Borstad Associates Ltd. - CSA Outreach Task 5

CSA Outreach Task 5

Project 5. Automatically calculate changes in the extent of built-up areas, agricultural land and in such natural features as forest cover.

The availability of up-to-date land use / land cover information is central to many watershed management, planning and monitoring tasks. Satellite remote sensing images with their synoptic view, map-like images and repetitive coverage can contribute to agricultural, hydrological (including wetlands) and ecological models. Traditional methods of gathering data for such models (or indicators of change) are field surveys and air photographs – both of which are expensive, especially when they have to be repeated at regular intervals.

Although the Manifold 5.5 image analysis/GIS software used for Projects 1 to 4 does not have an image classification capability, other more specialised image analysis software, such as PCI Geomatica 9.1, can handle classification tasks. One of the more straight forward of such tasks being mapping of surface water.

The identification of similar pixels may seem to be a simple process, but there are many complications that limit the accuracy of results. These factors include:

Geometric registration of the remote sensing image (especially when dealing with multi-temporal data); The choice of land use / land cover classification schemes;
Selection of the appropriate spectral bands from the sensor data set;
Training data characteristics may not be truly representative of the land cover classes;
The type of classifier (e.g. maximum likelihood classifier assumes a normal distribution for input data and this may not be true);
Statistical analysis of spectral reflectance values does not necessarily represent real variations in the land cover; and
The problem of mixed pixels along the boundary of classified areas.

The obvious conclusion is that classification tasks are best left to the image analysis professional and are thus beyond the scope of this web site.