Talk to be held at the workshop on image analysis and spatial statistics in forestry on November 2, 1999 at KVL, Frederiksberg, Denmark
Tomas Brandtberg
(tomas@cb.uu.se)
Centre for Image Analysis, Swedish University of Agricultural Sciences,
Uppsala, Sweden.
During the 90's many different laser scanning systems have been applied and tested on forests. Usually, the presented results of the analysis are based on extraction of maximum height value within a square unit area or, e.g., mean stand height. This means that very little of the captured information is utilized. On the contrary, an analysis approach based on the visible (i.e., typical overstory) individual trees has the ability to make use of more of the remotely sensed data. Such an approach is briefly described below.
To capture the local property of the data in the forest the height differences between the original laser height value and an optimally Gaussian smoothed surface is studied. Fuzzy set algorithms are used to extract different subsets of the data and for further processing of the derived artificial images. Local maxima in the calculated membership value image (i.e., an artificial image) are found and are similar to the tree-tops. Individual tree-based analysis of laser scanning data is thus feasible. The final result for each visible tree crown is a two-dimensional (2D) surface existing in three dimensions (3D). The surface describes the height information of the whole tree crown. The complete surface is of interest to predict stem properties, and not just its maximum height value. Further developments of our system might also include an estimate of the tree crown overlap.