Land cover data are available in image and vector formats with individual types of vegetation assigned to discrete classes. For example, in an image format each vegetation type would be assigned a unique numeric value and in a vector format each polygon would have attribute information that would describe the type of land cover in that polygon. These data are available with a wide range of thematic (classification scheme) and spatial (spatial resolution) detail.
The specific classification scheme used for a particular land cover data set can be as simple as forest/nonforest classes or as detailed as a species-level map. One important point related to thematic detail is that the more classes that are used, the lower the per-class accuracy will be. In other words, the classes in a forest/nonforest map will be more accurate than the individual classes in a species-level map. The spatial detail in a land cover data set is usually a direct result of the type of remotely sensed data on which the classification was based. Using aerial photography or high-resolution satellite imagery individual tree crowns can be discerned allowing improved capabilities for mapping species-level information.
For the most part land cover maps are created using data from optical sensors. One area where radar sensors excel is in mapping wetlands and water under forests, such as in flooded forests.
Land cover data sets can be created using manual and/or automated methods. The basic principle of land cover classification is to translate the pixel values in a satellite image into meaningful land cover categories. This is often accomplished using automated procedures, in which a computer algorithm is used to assign individual pixels or groups of pixels to one of the valid land cover categories. The classification process can also be accomplished using visual interpretation methods where the interpreter uses visual cues such as tone, texture, shape, pattern, and relationship to other objects to identify and group similar land cover types. In general, the human brain is better at interpreting the spatial characteristics in an image and automated algorithms are better suited for processing spectral (the many image bands) information. There are dozens of classification methods in use but there is not a single 'best' approach.
One of the possible limitations of classified land cover data is that the information is discrete instead of continuous. One way around this is to create a 'continuous fields' image data set for selected types of vegetation. In this data set each pixel value represents the percentage of that pixel covered by a particular land cover type. For example, in a broadleaf tree continuous fields data set a pixel value of 65 would mean that 65% of that pixel is covered by broadleaf tree species. In addition to different types of land cover it is also possible to create a continuous fields data set for imperviousness. This is called an impervious surface data set and it is being used increasingly in ecological modeling particularly when it is necessary to quantify water runoff.
Was this article helpful?