Epsilon Archive for Student Projects

Abstract

Sweden’s forestry legislation was updated in 2010 and a new definition of forest land was introduced. This definition was adapted to the one used by the Food and Agriculture Organization of the United Nations (FAO) for international statistics on the state of the world's forests. It is in short based on the lands ability to grow forest that reaches 5 meters, 10 % canopy closure and has a continuous distribution, according to FAO at least 0.5 hectares. A country-wide laser scanning is now carried out for the production of a new national elevation model; the laser data also provides information on forest height and density. The mountain birch forest makes up much of the border with other land types, and to map the distribution of forest land here would be of interest. A distribution map could provide information such as how much forest land that lies within protected areas.
In this study, laser data was used to classify the forest in the Abisko area, using reference data from sample plots. From the point cloud obtained from the laser scanning, different types of metrics were calculated and used to classify the forest. Classification results were evaluated by cross-validation, suggesting an overall accuracy of 92% and a kappa coefficient of 0.85. Despite the high accuracy, there were problems associated with a somewhat small sample of ground reference data. In order to separate forest that meets the requirements of forest land from forest which is high but not dense enough, more reference data would be needed. Steep and stony terrain also caused some problems, where the edges and rocks in some areas were mistaken for vegetation. The methods and problems that emerged from this study can be important experiences for potential future mapping of the forest land in the Swedish mountains.