Click images for larger view. Use back button to return to this page. Higher resolution images show the south-west corner only. All are at native resolution. These show the "wow" factor of Ikonos compared to Landsat TM and SPOT XS and how close it gets to airborne multispectral (CASI). The CASI flight-line is about 500 m wide.

We think this is partly because the paper focused on Ikonos satellite imagery, which was relatively new (launched 1999) and little tested for marine environmental mapping applications. Being the first commercial high-resolution multispectral satellite sensor to achieve 4-m spatial resolution, finding out what the new imagery could deliver in terms of thematic accuracy was of great interest to a wide range of people in the research and coastal management communities. Our research showed that although the spectral limitations of the sensor and loss of radiance-contrast due to the atmosphere meant that if you tried to carry out a supervised multispectral classification, there was no significant improvement in the accuracy of classified habitats, and if you incorporated the textural information available from the high spatial-resolution data into the classification, you could significantly enhance thematic accuracy. This finding was useful in practical terms to potential users of Ikonos but also bore out some interesting theoretical predictions by Lubin and colleagues in Remote Sensing of Environment in 2001, which suggested that Rayleigh scattering in the atmosphere would cause a basic limitation on satellite sensors by causing loss of radiance contrast. We were able to directly compare the radiance contrast of satellite (Ikonos and Landsat Thematic Mapper) and airborne data along a transect and demonstrate that the airborne sensor had 2.0-2.7 times the contrast range of the satellite sensors.

The paper showed that despite Ikonos’s 4 m spatial resolution, which was much higher than comparable earlier civilian sensors such as those on Landsat and SPOT satellites, the thematic accuracy that could be obtained in mapping marine environments was only improved if textural information on habitats was incorporated into the classification. We had data on the mean boundary spacing between the shallow water marine habitats in the study area and knew that this was around 20 m. Thus, with 4 m pixels (around 25 to 50 pixels for each SPOT XS or Landsat TM pixel) we could obtain some useful measure of habitat texture. To do this we passed a 5x5 pixel variance filter (corresponding to mean habitat width) over each of three depth-invariant bands (bands 1/2, bands 1/3, and bands 2/3, which we had created from the three Ikonos visible bands), to make three textural layers. Using the spectral information contained in the depth-invariant bands combined with the textural information we showed a 20% improvement in thematic accuracy for fine-level habitat discrimination over Landsat and SPOT satellite sensors. Thus, essentially we combined several methodologies in a new way to give a method that was useful to others.

The key implication is that high spatial resolution satellite sensors also need enhanced spectral resolution (more, narrower wavebands) if they are to be used for detailed coastal habitat mapping and deliver reliable thematic maps. At present, you still need to use airborne sensors if you want high accuracy fine-level habitat maps. Just being able to remotely sense smaller patches with satellites like Ikonos doesn’t solve all the problems, although it does help. However, the extra habitat-variability information, which you can get from high-spatial resolution sensors like Ikonos, can significantly improve the accuracy of habitat maps.

We had previously—during a project which we carried out for the UK Department for International Development from 1994-1997—evaluated the capabilities and limitations of satellite imagery of varying spatial and spectral resolution for a range of coastal management applications. We had also compared the thematic accuracy that could be achieved using the satellite sensors with that which could be achieved using airborne digital sensors such as Compact Airborne Spectrographic Imager (CASI) and using aerial photographic interpretation. With the launch of Ikonos in 1999, which has a spatial resolution close to that of airborne sensors but a spectral resolution in the visible very similar to Landsat TM and its successor, the Enhanced Thematic Mapper (ETM+), we were naturally very eager to find out how the new sensor compared with other satellite sensors and with airborne imagery. We already had good ground-truthing data and a range of imagery of the Caicos Bank, Turks and Caicos Islands with which to compare Ikonos and so all we needed was an Ikonos image of our study area. This was generously provided by Frank Muller-Karger and Serge Andrefouet under the NASA data-buy program.

Dr. Peter Mumby
Royal Society University Research Fellow
School of Biological Sciences
Hatherly Laboratory
University of Exeter
Exeter, United Kingdom

Dr. Alasdair Edwards
Senior Lecturer
School of Biology
University of Newcastle 
Newcastle upon Tyne, United Kingdom