1. Class 8: Radiometric Corrections
Sensor Corrections
Atmospheric Corrections
Conversion from DN to reflectance
BRDF Corrections

2. Two types of corrections: Geometric Corrections (class 5)
Radiometric Corrections (class 8)
Remote sensing images are contaminated by various radiative processes. The need to correct them varies with the applications and sensors used.
Every time two images need to be combined (e.g., in a mosaic) or compared, the corrections become obviously important.

3. Radiometric Correction
Correction is made on the brightness (gray level)
values of the image.
Source of errors to be corrected:
atmospheric degradation
sensor malfunctions
Illumination-view geometry
Corrections are usually different
for each band, and in theory for each pixel
Attempts to correct data may themselves introduce errors
Campbell 10.4

4. Radiometric Corrections
Correction for detector errors
Line drop
Destriping
2. Atmospheric corrections
Histogram adjustment
Atmospheric radiative transfer models
3. Conversion from DN to radiance
4. Conversion from radiance to reflectance
5. BRDF corrections

5. Sensor corrections Line Dropout Solution: Or use other spectral band43 47 51 57 40 46 50 54 38 42
Mean from above
and below pixels
Solution: 43 47 51 57 40 46 50 54 38 42 39 52
Or use other spectral band
Images: Lillesand-Kiefer
Campbell 10.4

6. Sensor corrections Striping Local averaging Normalization
Images: Lillesand-Kiefer
Campbell 10.4

7. Atmospheric Corrections
Histogram adjustment
Clear sky
Hazy sky
2) Physical Models
Campbell 10.4

8. Simple Atmospheric Corrections – Histogram Adjustment
Clear Atmosphere
Cloud shadowed region and water
bodies have very low reflectance
in infrared bands. This should give
a peak near zero on the histogram.
The shifted peak is due to
the low reflectance regions
with atmospheric scattering.
A correction can be obtain by
removing this value from
all pixels.
This method is called the
Histogram Minimum Method (HMM)
Narrow range of
brightness values
Small atmospheric
contribution to
brightness
Brightness values
Darkest values near zero
Campbell 10.4

9. Simple Atmospheric Corrections – Histogram Adjustment
Hazy Atmosphere
Wide range of
brightness values
In this case, the minimum
value is higher, and the
histogram shape has changed
Added brightness
of atmosphere
Brightness values
Darkest values far from zero
Campbell 10.4

10. Atmospheric Correction Models
Physical models simulate the physical process of scattering at the level of individual particles and molecules
Absorption by gases
scattering by aerosols
LOWTRAN 7
MODTRAN
CAM5S, 6S
Complex models that need many
meteorological data as input.
The data may not always be available
Campbell 10.4

11. Atmospheric Correction Models
Second Simulation of the Satellite Signal
in the Solar Spectrum: 6S
Input file example (Saskatchewan study site; Landsat imagery):
(landsat TM)
(month,day,hour,long,lat)
(mid lat summer)
(continental)
(visibility, km)
(TARGET ALTITUDE IN KM)
(SATELLITE CASE)
(Landsat band 1)
(HOMOGENEOUS CASE)
(NO BRDF effect)
(uniform target = vegetation)
(no atm. correction)

12. Atmospheric Correction Models ASAS Konza prairie reflectance spectrum
6S corrected reflectance
Top of atmosphere reflectance
ASAS Konza prairie reflectance spectrum
ASAS band central wavelength (nm)
Vermote et al., 1997

13. From DN to Radiance to Reflectance
Calibration Gain
Coefficient
(counts/(W/m2/sr/mm))
Characteristic
Wavelength
(mm)
Solar
Irradiance
(W/m2/mm)
LANDSAT TM
Spectral
Band
1 G=(-3.58E-05)*D
2 G=(-2.10E-05)*D
3 G=(-1.04E-05)*D
4 G=(-3.20E-06)*D
5 G=(-2.64E-05)*D
7 G=(-3.81E-04)*D
D = days since launch
The revised gain coefficient equations incorporating 1994 calibration updates not included in the IGARSS '94 paper are as follows, where D = days since launch (01-March-1984), L* = (DSL - Offset)/G, and DSL = digital signal level.
E (irradiance) = EoCos(SZA)/d^2
Radiance = (DN - Offset)/Gain
Reflectance = p.Radiance/Incident Solar Irradiance
Incident Solar Irradiance=Solar Irradiance *cos(SZA)
Source:
CCRS Web site

14. This is a common occurrence in land remote sensing systems
If the input signal exceeds the amount for which the sensor was designed,
the system response will become non-linear or reach the saturation level.
This is a common occurrence in land remote sensing systems
when they image bright clouds and/or snow cover, for example.
Saturation
Non-Linear Region
y (DN)
Linear Region y = a.x + b
(DN = gain*Radiance + offset)
Offset b
Input Value x (radiance)
Source:
CCRS Web site

15. Atmospheric Corrections
Ltot= radiance measured
by the sensor
r = reflectance of the target
E = irradiance on the target
T = transmissivity
of the atmosphere
Lp= path radiance (radiance
due to the atmosphere)
L & K 7.2

16. Atmospheric Corrections
E0 coss
E = d2
E0 = solar irradiance at the mean Earth-Sun distance
s =solar zenith angle
d = relative deviation of Earth-Sun distance from the
mean distance at the time of imaging
L & K 7.2

17. Bidirectional Reflectance Distribution Function (BRDF) Correction
Structures like trees cast shadows that change the amount of light that reaches a sensor depending on its
view zenith angle
To compare pixel reflectance from different images,
or even different part of an image, the target (pixel) reflectance must be measured under the same
view and solar geometry.
Sensor
Solar Zenith Angle
(SZA)
View Zenith Angle (VZA)

18. CCRS uses a modification of Roujean’s model
Some BRDF models
CCRS uses a modification of Roujean’s model
for BRDF corrections of AVHRR data
(Roujean + hotspot from 4-Scale, Chen and Cihlar, 1997)
GORT (Li and Strahler)
4-Scale (Chen and Leblanc)