Digital Image Processing:

Image Correction
(Radiometric)

GMA 320
SUT 3
06 August 2024
GGM Department
University of Pretoria
By: Dr. Philemon Tsele
 Outline
• Digital image acquisition process
• Sources of error in remotely-sensed imagery
• Types of radiometric corrections
– Simplified reflectance
– Improved surface reflectance
– True surface reflectance
– Temperature (K)
• Options for atmospheric corrections
Digital image acquisition process
• Surface radiance (Ls) at a
certain spatial coordinates (i,j)
and wavelength (k) is
converted to a list of digital
values, one for each band.
Later, those values can be
converted back to radiances.
 Sources of error in satellite imagery
• The process of acquiring satellite images
includes potential errors in both the radiance
detected and the position of resulting pixels
• Most frequent problems can be grouped into 4
categories
– Errors caused by the sensor platform
– Distortions produced by the Earth’s curvature
– Problems caused by the sensor operation
– Distortions created by the atmosphere
 Examples of Radiometric Corrections
• Solving mixed or deteriorated data
– Restoration of missing lines and pixels**
– Correction of stripping effects
• Conversion from DN to Radiance
• Conversion from Radiance to Reflectance
– Simplified reflectance (TOA)
– Atmospheric correction (BOA)
– Correction of bi-directional effects
– Topographic shadow corrections
• Conversion from thermal DN to Radiance>LST
 • Restoration of missing lines and pixels (1)
– Random problems in sensor performance, communications or reception
problems (interferences) cause missing lines/ pixels
– A simple way to estimate missing/ corrupt pixels is to rely on neighboring pixels
(i.e. spatial autocorrelation)
– Radiance values tend to be correlated with distances (closer areas, more similar;
farther areas, less similar)
– Several ways to use this vicinity criteria (Eq: 1-3) including a 3 x 3 moving window

Substitution of the DNs of each line/pixel by the preceding values, Eq. 1

DNi , j = DNi −1, j Eq. 1
where: DNi,j denote the digital value of the pixel in line i and column j (the corrupt one),
Dni-1,j denote digital value of the pixel in the preceding line

Substitution of the missing pixels by the average values of previous and subsequent values, Eq. 2
DNi , j = INT {( DN i −1, j + DNi +1, j ) / 2 } Eq. 2
Substitution of the missing pixels using pixels values in the second band (covar, corr) Eq. 3
s
DNi , j ,k = ( k ){ DNi , j ,r − ( DNi +1, j ,r + DNi −1, j ,r ) / 2 } + ( DNi +1, j ,k + DNi −1, j ,k ) / 2 Eq. 3
sr
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis
• Correction of Stripping Effects (2)
• Figure below depicts a typical Landsat MSS scenario, where 6 of its detectors does
systematic data recording/ scanning (simultaneously).
• However, one of the detectors (2nd) provide systematically lower signal than the rest
• This is called n-line stripping
Difference in means and standard deviations
between the detector’s recorded DN values suggests
that the detectors are not properly calibrated

range

Low DN High DN

Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis
• Correction of Stripping Effects…cont
– The process starts by computing the mean and standard deviation of the pixels
sampled by each detector (partial histograms). Then, the histograms are adjusted
to the reference histogram using Eq. 4:

DNi , j' = ai + bi DNi , j Eq. 4

where: ai and bi are the correction coefficients, computed as follows in Eq. 5 & 6
s
bi = Eq. 5
si

ai = DN − bi DN i Eq. 6
where: s and si indicate the standard deviation of the reference and partial histogram, respectively
and DN and DN i the mean DN for the reference and partial histogram

Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis
This image exhibits stripping every 16 lines
Landsat TM has 16 detectors
• Conversion to Physical units (3)
– Reflectance:
• Calibration (DL to radiance)….3.1
• Simplified reflectance….3.2
• Atmospheric correction….3.3
• Illumination correction.
• BRDF correction.
– Temperature:
• Calibration (DL to radiance).
• Atmospheric correction.
• Emissivity correction.
• Cloud masking.
• Image Calibration
Most space agencies use a terminology that identifies
satellite products according to their level of processing
Level Meaning
0 No correction has been applied
1 Geometric and radiometric corrections applied
2 Calculation of a secondary physical variable
associated to radiance
3 Resampling the original resolution to a certain
standard grid
4 Derived variable such as Land cover,
Evapotranspiration
 Sensor calibration
• Implies converting DNs to radiance
• DNs do not have an absolute physical meaning
• Range of radiance values that a sensor detects is
termed radiometric resolution
• Calibration follows a linear function DN

Radiance (W m-2sr-1μm-1)

Reflectance (%) Temperature (K)

 MANUAL-derivation GIAN & BIAS
• DNs can be converted back to radiance
calibration by first determining the calibration
coefficient for each band (k)

Eq. 7

• Bias coefficient, a0,k is also needed

Chuvieco (2020): Fundamentals of Satellite Remote Sensing, Taylor and Francis

• Converting DL to Radiance (linear function)…(3.1)
– Pre- and post-launch calibration of the sensor makes it possible to convert DNs to
radiance using a set of (band specific) calibration coefficients such as

Lsen ,k = a0 ,k + a1,k ( DN k ) Eq. 8

where: Lsen,k is the spectral radiance at the sensor in band k (in Wm-2 sr-1 µm-1),
a0,k and a1,k are the calibration coefficients for band k, and
DNk the digital number of a pixel in band k

-Note, calibration coefficients are usually

included in the image header file
(metadata file) or are provided by the
sensor manufacture

-Thus the conversion of DNs to simplified

reflectance is done in two steps: first, DNs
are converted to radiance values (Lsen,k)
through the use of sensor calibration
coefficients; while the simplified
reflectance P*k requires solar irradiance at
TOA lat/long of study area, day of year
acquisition including solar zenith angle
Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis
• Calculation of Simplified Reflectance (3.2)
– Reflectance is defined as energy reflected from a surface in the direction of the
sensor normalized by the incident energy on that target

– Reflectance can vary between 0 and 1

– Varying spectral signatures for various surface features/ land cover types

Chuvieco and Huete (2009): Fundamentals of Satellite Remote Sensing, Taylor and Francis
Simplified reflectance…cont
• Assumes sensor observed Earth
through: 1) transparent
atmosphere, 2) nadir view angle,
and 3) perfect Lambertian surface
• Typical of TOA – top of the
atmosphere reflectance

Apparent reflectance

Ideal diffuse reflection (Lambertian surface)

 Linear relationship between DNs and Reflectance

DNs are not practically comparable between images of different radiometric res
 Bidirectional effects due to sun-target-sensor geometry

Varying light intensities

Diffuse reflection with a

directional component
(not Lambertian)
On the left the sun is infront the camera; on the
right the sun is behind the camera.

Not considered in calculating simplified (apparent) reflectance: Lsen/ Lsun

 Improved Simplified Reflectance (3.2.1)
• Improved simplified reflectance
ought to take cognisance of
the following variables:
E0,k : solar irradiance at ToA in band k
Price, J.C., 1987. Calibration of satellite radiometers and the comparison of
vegetation indices. Remote Sensing of Environment, 21(1), pp.15-27.

 *k : apparent reflected radiance of the vertical ray

surface in band k
Solar ray

 i : zenith angle of the incident flux

(solar elevation angle)
Apparent reflectance
J : day of acquisition
2 ( J − 93.5) 2 : correction factor accounting for variations of the Earth-Sun
D = (1 + 0.01674(sin( )))
365 distance. Varies between 0.983 (January 3) at perihelion and
1.017 (July 4) at aphelion. J is the day of year, and sine term
is in radians
E0,k cos i  *k
Lsen ,k = Eq. 9
D
Doesn’t consider atmospheric influence & sensor observation (off-nadir) angles
• Correction factor accounting for variations
of the Earth-Sun distance
 Atmospheric Correction (3.3)
• Two components are needed to compute accurate
surface reflectance

• The atmosphere modifies both the downward

solar irradiance (1) reaching the Earth’s surface,
and the upward-surface leaving radiance (2) in the
direction of the satellite sensor

• These two components differ in that the path

length or thickness of the atmosphere is different
for these two directions

• The radiance measured by a satellite sensor is not

the same as the radiance leaving the Earth’s surface
due to the atmospheric attenuation
 Upward-surface leaving radiance
• The atmosphere has an important influence in modifying both the downward
solar irradiance (1) reaching the Earth’s surface, and the upward-surface
leaving radiance (2) in the direction of the satellite sensor
• These two components (1) and (2) are needed Transmittance is the surface
to compute accurate
reflectance counterpart of Absorptance
• These two components differ in that the path length or thickness of the
atmosphere is different for these two directions
• Thus the radiance measured by a satellite sensor is not the same as the
radiance leaving the Earth’s surface due to the atmospheric attenuation
• The radiance at the sensor is related to the following parameters Eq. 10:

Lsen ,k = Lg ,k k ,0 + L p ,k Eq. 10

where: Lsen,k is the spectral radiance at the sensor in band k (in Wm-2 sr-1 µm-1),
Lg,k is the apparent surface reflected radiance in band k, and
 k ,0 is the transmittance of the atmosphere between ground and sensor
Lp,k the atmospheric path radiance contribution (optical depth/ path length)

 k ,0 = exp((− 0 z ,k − a,k − r,k ) / cos  0 ) Eq. 11

The transmittance in the upward direction depends on the atmospheric attenuation,
and includes optical thickness of the ozone ( − 0 z , k ) aerosols ( − a,k) and the
molecular or Rayleigh optical thickness (− r,k ) for a given band at angle of obs.  0
 downward irradiance

For transmittance of incoming solar radiation:

• Sensor observation angle (cosϴ0 ) is replaced with angle of
incidence (cosϴi ) in Equation 11.
 True/ BOA Surface Reflectance
• Requires solving the following equation:
D ( L sen,k − La,k ) /  k ,0
k = Eq.12
E 0,k cos  i k ,i + E d ,k
Where:
D : correction factor accounting for variations of the Earth-Sun distance,

L sen,k : apparent spectral radiance at the sensor in band k,

La,k : atmospheric path radiance contribution (optical depth/ path length)
 k ,0 : upward transmittance of the atmosphere
E 0,k : solar irradiance at the ToA in band k
cos  i : zenith angle of incident flux

E d ,k : diffuse solar irradiance in band k

 k ,i : downward transmittance of the atmosphere

Options for atmospheric correction
• Empirical methods (simultaneous
measurements)
• Methods based on image data (dark-object,
split window).
• Methods based on radiative transfer models
(simulation).
 Chavez’s dark-object method (1996)

• Method assumes there is always high-absorption area in the image, which can be
reasonably accepted for shadowed areas or deep water bodies

• Also assumes that atmospheric scattering is constant throughout the image, such
that there is a minimum DN value is that commonly higher than 0 (attributed to
atmospheric scattering) – commonly affects shorter wavelengths ~ Rayleigh’s

• This minimum DN value decreases with increasing wavelength (in the SWIR)

• Simple approximation: subtract minimum DN value per band from all DN

values per band

• Shifts the origin of the histogram at zero

 Dark-object versus simplified
reflectance
0.6 0.6

0.5 0.5
Water Ref
0.4 0.4 Soil Ref
Reflectance

Veget Ref
0.3 0.3 Water Ch
Soil Ch
0.2 0.2 Veget Ch

0.1 0.1

0 0
ETM1 ETM2 ETM3 ETM4 ETM5 ETM7 Bands

Comparison of apparent reflectance (Ref) and atmospherically corrected curves

for water, bare soil and vegetation (Chavez method, Ch).