47
wavelengths that the sensor is measuring. Once pure pixels have been identified, a
linear regression of the image pixel values to the ground reflectance measurements
for each wavelength is undertaken and the resulting relationship is used to convert
the image to surface reflectance. Where multiple targets at each reflectance level are
available, a validation of the relationship can be carried out.
Dark Object Subtraction (DOS)
A dark object subtraction (DOS; Chavez 1988) is built on a simple assumption that
the darkest pixels within the scene have little or no surface reflectance and that the
radiance measured by the sensor is from the atmosphere. Therefore, while assuming
the atmosphere is consistent across the scene, subtracting that atmosphere component from the whole scene can be used to convert the at-sensor reflectance values to
surface reflectance. This is performed independently for each of the image bands
(i.e., wavelengths). However, there is a risk that the relative relationships between
the image bands can vary.
Modeled Atmospheres
Modeling the atmosphere is the most common way in which imagery is atmospherically corrected but this requires a radiative transfer (RT) model and associated
parameters, many of which are supplied in the image header file from the data provider (e.g., date and time of the acquisition). However, typically you, as the enduser, would perform this analysis through a software package that aids the
parameterization (e.g., automatically parses the supplied header file or associated
metadata), runs the atmospheric model and applies the model outputs to the image
file. There are a number of software packages and models that support this analysis
(Table 2), but they each only support a defined number of sensors. These lists are
being updated on a regular basis. Additionally, some products and analysis steps
may not be possible for all sensors and therefore functionality may not be equal
across all sensors (e.g., cirrus cloud correction uses bands only provided by
Sentinel-2 and Landsat-8 instruments).
More recently, there has been some effort to standardize these processing stages
and levels (Claverie et al. 2015; Feng et al. 2013; Ju et al. 2012; Roy et al. 2010) for
the Landsat and Sentinel-2 imagery. The United States Geological Survey (USGS)
is already supplying the Landsat archive (TM, ETM+, OSL) as an atmospherically
corrected product (Masek et al. 2006) and, in time, there may be a similar service
for Sentinel-2 imagery.
For the Second Simulation of the Satellite Signal in the Solar Spectrum (6S; Vermote
et al. 1997) model (others models are similar), the parameters needed are given in
Table 3. The sensor configuration and position parameters are well defined and
known so these can be parameterized using the image header information. However,
the parameters associated with the atmosphere at the time of the acquisition, specifically
Pre-processing of Remotely Sensed Imagery
wavelengths that the sensor is measuring. Once pure pixels have been identified, a
linear regression of the image pixel values to the ground reflectance measurements
for each wavelength is undertaken and the resulting relationship is used to convert
the image to surface reflectance. Where multiple targets at each reflectance level are
available, a validation of the relationship can be carried out.
Dark Object Subtraction (DOS)
A dark object subtraction (DOS; Chavez 1988) is built on a simple assumption that
the darkest pixels within the scene have little or no surface reflectance and that the
radiance measured by the sensor is from the atmosphere. Therefore, while assuming
the atmosphere is consistent across the scene, subtracting that atmosphere component from the whole scene can be used to convert the at-sensor reflectance values to
surface reflectance. This is performed independently for each of the image bands
(i.e., wavelengths). However, there is a risk that the relative relationships between
the image bands can vary.
Modeled Atmospheres
Modeling the atmosphere is the most common way in which imagery is atmospherically corrected but this requires a radiative transfer (RT) model and associated
parameters, many of which are supplied in the image header file from the data provider (e.g., date and time of the acquisition). However, typically you, as the enduser, would perform this analysis through a software package that aids the
parameterization (e.g., automatically parses the supplied header file or associated
metadata), runs the atmospheric model and applies the model outputs to the image
file. There are a number of software packages and models that support this analysis
(Table 2), but they each only support a defined number of sensors. These lists are
being updated on a regular basis. Additionally, some products and analysis steps
may not be possible for all sensors and therefore functionality may not be equal
across all sensors (e.g., cirrus cloud correction uses bands only provided by
Sentinel-2 and Landsat-8 instruments).
More recently, there has been some effort to standardize these processing stages
and levels (Claverie et al. 2015; Feng et al. 2013; Ju et al. 2012; Roy et al. 2010) for
the Landsat and Sentinel-2 imagery. The United States Geological Survey (USGS)
is already supplying the Landsat archive (TM, ETM+, OSL) as an atmospherically
corrected product (Masek et al. 2006) and, in time, there may be a similar service
for Sentinel-2 imagery.
For the Second Simulation of the Satellite Signal in the Solar Spectrum (6S; Vermote
et al. 1997) model (others models are similar), the parameters needed are given in
Table 3. The sensor configuration and position parameters are well defined and
known so these can be parameterized using the image header information. However,
the parameters associated with the atmosphere at the time of the acquisition, specifically
Pre-processing of Remotely Sensed Imagery