The pattern size depends on the current resolution; you may adjust the multiplier to your needs (we used multiplier 5 above). Figure 9.7 shows a portion of an overlay with the roads raster map, resampled from the vector map to 10 m resolution. This approach applies when referencing with topographic raster maps. If the reference map is a vector map, it may be simply overlayed over the rectified image(s) with d.vect to verify the geocoding accuracy. If the accuracy is not sufficient, you have to go back to the xy LOCATION and select more or different ground control points. You should make sure that sufficient GCPs are selected and the GCPs are well distributed in the satellite scene.

9.4.2 Radiometric preprocessing

Besides changes to the color lookup tables (LUTs) that are used to enhance the visual perception of an image, satellite data often have to be radiometrically preprocessed so that each pixel represents the apparent radiance measured at the satellite sensor. Up to three major effects have to be corrected, depending on the project goal, the image type, and the observed targets:

the pixel values are usually a linear transformation of the original data performed by the data provider to fit into the range of 8 bit (0 - 255). By applying gain and bias (also called offset ) values which are delivered in the image header files or available from the data provider, the DN values (DN: digital numbers) can be recalculated to apparent radiance at sensor values;

optical data, depending on their spectral range, are influenced by atmospheric effects. To reconstruct the reflectance values at earths surface (image includes only the values measured at the satellite), each satellite channel has to be atmospherically corrected;

when the observed target area contains hilly or mountainous regions, the slopes cause variations in the brightness reflectance (terrain effects), which can lead to wrong reclassification results. To overcome this problem, a terrain correction (illumination correction) based on the local slopes derived from an elevation model has to be applied.

Image calibration from DN to apparent radiances at sensor. The gain/bias correction is applied channel-wise to the image data set. The values for the gains and biases are available from the channel headers (leader file) or the data providers.

For LANDSAT-TM7, there are two gain states (low and high gain, see GSFC/NASA, 2001). The rationale behind switching gain states is to maximize the instruments 8 bit radiometric resolution without saturating the detectors. For thermal data, both low and high gain data are available by default.

For other bands (1 to 5, and 7) the satellite will acquire image data in one of two possible gain settings. High gain measures a lesser radiance range with increased sensitivity over areas of low reflectance. Low gain setting measures a greater radiance with decreased sensor sensitivity for very bright regions to avoid detector saturation.

The leader file of a satellite data set can be analyzed with gdalinfo (delivered with GDAL library) for GDAL supported data formats. The program prints important metadata information, including the gain and bias values, if they are present in the data set. Because the original SPOT-1 data are not available, we show an example for a LANDSAT-TM7 data set in CEOS format (first channel):

gdalinfo /cdrom/scenel/dat 01.001 Driver: SAR CEOS/CEOS SAR Image Size is 6920, 5960 Coordinate System is Metadata: [. . .]

CEOS OFFSET A0= - 6.200000OOOOe + 00

CE0S GAIN A1= 7.7568627451e-01

CEOS GAIN SETTING=H Corner Coordinates: Upper Left ( 0.0, 0.0) Lower Left ( 0.0, 5960.0) Upper Right ( 6920.0, 0.0) Lower Right ( 6920.0, 5960.0) Center ( 3460.0, 2980.0)

Band 1 Block=6920xl Type=Byte

Min=0.000/0, Max=255.000/0

ColorInterp=Undefined

From this output, we obtain gain and bias (offset) as well as the gain level

(high or low gain). The units for gain/bias are usually

nfisrfun

The gen-

eral equation for calculation of the apparent pixel radiance at sensor is (Schowengerdt, 1997:313):

with:

Lf. apparent pixel radiance of channel j [ biasyi offset of linear equation for channel j gam/, gain of linear equation for channel j

To apply a gain/bias correction, the module r.mapcalc can be used. For our example, it will be the following command (we assume, that we have imported the first channel as tm.1):

r.mapcalc tm.lrad=0.77568627451 * tm.l - 6.2

Note that the values depend on the data provider and the image acquisition date as gain/bias values regularly change for various reasons.

With further calculations, it is also possible to convert apparent pixel radiance at sensor to planetary reflectance or albedo (see Mather, 1999:93 and Schowengerdt, 1997:317). These planetary reflectances can be computed to achieve a reduction in between-scene variability through a normalization for solar irradiance. Please refer to the remote sensing literature for details.

Correction of atmospheric effects. Satellite signal distortions are caused by several effects. Diffuse irradiance from sky may increase the radiance of an observed object. Path radiance (atmospheric intrinsic radiance) leads to haze effects. Local effects such as environmental radiance from neighborhood objects change the objects radiance, as well as a locally reduced upward trans-mittance. Finally, there is the adjacency effect, when a brighter adjacent object influences the surrounding objects radiation. All these problems are widely discussed in the remote sensing literature, see for example Schowengerdt, 1997 (Chapter 2). Atmospheric effects are visible in color composites as a whitish-bluish haze.

The correction of such atmospheric effects is a complex issue. Using an atmosphere model like 6S (Second Simulation of the Satellite Signal in the Solar Spectrum, Vermote et al., 1997), the radiance at earths surface can be reconstructed from the apparent radiance at sensor if the local weather conditions at image acquisition time are known. For a method to use the 6S model within GRASS see Neteler, 1999.

As detailed information about the local weather conditions and gaseous contents are often unknown, the atmospheric effects can be retrieved statistically from the image channels themselves. Known dark objects (e.g. water bodies or coniferous forest) can be used to do this. In an unconnected image, these objects do not appear dark due to atmospheric effects. The amount of path radiance is approximately identified by calculating pixel-wise the difference between the actual radiance for a dark object and zero (full absorption, given for water in infrared). This difference value can be removed from all pixels of the channel. Details are described in Moran et al., 1992 and Chavez Jr., 1996. The modules d.what.rast or r.what can be used to calculate the path radiance for dark objects, the subtraction can be done with r.mapcalc. A simpler method, not considered in detail here, is based on the Tasseled Cap transformation. It does not require the manual identification of dark objects and corrects the data set through a haze image (Tasseled Cap component TC4) and linear regression. It is implemented in i.tm.dehaze for LANDSAT-TM5.

Correction of terrain effects. When observing hilly or mountainous areas, a terrain correction should be applied to the image to correct local brightness