Model-based atmospheric transformation

Model-based atmospheric transformation consists of either the transformation of an at-sensor radiance data set to a surface reflectance data set, or the transformation of a surface reflectance data set to an at-sensor radiance data set. The at-sensor radiance to surface reflectance transformation provides a more rigorous alternative to the techniques described at Simple atmospheric correction for atmospheric correction.

By "model-based" we mean a transformation technique that involves a model of the interaction of light with the atmosphere (i.e., radiative transfer), and of the sensor. Inputs to the model are data on the terrain elevation, atmospheric conditions, solar illumination geometry, sensing wavelength intervals, and sensing geometry within the scene.

In the CATALYST Pro V10 atmospheric transformation system, the model generates an estimate of the at-sensor radiance for scene surfaces with spectrally uniform low and high reflectance (typically, with reflectance values of 0.05 and 0.60). The radiance components are 1) the at-sensor radiance reflected by the scene surface and 2) the at-sensor radiance due to all other sources. These estimates are idependently generated for each image pixel, and are used to evaluate parameters of a transformation equation for each pixel.

The CATALYST Pro V10 atmospheric transformation model implementation incorporates the MODTRAN4 radiative transfer model computer program. A prohibitive amount of computing would be required to run the MODTRAN4 program for each pixel in a realistically sized image. Therefore, the estimates are pre-computed and stored in a multidimensional radiance lookup table (RLUT). Each dimension of the RLUT corresponds to one of nine independent physical variables. For each pixel, the coordinates of the pixel along the RLUT dimensions are determined and then the radiance estimates are taken from the RLUT by linear interpolation between backeting samples.

Atmospheric water vapor content in the scene is one input into the atmospheric transformation process. A nominal value for the entire image extent may be used, but this is often inadequate since atmospheric water vapour content can vary rapidly over horizontal distance, especially if the scene is mountainous or contains water bodies. In situ measurements of atmospheric water vapour content throughout an imaged scene during the time of data acquisition would be very expensive to acquire. However, the Hyperspectral Analysis Package includes a program that generates an atmospheric water vapour content map from the image data set itself and an associated RLUT. This map can be used as an input into the atmospheric transformation process.

Many imaging spectrometers are pushbroom sensors that incorporate a rectangular detector array. One dimension of the detector array is the scanline. Incoming radiation is distributed on the basis of wavelength by the sensor optics so that different scanlines are exposed to radiation belonging to different wavelength intervals. Also, adjacent scanlines are exposed to radiation belonging to adjacent wavelength intervals. In principle, each detector belonging to the same scanline should be exposed to radiation belonging to the same wavelength interval. In reality, this may not be so. An optical distortion called "spectral line curvature" or "spectral smile" may cause a gradual shift of the center wavelength and width of the wavelength interval of radiation falling on the scanline, as a function of along-scanline position. So, pixels in the at-sensor radiance image will represent a slightly different spectral band depending on their positing in the along-scanline dimension.

Spectral line curvature may significantly distort the atmospheric correction results for a hyperspectral data set. In particular, it may cause incomplete removal of atmospheric absorption features and reflectance spikes adjacent to material absorption features. Therefore, even though the detection of spectral line curvature effects in a data set and their removal are not atmospheric correction operations, they may improve atmospheric correction results. Furthermore, the spectral line curvature detection and correction technique included in the Hyperspectral Analysis Package uses the same RLUT that is used for atmospheric transformation.

The programs that support model-based atmospheric transformation are listed in the following table:

We discuss the use of these programs for model-based atmospheric correction in the following sections:

1. Radiance lookup table generation
2. Atmospheric water vapour content map generation
3. Spectral line curvature detection (for pushbroom sensor data)
4. Model-based atmospheric transformation
5. Spectrum smoothing

References

Staenz, K., T. Szeredi, J. Schwarz, 1998., “ISDAS – A System for Processing/Analyzing Hyperspectral Data”, Canadian Journal of Remote Sensing, Vol. 24, No. 2, pp. 99 – 113.