This document discusses several approaches for atmospheric correction of remote sensing imagery: 1) Image-based methods like the dark pixel method and regression method estimate and remove atmospheric path radiance. 2) The empirical line method uses ground targets of known reflectance to model atmospheric effects. 3) Radiative transfer models precisely account for atmospheric conditions using numerical models like MODTRAN or 6S to convert pixel values to surface reflectance. 4) Relative correction methods normalize images without absolute calibration to surface reflectance. Atmospheric correction is needed to accurately analyze surface properties from remote sensing data and compare images acquired at different times or wavelengths.

1. APPROACHES FOR
ATMOSPHERIC
CORRECTION
NIRMAL KUMAR
AP:140

2. Atmospheric correction
To retrieve surface reflectance from RS imagery
 relationship between radiance
received at the sensor (above
atmosphere) and radiance
leaving the ground
Ls =H.ρ.T + Lp
Ls – at sensor radiance
H – total downwelling radiance
ρ – reflectance of target
T – atmospheric transmittance
Lp – atmospheric path radiance
(wavelength dependent)

3. Why do atmospheric correction?
 Physical relation of radiance to surface
property (surface normal, surface roughness, reflectance).
 Atmospheric component needs to be removed
 Image ratios (NDVI) leads to biased estimate
 Scattering increases inversely with wavelength
 The involved channels will be unequally affected
 Time difference between image acquisition and
ground truth measurements
 Comparison of RS data captured at different times
 Conditions may be different

4. Atmospheric correction methods
 Image – based methods
 Dark pixel method
 Regression method
 Empirical line method
 Radiative transfer models
 Relative correction method (PIFs)

5. Dark pixel subtraction method
Ls =H.ρ.T + Lp
 Pixel
values of low reflectance areas
near zero
 Exposure of dark colored rocks
 Deep shadows
 Clear water
 Lowest pixel values in visible and NIR are
approximation to atmospheric path
radiance
 Minimum values subtracted from image

6. Regression method
 NIR pixel values are plotted against values
in other bands
 Apply a straight line using the least square
method
 If there was no haze, the line would pass
through origin
 resulting offset is approximation for
atmospheric path radiance
 offset subtracted from image

7. Empirical line correction method
 Use target of “known”, low and high reflectance
targets in one channel e.g. non-turbid water & desert,
or dense dark vegetation & snow
 Assume radiance, L = gain * DN + offset
 Offset is assumed to be atmospheric part of signal
Radiance, L
Offset assumed to be atmospheric
path radiance
Regression line L = G*DN + O
Target DN values
DN

8. Conversion of DNs to absolute
radiance value
3 steps
• Convert DN to apparent radiance Lapp
• Convert Lapp to apparent reflectance (knowing
response of sensor)
• Convert to at-ground reflectance i.e. intrinsic surface
property by accounting for atmosphere
Use Radiative transfer models

9. Radiative transfer models
 Limited by the need to supply data about
atmospheric conditions at time of acquisition
 Mostly used with "standard atmospheres"
 Available numerical models
 􀁸 LOWTRAN 7
 􀁸 MODTRAN 4
 􀁸 ATREM
 􀁸 ATCOR
 􀁸 6S (Second Simulation of the Satellite Signal in
the solar spectrum)

10. THANK YOU