FAQ: Frequently Asked Questions

  1. Question: Which radiance unit is used by ATCOR ?

    Answer: ATCOR employs the unit [mW cm-2 sr-1 micron-1]. For each channel the at-sensor radiance L(i) and digital number DN(i) are related by
    L(i) = c0(i) + c1(i)*DN(i)
    where c0(i) and c1(i) are the offset (bias) and slope (gain) of the linear calibration equation. The metadata files of different sensors use different units, so care has to be taken to convert these numbers into the ATCOR radiance unit.
    The c0, c1 coefficients (with unit mW cm-2 sr-1 micron-1) are specified in a "sensor.cal" file, e.g. "Ikonos.cal".

    Exception: for the SPOT cameras the calibration coefficient A(i) is specified in the metadata file as
    DN(i) = A(i) * L(i)
    and the A(i) coefficients (with unit [m2 sr micron W-1]) enter the ATCOR calibration file and c0(i)=0. ATCOR converts the A(i) internally as
    c1(i) = 0.1 / A(i)
    to obtain the required unit [mW cm-2 sr-1 micron-1]. Please use the values given in the "metadata.dim" file for SPOT-5 imagery, not the values of the "vol_dir.pdf". The values of "vol_dir.pdf" should be used for SPOT-4 !

  2. Question: The Landsat-7 ETM+ calibration coefficients are specified with the unit [W m-2 sr-1 micron-1 in the metadata file. What is the conversion factor for ATCOR ?

    Answer: Multiply the coefficients in the metadata file with 0.1 to convert into mW cm-2 sr-1 micron-1.

  3. Question: Landsat-7 ETM+ has commandable gain settings (low: g=1, high: g=1.5, for the solar bands, and high: g=2 for the thermal band). Do I have to consider this for the c0, c1 calibration coefficients ?

    Answer: The radiance-DN relationship for sensors with gain setting g is
    L = c0 + {c1/g} * DN
    However, in the Landsat-7 metadata file the slope (gain) value c1 is already adapted to g=1. Therefore, multiply the value with the factor 0.1 to convert the Landsat radiance unit (see Question 2) into the ATCOR unit [mW cm-2 sr-1 micron-1], see Question 1, and specify a gain setting g=1 in the ATCOR panel.

  4. Question: The tilt direction of the SPOT satellites is specified as "R"=right or "L"=left in the metadata. What does that mean in terms of geographic east/west ?

    Answer: The SPOT tilt angles are specified with respect to the direction of motion of the sub-satellite point, so "right" means geographic west for the descending orbit, where daytime data acquisition takes place, "left" means geographic east.

  5. Question: The reflectance of dark targets (water, vegetation) is too small or even negative in the blue-to-red spectral region. What action is necessary to obtain reasonable reflectance values ?

    Answer: There are several possibilities for this behavior:

    1. Check the solar zenith angle: is it correct ? A too small zenith angle, i.e. high sun, reduces the reflectance values.
    2. Check the visibility with the SPECTRA module: the default value may not be appropriate. You have to increase the visibility, i.e. decrease the path radiance, to raise the reflectance values.
    3. Try a different aerosol type (on the ATCOR main panel), e.g. switch from a maritime to a rural aerosol.
    4. If the actions (1)-(3) do not lead to reasonable reflectance values then change the radiometric calibration coefficients:
      (a) Edit a copy of the "sensor.cal" file, e.g. "sensor_edit2002.cal" and raise the offset c0 in the appropriate channels until the reference spectrum (e.g., water or vegetation) is matched.
      (b) use the "Inflight Calibration" module and select some ground target(s) whose reflectance spectra serve as reference and will be matched. A new calibration file will be generated.

  6. Question: Which aerosol model should be used for what location and weather condition ?

    Answer: The composition of the aerosol particles is primarily controlled by natural and man-made sources at the earth's surface. The aerosol content of the atmosphere at a given location will therefore depend on the trajectory of the local air mass during the preceding several days. However, some general easy-to-use guidelines for the selection are :

    1. rural aerosol
      If in doubt, select the "rural" aerosol type. It represents conditions one finds in continental areas not directly influenced by urban/industrial sources. This aerosol consists of dust-like and organic particles.
    2. urban aerosol
      In urban areas, the rural aerosol background is often modified by the addition of particles from combustion products and industrial sources (carbonaceous soot-like particles). However, depending on wind direction and shortly after raining, the rural aerosol might also be applicable in urban areas.
    3. maritime aerosol
      In areas close to the sea or close to large lakes, the aerosol largely consists of sea-salt particles, mixed with continental particles. In these areas, the aerosol choice would depend on wind direction: for an off-shore wind, the rural aerosol (or urban) type is still the best choice, otherwise the maritime. Again, since the wind conditions are often not known, select the rural type if in doubt.
    4. desert aerosol
      As the name implies, this type is intended for desert-like conditions, with dust-like particles of larger size.

      In forested, agricultural areas and scrub land the rural aerosol is usually the adequate choice. This also holds for polar, arctic, and snow covered land.
      Notice :
      (i) The SPECTRA module can be used to assess the influence of the selected aerosol type. The target spectrum for a certain target (e.g. vegetation, water) can be displayed for the different aerosol types and compared to library spectra. The aerosol with the closest match would be the logical choice.
      (ii) If the sensor has a blue spectral band and a 1.6 or 2.2 um band (e.g. Landsat TM), ATCOR starts with the user-selected aerosol/atmosphere, but is able to adapt the path radiance, i.e. aerosol properties, in cases where reference areas (coniferous forest, dark soils) are found in the scene. Thus, ATCOR will cope with non-standard aerosol conditions, e.g. a mixture of rural and urban aerosols.

  7. Question: What is the required DEM spatial resolution for the combined atmospheric/ topographic correction ?

    Answer: A DEM spatial resolution comparable to the image resolution is sufficient for most applications. Even if the DEM pixel size is larger than the image pixel a clear reduction of topographic illumination effects is usually achieved. However, areas of steep slopes that are smoothed in the coarse-resolution DEM may lead to artifacts.
    In practice we often have to live with DEM resolutions of 50 - 100m, because higher resolution DEM's are not available or too expensive.

  8. Question: When calculating the slope and aspect images from very high resolution DEM's (typically 1-10 m) steep slope gradients occur from one pixel to the neighboring pixel. This causes the slope pattern to be visible in the processed image. How can I avoid this ?

    Answer: Use a filter size of 3-10 pixels for the slope/ aspect calculation to smooth the gradients.

  9. Question (hyperspectral sensors): Reflectance spectra have spikes in the spectral regions with atmospheric water vapor. How can I remove these spikes ?

    Answer: There are 3 possibilities:
    - reduce the atmospheric water vapor column
    - do a re-calibration ("Inflight Calibration" with reference targets)
    - try the "Spectral Polishing" module (only for small-amplitude spikes)

  10. Question: When processing imagery in rugged terrain I get some bright areas with very high reflectance values that are obviously overcorrected. How can I avoid the overcorrection ?

    Answer: The overcorrection usually occurs in areas with steep slopes oriented away from the sun. These areas with a high local solar incidence angle (or low local solar elevation angle) appear dark in the illumination image cos(solar_incidence) that is automatically generated. In these areas the reflectance shows a behavior that depends on the illumination and viewing conditions (the bidirectional reflectance distribution function, BRDF). There is no general solution to this problem. However, the next Question/Answer discusses an empirical correction method implemented in ATCOR.

  11. Question: For the combined atmospheric / topographic processing ATCOR offers an empirical BRDF correction that depends on the illumination and viewing geometry. In its simplest form it reads :
    rho(corrected) = rho(isotropic) * cos(beta) / cos(beta_t)
    where beta is the local solar incidence angle on a surface element, and beta_t is a threshold angle to be specified by the user. How do I have to specify the beta_t to reduce the reflectance of overcorrected bright areas in the processed scene ?

    Answer: Two output files are generated during the atmospheric / topographic correction: the processed scene and the illumination image cos(beta) which is coded as byte data, i.e. illumination illu = 100*cos(beta).
    Link both files (e.g. with the ENVI software) and look for the dark areas in the illumination image. These correspond to steep slopes oriented away from the sun and these areas usually show strong BRDF effects, i.e. they correspond to the bright overcorrected areas in the processed scene.
    Example: the scaled overcorrected reflectance is 320 (scale factor 4), i.e., rho(isotropic) = 80%. The illumination file for this pixel says illu=34, i.e., beta = arccos(illu/100)=70 degree. Let us assume the 80% reflectance should be reduced to 40%, which is the value for pixels in the flat-terrain neighborhood. Then the threshold angle beta_t has to be specified such that cos(beta) / cos(beta_t) = 0.5, in this case beta_t=47 degree. So if the desired geometry-dependent reflectance reduction factor is G the required threshold angle can be calculated as
    beta_t = arccos { cos(beta) / G } = arccos { illu / (100*G) }
    Unfortunately, there is no general solution to this problem, the threshold parameter is scene dependent and has to be defined interactively. A rule of thumb is
    beta_t = solar_zenith_angle + 20 [degrees]

  12. Question: ATCOR offers the calculation of VALUE ADDING channels such as LAI and FPAR. These are based on equations with 3 empirical parameters that depend on vegetation cover type. Should I use the default parameter values, e.g. for corn, even if my scene contains other vegetation types as well ?

    Answer: Since the current LAI and FPAR equations are simple empirical relationships based on the selected vegetation index (either NDVI or SAVI) they can only approximate typical trends. This application is mainly intended to get the correct trends in multi-temporal studies, therefore it is recommended to use the same parameter set for all multitemporal scenes, whether it is the default or user-specified parameters.
    A quantitative agreement with field measurements of different crop types in different seasons cannot be expected.


Last Updated: 23-Jan-2008, RR, DLR, r.richter@dlr.de