Example 3. Atmospheric correction of an Ikonos multispectral image.
Introduction
The ability to survey large areas is one of the primary advantages of remote sensing, but for many applications the size of the area covered is less important than having data with a high spatial resolution from a number of well-defined targets. This is often the situation faced by nature conservation organisations responsible for managing a number of scattered small sites, and in this example we describe research by English Nature to incorporate remote sensing into the procedures for managing and monitoring the lowland raised bogs in the UK.
Lowland raised bogs are an important and declining habitat throughout Western Europe. The majority of lowland raised bogs in the UK have been damaged to varying degrees and by various human activities over a very long period. The classification and categorisation of the extent of this damage, the extent of natural or near natural active peat growth and the ability to restore active peat formation is central to the management of remaining sites and the application of appropriate restoration measures on degraded sites.
This example is based on a study commissioned by Dr Roger Meade (English Nature) and was undertaken by a multidisciplinary team from the University of Southampton GeoData Institute.
Study site
A total of eight lowland raised bogs were studied (Milton et al. 2004), but only results from one of these will be presented here : Wedholme Flow in Cumbria. This site contains bog in various states, from almost intact 'primary' bog, through bog in a range of degraded states, to large areas from which the surface peat has been removed by mechanical excavation. The challenge facing remote sensing is more than simply classifying an Ikonos image, it is to deliver an information system that is robust and repeatable, so that changes in the area and status of the various communities present on the bog surface can be monitored into the future. One of the first steps necessary to achieve this is atmospheric correction of the Ikonos data.
Figure 1. Ikonos image of Wedholme Flow, Cumbria. © Space Imaging, 2001.
The Ikonos image supplied by Space Imaging was centred on Wedholme Flow, but also had areas of cloud and part of the estuary of the Solway Firth at the north of the image. Both of these could have introduced problems into the atmospheric correction, so the image was subset to extract just Wedholme Flow and its immediate surroundings (Figure 2).
Figure 2. The image subset of Wedholme Flow, Ikonos bands 3,2,1 [R,G,B]. © Space Imaging, 2001.
The Ikonos image was acquired at 11:17 GMT on 24th October 2001. The solar elevation angle was quite low at that time of the year (23°), but previous experience had shown that the vegetation communities of interest would be spectrally distinct in the autumn. A further reason for choosing an autumn image was to avoid the hazy skies of mid-summer, and thus make the atmospheric correction easier.
The Antunes-6S model used for Example 1 was not applicable in this case because Ikonos data are distributed over an 11-bit dynamic range and Antunes-6S is limited to 8-bit image data. Conversion to 8-bit data was undesirable due to the need to preserve the maximum signal-to-noise ratio. Consequently, the ATCOR2 atmospheric correction program distributed by ReSe Applications Schläpfer was chosen for use in this example. This is one of a suite of ATCOR programs, ATCOR2 being suitable for satellite sensor data acquired from fairly flat terrain, ATCOR3 for satellite sensor data from rugged terrain, and ATCOR4 for airborne sensor data. All of them are based on MODTRAN, but rather than run the complete MODTRAN model each time a correction is required, they use a pre-determined 'look-up table' containing the results from MODTRAN for a range of possible conditions. The look-up table provided with ATCOR2/3 contains data applicable to a number of different satellite sensors, whereas the user has to define some sensor parameters in order to use ATCOR4 as the range of possible airborne sensors is very large.
The metdata file provided by Space Imaging contained many of the parameters required to run ATCOR2:
Nominal Collection Azimuth : 70.8299 degrees
Nominal Collection Elevation : 68.23434 degrees
Sun Angle Azimuth : 169.2586 degrees
Sun Angle Elevation : 22.87980 degrees
Acquisition Date/Time : 2001-10-24 11:17 GMT
...
Pixel Size X: 4.00 meters
Pixel Size Y: 4.00 meters
In addition, the approximate ground elevation, and latitude and longitude of the centre of the image subset were obtained from a map.
ATCOR2 version 6.1 was run within the freely available IDL Virtual Machine 6.1. Three Ikonos sensor calibration files were provided with ATCOR2, that for 2001 onwards was chosen. The main data entry screen for ATCOR2 is shown in Figure 3 and the meaning of some of the key parameters is explained below.
Figure 3. The main data entry screen for ATCOR2.
Atmospheric file
ATCOR2 has several pre-determined atmospheres, and the one selected as most
appropriate for the Wedholme image was 'fall (autumn), rural area'. There are
several versions of each atmosphere, depending upon the sensor view geometry.
The Ikonos metadata file gave the Nominal Collection Azimuth as approximately
70°, at an elevation angle of approximately 68°, which equates to the
sensor viewing the site from an orbital path to the east, at a view zenith angle
of approximately 20°. Consequently, the '20 deg. east, fall, rural' atmosphere
file was initially selected (faru2e). This was later changed to use the '20
deg. east, winter, mid-latitude, urban atmosphere (mwur2e) for reasons that
are described here.
Adjacency range
The ATCOR programs include a correction for the adjacency effect, which is important
if juxtaposed surfaces with very different reflectance properties are present
in the image. The effect is to sharpen-up the detail in the image, as the spatial
averaging introduced by the atmosphere is moderated to some extent. Choosing
a suitable value for the adjacency range is largely trial-and-error in the absence
of information on the point spread function of the sensor. A value of 0.10 km
was chosen for use with this Ikonos image.
Visibility
The most important parameter is the horizontal visual range, referred to as
the visibility parameter. In the absence of contemporaneous ground data, this
is an educated guess. However, the ATCOR programs provide an interactive way
to choose the visibility parameter which can be very helpful. A module within
ATCOR called SPECTRA displays the image and two windows in which the analyst
can display corrected spectra from known surfaces, for checking before running
the model on the whole image (Figure 4).
Figure 4. Using the SPECTRA module within ATCOR2 to check the plausibility of two spectra, the top from grass, the bottom from asphalt. (click the image to enlarge)
The final image, corrected using ATCOR2 is shown in Figure 5. Visually, it appears almost identical to the original subset shown in Figure 2, but on closer study it is clear that the ATCOR2 corrected image is slightly sharper and the vegetation boundaries have better definition (see Figure 6). However, the main difference between the two data sets is that the atmospherically corrected image contains data values which are meaningful physical units and capable of being extrapolated across space and through time. There is still a great deal to do before the reflectance values contained in remotely sensed images are truly independent of the conditions of measurement, but atmospheric correction is an important first step.
Figure 5. The atmospherically corrected Ikonos subset of Wedholme Flow.
Figure 6. Comparison between a small area from the atmospherically-corrected image (left) and the same area from the original image (right). The corrected image is slightly sharper and has marginally better visual discrimination between vegetation classes, although some of this difference is lost in the compression necessary for presentation on the website.
References
Milton, E. J., Hughes, P. D., Anderson, K., Schultz, J., C.T.Hill and Lindsay, R., 2004. Remote sensing condition categories on lowland raised bogs in the United Kingdom. Development and testing of methods. Proceedings of the Peterborough Remote Sensing Workshop, Peterborough, English Nature, 26-35.
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
