Spatial tree pattern analysis from aerial photographs

Figure 5: Aerial photo of a forest stand, with traces of image processing

High resolution aerial photographs of a forest can, in the hands of foresters, statisticians and computer specialists, give surprisingly precise answers as to the number of stems in the forest stand, the positions of the single trees, health status, the estimated volume of wood and other properties of interest for forest and landscape monitoring. One key project in Dina has been to develop such methods, building on the theory of spatial point processes which has been growing vigorously in later years. A joint effort between KVL and FSL, the project has surpassed even optimistic expectations and brought Denmark into a front position in the area. The visible expression of this achievement consists of two PhD theses and a series of papers in international journals by Dina researchers: K. Dralle, M. Larsen, J. Lund and M. Rudemo.
The number of stems can be estimated by smoothing the digitized aerial photograph by a so-called Gaussian kernel which can furthermore be used to find tree positions. Another method based on an optical single-reflection model allows for an accurate analysis of images obtained from varying viewing angles. Observational errors and simulations to support these forestry applications have also been studied in detail.
The methods developed so far mostly pertain to homogeneous forest stands where all trees are of the same species and the same age. Further improvements are needed to develop the methods into a useful tool for monitoring forests on a regular basis, for example from aerial photographs every fifth year, or from the new generation of high spatial resolution satellites. In the near future the Dina group will investigate the problem of mixed forests and also focus on spatial analysis of multiple photos.