Choosing ATCOR parameters
The bands sensed by Ikonos (and many other satellite sensors) were deliberately chosen to avoid interference by the atmosphere as much as possible. Whilst this is sensible, it does mean that the Ikonos data carry little or no information on the water vapour content of the atmosphere or on the amount and type of aerosols present. Consequently, if we wish to calibrate Ikonos data to reflectance we must obtain information on the atmosphere at the time of sensing from other sources or assume 'standard' values an apply these throughout the image. In most cases, contemporaneous data from other sources are not available, or are not cost-effective, so we often must use estimated values based on the time of year, the latitude of the site and its location relative to known sources of moisture and aersols. The result is that ATCOR, like many other models, has options for the different seasons of the year, for mid-latitudes and the tropics, and for situations dominated by maritime aersols (salt), urban aerosols (carbonaceous particles) and desert aerosols (dust).
Wedholme Flow is in the north-western part of England, so it is reasonable to select a mid-latitude model, and the the data were acquired in late October, so either an autumn or winter model is appropriate. The area is close to the sea, but the image was acquired in late morning in autumn, so it unlikely that a sea breeze was occurring which would bring maritime aerosols over the site. The area is a long way from sources of urban pollution, so it is reasonable to select a rural atmosphere.
Putting all of this together, we will start the investigation using a mid-latitude, autumn (fall) rural atmosphere. The appropriate atmospheric file is called 'faru2e' by ATCOR2. To keep things simple, let's switch off the adjacency correction (i.e. set it to 0.0), and let's start by assuming that the atmosphere was very hazy, say a visibility of 20km. Running ATCOR2 with these parameters produces the following spectrum from the area of asphalt at the end of Kirkbride airfield runway:
Figure 1. Asphalt spectrum. Rural atmosphere in autumn (fall), 20km visibility, no adjacency correction.
We don't have an actual spectrum from this area of asphalt, so we can't be sure whether the dip in red wavelengths is plausible, but it is clear that the model has over-corrected the blue wavelengths, which are giving negative reflectance values. The most likely reason for this is that we have under-estimated the clarity of the atmosphere at the time the data were acquired, so let's try again, this time setting the visibility to 30km.
Figure 2. Asphalt spectrum. Rural atmosphere in autumn (fall), 30km visibility, no adjacency correction.
This is more plausible, but the dip in red wavelengths remains. Also, the reflectance in blue wavelengths, although positive, is still too low (1.1% at 486nm). Let's increase the visibility to 40km.
Figure 3. Asphalt spectrum. Rural atmosphere in autumn (fall), 40km visibility, no adjacency correction.
Increasing the visibility still further leads to progressively higher values of reflectance, especially in blue wavelengths, but has no effect on the dip in red wavelengths (Figure 4).
Figure 4. Asphalt spectrum. Rural atmosphere in autumn (fall), no adjacency correction. The lowest line is for a visibility of 40km, then successively higher lines are for 50km, 60km and 70km visibility.
Let's assume 40km visibility for the time being, and turn our attention to the effect of the other parameters. The next one to investigate is the atmospheric model. So far we have assumed a mid-latitude autumn atmosphere, but lets compare that with some other possible atmospheres (Table 1).
|
Blue
|
Green
|
Red
|
Near IR
|
|
| Autumn, rural |
3.2 %
|
4.8 %
|
4.1 %
|
14.2 %
|
| Mid-latitude, winter, rural |
4.3 %
|
5.8 %
|
4.8 %
|
14.1 %
|
| Mid-latitude, winter, urban |
6.6 %
|
7.8 %
|
6.2 %
|
15.6 %
|
| US standard, maritime aerosols |
3.7 %
|
5.2 %
|
4.3 %
|
14.3 %
|
Table 1. The reflectance of the area of asphalt, assuming different type atmospheres.
The effect of changing the atmospheric model is to cause a small change in the reflectance of the asphalt surface, but without ground data it is impossible to say which is the most accurate. Experience from similar asphalt surfaces elsewhere suggests that the mid-latitude, winter, urban is probably closer to the true value.
Figure 5. The effect of changing the atmosphere type on the spectrum of an asphalt surface.
Expanding the y-axis scale to investigate more closely the nature of the reflectance from the asphalt surface reveals a dampened green vegetation spectrum (Figure 5). The presence of a red-edge in the asphalt spectrum for all the atmospheric types suggests that the signal from the asphalt runway may be contaminated by light scattered from the adjacent areas of grass. In order to account for this effect we can introduce an adjacency correction.
Figure 6. Asphalt spectrum. Mid-latitude urban atmosphere in winter, 40km visibility. The upper line is no adjacency correction, the lower dotted line is with 0.1km adjacency correction.
The adjacency correction causes a reduction in the near infra-red reflectance as would be expected as the asphalt is surrounded by dense short grass with high near infra-red reflectance. Increasing the adjacency correction above 0.1km had little further effect upon the reflectance.
It is possible to invest a great deal more time and effort into performing an atmospheric correction of this image, but the absence of independent ground data mean that this is unlikely to be justified. The correction described above was easily achieved using the ATCOR2 program and is sufficient to provide a basis for using Ikonos data to map and monitor the status of lowland raised bogs in the UK.
This work is licenced under a Creative Commons Licence.
©
NCAVEO, 2005
Network for Calibration and Validation of Earth Observation data
School of Geography, University of Southampton
Southampton SO17 1BJ, UK
