Atmospheric correction of remote sensing data. This PPT describes development of a region sensitive atmospheric correction method for hyperspectral image processing
Atmospheric Correction of Remote Sensing Data_RamaRao.pptx
1. Atmospheric Correction of Remote Sensing Data
Global vs Regional perspective , Application potential β¦
Dr. Rama Rao Nidamanuri
Department of Earth and Space Sciences
Indian Institute of Space Science and Technology
Thiruvananthapuram β kerala
rao@iist.ac.in
3. Atmospheric correction : Radiative transfer process
1 Amount of irradiance reaching the terrain
as a function of the atmospheric
transmittance
2 Spectral diffuse sky irradiance
Radiation after scattering (Rayleigh, Mie,
and/or nonselective) and absorption and
reemission
3
Radiation reflected or scattered by nearby
terrain into the IFOV of the sensor system
4
Energy that was also reflected from
nearby terrain into the atmosphere, but
then scattered or reflected onto
the study area
5
Source : Jensen, J. R. (2009). Remote sensing of the environment: An
earth resource perspective 2/e. Pearson Education India.
4. Atmospheric correction : Approaches
Empirical based
ο Flat fielding
ο Quick atmospheric correction
(QUAC)
ο Empirical line methods
Physics / Radiative Transfer
Model based
Works well on small geographic regions
Fails on large area/ time β space series data
ο MODTRAN based (e.g. FLAASH, ATCOR, etc.)
ο Second simulation of the satellite signal
in the solar spectrum (2S ----- 6S)
ο Discrete Ordinates Radiative Transfer
Program for a Multi-Layered Plane-
Parallel Medium (DISORT)
Accurate! ,,,But computationally and financially
expensive
5. Geometrical conditions
Computation of apparent
Reflection
Ground Reflectance
Atmospheric condition
Spectral Bands
Interpolation of scattering
atmospheric Function
Integration over the
Spectral Band
Atmospheric Correction if
requested
Computation of BRDF
Atmosphere Coupling
Aerosol Model(Type)
Computation of
Atmospheric functions
Solar Spectral Irradiance
Aerosol Concentration
Output
Environmental Reflectance
Computation of Gaseous
Absorption Function
Internal view of RT based atmospheric correction
10. Fact from literature: Local Vs Global Aerosol Optical Depth
Aerosol Optical Depth Variation during Summer (2008 & 2014) Carbon Monoxide mixing ratio averaged for (a) May of 2000-2007 and (b)
2000-2007
S. N. Palve et al 2016 IOP Conf. Ser.: Earth Environ. Sci. 37 012076
11. Literature example: Influence of region sensitive aerosol models
Continental - Aeronet Urban - Aeronet
Maritime - Aeronet GEOS-Chem -
Aeronet
Relative difference at 550 nm between the surface reflectance values obtained from the CHRIS@CRI atmospheric
correction algorithm and those obtained with AERONET data, used as a reference.
Tirelli, C., Curci, G., Manzo, C., Tuccella, P. and Bassani, C., 2015. Effect of the aerosol model assumption on the atmospheric
correction over land: Case studies with CHRIS/PROBA hyperspectral images over Benelux. Remote Sensing, 7(7), pp.8391-8415.
12. Model research gap : Region-sensitive atmospheric parameters
5 10 15 20 25
0
5
10
15
20
Column
Water
Vapor
(g/m3)
Altitude (Km)
Standard Profile
Regional Model (JAN)
Regional Model (MAY)
Regional Model (DEC)
5 10 15 20 25 30 35 40 45 50
200
250
300
Temperature
(Kelvin)
Altitude (Km)
Standard Profile
Regional Model (JAN)
Regional Model (MAY)
Regional Model (DEC)
Standard Tropical Atmospheric Profile Vs Indian Humid Sub-Tropical Profile
13. Region sensitive atmospheric correction model : Method
Transmittance
from ground to
sensor
Transmittance
from sun to
ground
Spherical Albedo
Path reflectance
Input Radiance Image
Visibility
Estimate
Water Vapour
Estimate
Radiative Transfer
Inversion Equation
Geometry
Condition
Atmospheric
Model
Aerosol
Model
Flight and
Target
Altitude
Output Reflectance Image
Sensor Spectral
Response
Function
View Zenith Angle
Atmospheric
Model
Aerosol Model
Water Vapour
Aerosol Optical
Thickness
Altitude of the
flight
Altitude of the
target
Solar Zenith Angle
View Azimuth
Angle
Solar Azimuth
Angle
Input Parameters
for the LUT
Neighbourhood
Parameter
Matching
&
N-D
Interpolation
Input Variables
14. Flexible Atmospheric Compensation Technique (FACT): Computational Design
Parameters Range Interval Values
Aerosol Model 6S standard models
Atmospheric Model 6S standard models
View Zenith Angle 0-600 150 0,15β¦..60
Solar Zenith Angle 0-750 150 0,15β¦..75
View Azimuth Angle 0-3600 800 0,80β¦..360
Solar Azimuth Angle 0-3600 800 0,80β¦..360
AOT 0-0.7 - 0.03,0.12,0.2,0.3,0.4,0.7
Water Vapour 0-5 (g/cm2
) - 0,0.4,0.8,1.2,1.6,2,3.5,5
Altitude of the Flight(for
Airborne sensors) 0-6 Km - 1,3,6
Altitude of Target 0-7 Km - 0,1,2,4
πππ =
π
π + ππ Γ π
π¦ = π₯π Γ πΏπππ΄ β π₯π
π₯π =
π
πππ» π½π π» π½π ππ πΈπ
π₯π =
ππ
π ππ π ππ£
ππ = π
Radiative Transfer Inverse Modeling
15. Climatic Zones Used for
Atmospheric Profile Preparation
Indian-Wet Dry Tropical : Gorakhpur Indian Arid : Jodhpur Indian Tropical Wet and Dry : Chennai
FACT : Regionalization of atmospheric parameters
16. FACT : Results and validation β Hyperspectral sensor
0.00 0.02 0.04 0.06 0.08 0.10
0
5000
10000
15000
0.00 0.02 0.04 0.06 0.08
0
10000
20000
30000
Pixel
count
Mean absolute error (0-1)
FLAASH Vs FACT
FLAASH Vs NASA JPL
NASA JPL Vs FACT
Pixel
count
Root mean square error (0-1)
FLAASH Vs FACT
FLAASH Vs NASA JPL
NASA JPL Vs FACT
RMSE of 0.0563 for
99.9% of total pixel:
FACT and FLAASH
Results of atmospheric correction of AVIRIS-NG imagery: (a) β (c) False Colour Composites of the corrected imagery from the
proposed FACT, FLAASH and NASA JPLβs atmospheric correction scheme respectively; (d) β (f) Root Mean Square Error (RMSE) image
between the proposed FACT-FLAASH, FLAASH-NASA JPLβs and FACT-NASA JPLβs scheme respectively
(a) (b) (c)
(d) (e) (f)
(f)
17. FACT : results and validation β multispectral sensor
0.00 0.01 0.02 0.03
0
100000
200000
300000
Pixel
count
Mean absolute error (0-1)
0.00 0.01 0.02 0.03
0
100000
200000
300000
400000
Pixel
count
Root mean square error (0-1)
Results of atmospheric correction of WorldView-3 imagery: (a) False Colour
Composites of the corrected imagery from the proposed FACT, (b) Mean Absolute
error (c) and Root Mean Square Error image between the proposed FACT and
FLAASH atmospheric correction method
The RMSE histogram for the
WorldView-3 imagery
calculated between
atmospherically corrected
image by FLAASH and the
proposed FACT
18. 500 1000 1500 2000 2500
0.00
0.05
0.10
0.15
0.20
500 1000 1500 2000 2500
0.00
0.01
0.02
0.03
0.04
0.05
500 1000 1500 2000 2500
0.00
0.01
0.02
0.03
0.04
0.05
400 600 800 1000
0.00
0.01
0.02
0.03
0.04
0.05
Band
RMSE
(range
0
to
1)
Wavelength (nm)
FLAASH Vs FACT
FLAASH Vs NASA JPL
NASA JPL Vs FACT
Band
RMSE
(range
0
to
1)
Wavelength (nm)
Band
RMSE
(range
0
to
1)
Wavelength (nm)
Band
RMSE
(range
0
to
1)
Wavelength (nm)
(a) (b)
(c) (d)
FACT : results and validation β spectral measure
A spectral analysis (band wise RMSE calculation) for the imagery data of (a) AVIRIS-NG (b) Hyperion (c) LANDSAT-8
OLI and (d) WorldView-3 corrected by the FLAASH, the NASA JPLβs and the proposed FACT correction method in case
of AVIRIS-NG and by the FLAASH and the proposed FACT in other three cases.
ππ β ππ
2+ . . . + ππ β ππ
2
k
πΏ = FACTβs estimated reflectance cube
π = Reference reflectance Cube
k = Total number of bands
19. FACT 2.0 : Summary
ο§ Development of open-source atmospheric correction method FACT was
successfully carried out
ο§ Spectral analysis shows marginal disagreements between FACT and other
atmospheric correction methods in water absorption region
ο§ Spectral conformity of about 95% for hyperspectral imaging sensors and
about 98% for multispectral imaging sensors in comparison to other
atmospherically corrected imagery is achieved
ο§ Overall performance of FACT is found to be within 3-5% error margin,
which is considered satisfactory
20. Simulation and modelling of hyperspectral target detection
under varying atmospheric state variables
Example applications of atmospheric influences
Design and conduct experimental studies of the role played by
atmospheric state variables such as atmospheric profile (function
of columnar temp, pressure, ozone and water vapour), aerosol
composition (function of particle size and density), on target
detection performance
21. Literature example: Impact of atmospheric correction models on image classification
ICOR
DOS
TOA
S2AC
Ground Truth
Satellite Imagery
Rumora, L., Miler, M. and Medak, D., 2020. Impact of Various Atmospheric Corrections on
Sentinel-2 Land Cover Classification Accuracy Using Machine Learning
Classifiers. ISPRS International Journal of Geo-Information, 9(4), p.277.
23. Multi-platform TD : Target positioning
AVIRIS-NG HSI
Black and Yellow
nylon sheet targets
Red, Green and White
colour nylon sheet
targets
Ground
Truth Map
Yellow Nylon
sheet (N3Y)
Black Nylon
sheet (N4B)
Red Nylon
sheet (N2R)
Green nylon
sheet (N1G)
White Cotton
sheet (C1W)
29. Results : Impact of varying aerosol optical thickness (AOT)
0 1 2 3 4 5
0.000
0.036
0.072
0.108
0.144
P
FA
(@P
D
=75%)
Aerosol Optical Thickness (AOT)
0.0192
0.0240
0.0288
0.0336
0.0384
0.0001
0.0002
0.0003
0.0004
0.0005
0 1 2 3 4 5
0.370
0.444
0.518
0.592
0.666
P
FA
(@P
D
=75%)
Aerosol Optical Thickness (AOT)
0.072
0.078
0.084
0.090
0.096
0.000
0.056
0.112
0.168
0.224
0 1 2 3 4 5
0.0
0.2
0.4
0.6
0.8
Aerosol Optical Thickness (AOT)
0.0128
0.0192
0.0256
0.0320
0.0384
P
FA
(@P
D
=75%)
0.0
0.2
0.4
0.6
0.8
1.0
0 1 2 3 4 5
0.684
0.722
0.760
0.798
0.836
P
FA
(@P
D
=75%)
Aerosol Optical Thickness (AOT)
0.00124
0.00186
0.00248
0.00310
0.00372
0.000
0.048
0.096
0.144
0.192
0 1 2 3 4 5
0.34
0.36
0.38
0.40
0.42
P
FA
(@P
D
=75%)
Aerosol Optical Thickness (AOT)
0.002
0.004
0.006
0.008
0.010
0.00
0.02
0.04
0.06
ACE
MF
CEM
N1G N2R
C1W N3Y
N4B
30. Results : Impact of atmospheric and aerosol models (N1G)
N-Aero. Cont. Mar. Urb. Des.
0.000
0.021
0.042
0.063
0.084
0.105
0.126
0.147
N-Aero. Cont. Mar. Urb. Des.
0.000
0.023
0.046
0.069
0.092
0.115
0.138
0.161
N-Aero. Cont. Mar. Urb. Des.
0.000
0.094
0.188
0.282
0.376
0.470
0.564
0.658
N-Aero. Cont. Mar. Urb. Des.
0.0000
0.0038
0.0076
0.0114
0.0152
0.0190
0.0228
0.0266
N-Aero. Cont. Mar. Urb. Des.
0.0000
0.0038
0.0076
0.0114
0.0152
0.0190
0.0228
0.0266
N-Aero. Cont. Mar. Urb. Des.
0.0000
0.0071
0.0142
0.0213
0.0284
0.0355
0.0426
0.0497
N-Aero. Cont. Mar. Urb. Des.
1.40x10-4
1.43x10-4
1.46x10-4
1.48x10-4
1.51x10-4
1.54x10-4
1.57x10-4
1.60x10-4
N-Aero. Cont. Mar. Urb. Des.
1.39x10-4
1.43x10-4
1.47x10-4
1.51x10-4
1.55x10-4
1.60x10-4
1.64x10-4
1.68x10-4
N-Aero. Cont. Mar. Urb. Des.
1.58x10-4
2.37x10-4
3.16x10-4
3.95x10-4
4.74x10-4
5.53x10-4
6.32x10-4
7.11x10-4
P
FA
(@P
D
=75%)
P
FA
(@P
D
=75%)
P
FA
(@P
D
=75%)
Aerosol Model Aerosol Model Aerosol Model
AOT=0.48
AOT=0.89
AOT=4.94
ACE CEM MF
Tropical
MidLatitude Summer
MidLatitude Winter
Subartic Summer
Subartic Winter
US62
Atmospheric Profiles
Aerosol Models
N-Aero : No Aerosol
Cont : Continental
Mar : Maritime
Urb: Urban
Des : Desert
31. Conclusions : Ramifications of mismatch of atmospheric state
variables on target detectability
ο§ Visibility (AOT)
β Substantial variation in detectability by different algorithms
β Different detectability from material to material
β Effect of AOT is profound after certain threshold, such as
after AOT=2.5
ο§ Aerosol and atmospheric models
β Inherent randomness in performance
β Performance heavily penalized for the wrong choice of
atmospheric model when compared to a mismatch of
aerosol model