417
• Spectral correction: Every pixel contains spectral information recorded at
slightly different wavelengths, which are standardized to a common wavelength
interval in this step.
• Atmospheric correction: Atmospheric correction transforms TOA reflectance to
surface reflectance using information about atmospheric conditions and aerosol
properties at the time of image acquisition. This can be done with atmospheric
radiative transfer models (e.g., MODTRAN, Berk et al. 1987), some of which are
included in image processing software; dark subtraction, the subtraction of values from dark image pixels; or ELC (explained below).
• Geometric correction (including geometric resampling, orthorectification, and
georeferencing): Geometric resampling makes pixels square (initially they are
elliptic). Orthorectification corrects image distortions caused by sensor tilt, flight
altitude, and changes in surface terrain, creating planimetric images, which allow
direct measurement of distances, areas, and angles. Geometric resampling and
orthorectification require information about camera properties, the sensor position recorded by the inertial measurement unit (IMU), and an accurate digital
elevation model (DEM), which provides information about terrain height (elevation above sea level). Georeferencing aligns images to a specific coordinate system. This is frequently done using ground control points (GCPs).
Ground reference data collected for image processing frequently include GCPs
for georeferencing and reference spectra for image calibration/validation procedures and atmospheric correction with ELC. Generally, GCPs need to be easily
identifiable in the acquired images; road intersections, corners of buildings, and
trees are good choices. Accurate and precise coordinates of GCPs can be either
determined from existing georeferenced imagery, in which case it is good practice
to check if the features still exist, or measured on the ground. For ELC, the reflectance of large calibration targets on the ground is measured with a handheld spectrometer during the overflight. At-sensor radiance is transformed to ground- level
reflectance by applying bandwise transformation coefficients estimated based on
the difference between sensor and calibration target spectra (Smith and Milton
1999). Essentially, this subtracts atmospheric influences from the spectra recorded
by the sensor. Calibration targets should meet several requirements. They need to be
identifiable in the images; when in doubt, location data should be collected.
Furthermore, calibration targets should allow the extraction of several pixels from
the image, so they should be sufficiently large (e.g., targets measuring 7 × 7 pixels
allow using 5 × 5 pixels from the image after removing the pixels at the edges). In
addition, calibration targets should be Lambertian as possible (i.e., they should
reflect light equally in all directions, independent of viewing angle). Ideally, calibration targets should include the range of values in the image, which means including
targets with low and high reflectance. Good options are calibration tarps in different
shades (e.g., white, gray, and black) that are as spectrally invariant as possible. Tarps
can be made of boat canvas (e.g., acrylic-coated woven polyester) and should
include grommets, so that they can be tightly pulled and secured with pegs.
Calibration tarps should be placed in flat areas with short vegetation. Ideally, one set
of tarps is placed in the center of each flight line and measured exactly at the time
15 Spectral Field Campaigns: Planning and Data Collection
• Spectral correction: Every pixel contains spectral information recorded at
slightly different wavelengths, which are standardized to a common wavelength
interval in this step.
• Atmospheric correction: Atmospheric correction transforms TOA reflectance to
surface reflectance using information about atmospheric conditions and aerosol
properties at the time of image acquisition. This can be done with atmospheric
radiative transfer models (e.g., MODTRAN, Berk et al. 1987), some of which are
included in image processing software; dark subtraction, the subtraction of values from dark image pixels; or ELC (explained below).
• Geometric correction (including geometric resampling, orthorectification, and
georeferencing): Geometric resampling makes pixels square (initially they are
elliptic). Orthorectification corrects image distortions caused by sensor tilt, flight
altitude, and changes in surface terrain, creating planimetric images, which allow
direct measurement of distances, areas, and angles. Geometric resampling and
orthorectification require information about camera properties, the sensor position recorded by the inertial measurement unit (IMU), and an accurate digital
elevation model (DEM), which provides information about terrain height (elevation above sea level). Georeferencing aligns images to a specific coordinate system. This is frequently done using ground control points (GCPs).
Ground reference data collected for image processing frequently include GCPs
for georeferencing and reference spectra for image calibration/validation procedures and atmospheric correction with ELC. Generally, GCPs need to be easily
identifiable in the acquired images; road intersections, corners of buildings, and
trees are good choices. Accurate and precise coordinates of GCPs can be either
determined from existing georeferenced imagery, in which case it is good practice
to check if the features still exist, or measured on the ground. For ELC, the reflectance of large calibration targets on the ground is measured with a handheld spectrometer during the overflight. At-sensor radiance is transformed to ground- level
reflectance by applying bandwise transformation coefficients estimated based on
the difference between sensor and calibration target spectra (Smith and Milton
1999). Essentially, this subtracts atmospheric influences from the spectra recorded
by the sensor. Calibration targets should meet several requirements. They need to be
identifiable in the images; when in doubt, location data should be collected.
Furthermore, calibration targets should allow the extraction of several pixels from
the image, so they should be sufficiently large (e.g., targets measuring 7 × 7 pixels
allow using 5 × 5 pixels from the image after removing the pixels at the edges). In
addition, calibration targets should be Lambertian as possible (i.e., they should
reflect light equally in all directions, independent of viewing angle). Ideally, calibration targets should include the range of values in the image, which means including
targets with low and high reflectance. Good options are calibration tarps in different
shades (e.g., white, gray, and black) that are as spectrally invariant as possible. Tarps
can be made of boat canvas (e.g., acrylic-coated woven polyester) and should
include grommets, so that they can be tightly pulled and secured with pegs.
Calibration tarps should be placed in flat areas with short vegetation. Ideally, one set
of tarps is placed in the center of each flight line and measured exactly at the time
15 Spectral Field Campaigns: Planning and Data Collection