mean_quality": 255.0"ozone_method": "fixed""ozone_source": "mod09cmg""ozone_status": "Data Found""ozone_std": 0.0"ozone_used": 0.2924999925825331
"satellite_azimuth_angle": 0.0"satellite_zenith_angle": 0.0"solar_azimuth_angle": 150.32522363"solar_zenith_angle": 55.4013318347"sr_version": "1.0""water_vapor_coverage": 0.5"water_vapor_mean_quality": 1.0"water_vapor_method": "fixed""water_vapor_source": "mod09cma""water_vapor_status": "Data Found""water_vapor_std": 0.0099380845107"water_vapor_used": 0.51888889736}}
100 x 100( 30 m pixel )
1000 x 1000( 3 m pixel)
L8_OLI
PlanetScope300 x 300 ( 10 m pixel )
S2 MSI
?
Year Month Date
L8 OLI
100 x 100( 30 m pixel )Center : AERONET
1000 x 1000( 3 m pixel)
L8_OLI PlanetScope
E0
N0
E0N0
Image resamplingE0+30
N0+300 100
No Image ShiftL8_OLI30m native
PlanetScopeResampled to 30m
Excessive dispersion Relative georeferencing error
100 x 100( 30 m pixel )Center : AERONET
1000 x 1000( 3 m pixel)
L8_OLI PlanetScope
E0
N0
E0+∆E
N0+∆N
Image Shift( ∆E, ∆N ) E0+∆E+30
N0+∆N+300 100
Easting : -3 pixel ( - 9 m )
Northing : 8 pixel ( 24 m )
Spatial shift : L8_OLI vs Planet ∆E=-30 ∆E=+30
∆N=+30
∆N=-30
-3, 8
-3, 8
-3, 8
-3, 8
Image Shift( 0, 0 ) ( -3, 8 )
+ 𝜕𝑁𝐷𝑉𝐼
𝜕𝜌𝑁𝐼𝑅|𝜌𝑅,𝜌𝑁𝐼𝑅 · σ𝜌𝑁𝐼𝑅
2
= 𝜕𝑁𝐷𝑉𝐼
𝜕𝜌𝑅|𝜌𝑅,𝜌𝑁𝐼𝑅 · σ𝜌𝑅
2σ𝑁𝐷𝑉𝐼
2
Strictly speaking, mismatch in SR will cause error in the downstream application
SR apparent mismatch
Other reasons for the mismatch?
* Spectral Response * Atmospheric Correction* Improper Radiometric Correction
( Stray light, post-launch drift )
* Georeferencing error ? Checked
S R
0
0.2
0.4
0.6
0.8
1
400 450 500 550 600 650 700 750 800 850 900
rela
tive
sp
ect
ral r
esp
on
se [
]
wavelength [nm]
In-Band Band-Average Relative Spectral Response
0
0.2
0.4
0.6
0.8
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
Blue_0f10 Green_0f10 Red_0f10 NIR_0f10Blue_0c0d Green_0c0d Red_0c0d NIR_0c0d
Landsat 8 OLI
Planet
Sentinel2 MSI
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI vs Planet
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI vs Planet
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI < Planet
𝜌 =∑ℜ𝑖𝜌𝑖∑ℜ𝑖
𝜌 =𝜌 𝜆 ∙ ℜ 𝜆 ∙ 𝑑𝜆
ℜ 𝜆 ∙ 𝑑𝜆
0.201 0.212
0.0940.106
5 %sand
10%veg+soil
20%DDV
0.0390.048
0.0410.042
2 %Dark soil OLI
Planet
DDV
dry V
Sand
No constant conversion! class-dependent
Interoperability ???( hardware protocol )
Asphalt
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI > Planet
0.250 0.236
0.1500.133
7 %
14%
25%0.0860.068
OLI
Planet
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI vs Planet
0.272 0.266
0.1480.146
≈
≈
20%0.0440.053
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9
OLI vs Planet
0.3060.299
0.5610.468
≈
20%
≈0.7050.693
Spectral Response Function IssueThe Correction factor should be different for
* varying reflectance (dark to bright) * different classes (spectral shape)* different bands (channels) Ideally, (1) extensive modeling study (2)classification(3) apply varying degree of correction based on (1) & (2)
( it is very difficult, if not impossible ) Let’s leave it uncorrected. This effect definitely adds
substantial uncertainty in SR comparison
Atmospheric Correction
Non-vegetated wet dark soil pixel
Vegetation pixel
Bare soil pixel
Non-vegetated wet dark soil pixel
AOT = 0.01 0.05 0.1 0.15
If AOT SR If AOT SR
AOT = 0.02 AOT = 0.035
AOT (CWV, O3) in ACshould be no issue !
TOAR SR
SR
TOAR
SR
TOARSoil
Veg
Soil
Veg
Root cause isnot AC,but Radiometry.
Radiance Invariance law
TOA Radiance : L8_OLI vs PlanetScope
OLI vs MSI Easting : -1 pixel ( -10 m )
Northing : 2 pixel ( 20 m )
Spatial shift : L8_OLI vs MSI
OLI vs PlanetEasting : -3 pixel ( - 9 m )
Northing : 8 pixel ( 24 m )
(One day ahead,similar local time)
(0, 0) No Image Shift (-1, 2) (-10m, 20m) Image Shift
TOA Radiance TOAR
Pre-launch calibrationOn-orbit monitoring, correctionStraylight correction, …
Earth-Sun distanceSolar zenith angleSolar Flux (irradiance) all, precisely known
~ no uncertainty
SRTOAR
Overall spectral pattern looks decent.It is not easy to really screw up AC !!!
Atm. Corr.
TOARTOAR
𝐿𝑔𝑟𝑜𝑢𝑛𝑑, 𝑜𝑢𝑡𝑔𝑜𝑖𝑛𝑔
𝐿𝑠𝑒𝑛𝑠𝑜𝑟400𝑘𝑚, 𝑖𝑛𝑐𝑜𝑚𝑖𝑛𝑔
Atmosphere
𝐿𝑠𝑒𝑛𝑠𝑜𝑟700𝑘𝑚, 𝑖𝑛𝑐𝑜𝑚𝑖𝑛𝑔
Radiance Invariance Law
𝑓 ≈ 1/𝑛2
𝐿𝑔2 ∙ 𝑓 = 𝐿400
2
𝐿𝑔2 ∙ 𝑓 = 𝐿700
2
Regardless of* Altitude* Aperture* IFOV
Superb spatial resolutionHigh temporal frequency
But, Poor radiometric quality
Global radiometric correction??
TOAR
175 +
ID=1018
Sep. 8 2018WRS2 (Path=033)
Denver, CO
Correction coefficients derived from Denver imagefor Camera ID 1018
Crestone, CO
Rapid City, SD
Cheyene, WY
LaSRC (OLI AC) :(1) solar & view geometry (θs,θv,Δφ)
(2) image-derived ( τa550 )
(3) external data (P, H2O , O3)
Pressure(DEM)
H2OMODIS
OzoneMODIS
MODIS
Forget about external data (water vapor, ozone)but, focus on the AOT algorithm!
DEM7200 x 3600
GSD = 5.5km (along equator, D = 12800 km)
Lat/Lon gridΔ=0.05o
Denver
Sioux Falls
[ Meter ]𝑃 ℎ = 1013𝑒−ℎ(
𝑔𝑅𝑇
)
= 𝟏𝟎𝟏𝟑𝑒−ℎ/8500
?
Realistic sea-level pressure range ( 1010 – 1030 mb)
P = 1000 mbP = 1050 mbSZA = 20o
P = 1000 mbP = 1050 mbSZA = 70o
SZA, VZA ?
𝑃 ℎ = 𝟏𝟎𝟏𝟑𝑒−ℎ(𝑔𝑅𝑇
)
OZONE
Scale factor: 400
[cm atm]
300 DU
340 DU
200 DU( 0.2 atm cm )
500 DU( 0.5 atm cm )
Chappuis absorption
200 DU
400 DU
200 DU300 DU400 DU
Ozone Sensitivity Analysis : 200 - 400 DU
Vegetation Dark soil Brighter soil
Strong CHL signature
Strong Ozone absorptionLow Ozone Transmission
Weak CHL signature
Weak Ozone absorptionHigh Ozone Transmission
Varying vegetation signature strength (spatio-temporal)& varying ozone amount mutually compensate!
No unique ozone solution ! infinite solutionsVegetation pixel is not useful for ozone inversion !
Underestimated Ozone
Overestimated Ozone
Sand (soil, road, asphalt, … ) with straight spectral feature
may provide a chance to estimate ozone!
The question is : Does a (sand, soil-like) spectra have a straight or constant universal spectral curvature ?
4 known bands : B1, B2, G, R2 unknowns : AOT, ozone
Non-liner optimization
Simultaneous solution ofAOT & ozone
2018
May
June
July
August
July 7, 2018
July 23, 2018
AOT = 0.07
AOT = 0.08
CWV = 2.2
Ozone= 300
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
0.229
0.261
0.331
0.360
0.308
+ 7.6 %
210310410
O3
Ground truth ozone = 300 DUEM-based estimation = 310 DU
If spectral signature is (pseudo) invariant Accurate ozone estimation is possible!
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
5-Jul 23-Jul 15-Aug 2-Sep
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
28-May 14-Jun 26-Jul 8-Aug
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
11-Apr 30-Apr 22-May 1-Jun
Surface Reflectance
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
RVUS_avg
BTCN_avg
GONA_avg
TOAR = 𝒇( 𝑨𝑶𝑻,𝑶𝒛𝒐𝒏𝒆, 𝑪𝒔)
4 known bands : B1, B2, G, RNon-liner optimization
( Levenberg-Marquardt )
Ozone = 347(regular pixel)
Ozone = 420(brighter pixel)
0.15
0.2
0.25
0.3
0.35
400 450 500 550 600 650 700
5-Jul 23-Jul 15-Aug 2-Sep
RVUS yearly Ozone range : 280 – 320
negligible difference
Uncertainty due to using global average
200 - ( 300 ) – 400 DU 32.3 - (33) - 33.7
MODISOzone
305 DU
320 DU
200 ( 300 ) 400 DU32.3 (33%) 33.7 ( 2%)
270 (350) 430 DU: less than 1% error !!!
T2-UncertaintySummaryReport-RVUS.pdf
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
5-Jul 23-Jul
15-Aug 2-Sep
BOA Reflectance (Railroad Valley Playa)(Intra-day & daily variation combined)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
400 450 500 550 600 650 700
Pseudo-Invariant ?
Summary : Ozone mitigation
(1) Sand-like EM-based approach allows ozone estimation(2) Unique spectral curvature does not exist (only site specific PI)(3) Blind use of global average ( 300-350 DU) creates
acceptable error ( ~ 1% relative error)(4) Considering TOA & BOA measurement uncertainty ( ~4-5%)
1% relative reflectance error is definitely acceptable!(5) No worry, there always will be ozone satellite!
Water vapor (CMG)
Scale factor : 200
[ g /cm2 ]
0.3
2.5
0
0.2
0.4
0.6
0.8
1
1.2
0.35 0.45 0.55 0.65 0.75 0.85 0.95
column water vapor 0.3 cm
column water vapor 3.0 cm
[g/cm2]
[g/cm2]
H2O [g/cm2]0.51.02.03.04.0
1 g/cm2 5 g/cm2
L8 OLI bands immune to the water vapor !!
SWIR2.1 bandH2O sensitivity
Need for 940 nm H2O bandin future L10