Land Cover/Use Mapping Using Object Based Classification of SPOT Imagery

National authorities and international organisations worldwide are interested in the division of the landscape according to the various classes of land cover/use, for example, into urban areas, arable land, grasslands, forests or wetlands. The creation of a land cover/use map emerges not only from the need to generate information that would be useful for general policy purposes, but also from the need to control development at a local level; for example, the need to conserve natural resources, to deal with problems incurring as a result of tourism development and local authority planning, among others.

Today, a wide range of satellite sensors such as Landsat TM, IRS and SPOT HRVIR, are extensively used for land cover/use mapping on different scales, by employing a large number of image interpretation techniques. A classification technique that has recently been developed is object-based image analysis (OBIA). The concept of OBIA is that information necessary to interpret an image is not represented in single pixels, but in meaningful image objects. The technique is an approach that uses not only spectral information, but also spatial information of image objects.

The aim of this work was to examine the potential of OBIA in the mapping of basic land cover/use classes when employing SPOT HRVIR imagery. The specific objectives were:
• to develop an object based classification model for mapping the land cover/use classes using a SPOT image,
• to examine the transferability of the model by applying the same model on a second SPOT image of a different area, and
• to estimate the accuracy of the model by comparing the results with data collected in the field, as well with data derived from photo-interpretation of very high resolution imagery.

The combination of SPOT data and OBIA revealed promising results (79.11% overall accuracy) in the classification of the first image. Also, the transferability of the model proved to be successful (81.5% overall accuracy) when applied to the second image. However, it should be noted that adjustment of the thresholds of the feature values that were incorporated in the model, was necessary in order to yield the results.