wRESEx - Followup of environmental objectives and preservation strategies for forest and wetland using remote sensing and GIS (2001-2003)
Partners: Örebro University, Metria Miljöanalys, County Administrative Boards of Dalarna and Gävleborg
The quality and dynamics of habitats, their suitability and temporal and spatial behaviour need to be monitored. Evaluations and analyses within environmental monitoring have to this day been based on information from samples of field data together with ancillary map data. The new development within regional environmental work has created a need of new and complementary information with a total geographic cover. This need was the basis of the project wRESEx.
Biotopes and elements that indicate high environmental quality from a biodiversity perspective are today quite rare and the existing monitoring programs linked to landscape surveillance, e.g., the Swedish National Forest Inventory, do not permit evaluation of data at a local or of smaller regions. Remote sensing of satellite images and/or aerial photographs offer potential to assess wall-to-wall changes in the health of ecosystems, identify risks of further degradation and opportunities for restoration. So far, however, little of this potential has been realized.
In the project three environmental indicators were tested: the deciduous component, old-growth forests and clear-cut areas. Classification of this type of objects is extremely complicated and reported classification rates are usually poor. For the following reasons traditional methods will not give satisfactory results:
· The number of field plots in different classes varies too much leading to an unfavourable treatment of sparsely occurring objects
· Non-contextual methods neglect information from neighbouring plots and the feature vector is not normally distributed.
· Parametric contextual classification methods do utilize information from neighbouring plots but the demands for field data increases.
· Estimation of some parameters in the model is unreliable and in many situations impossible.
To overcome these problems semi-parametric approaches should be used. Information from neighbouring pixels is utilized by using the wavelet transform with the following purpose:
· Denoising the original images by removing small wavelet coefficients and inverting the wavelet transform. Hereafter an extended and improved version of the Nearest Neighbour classification algorithm (NN-classifier) is applied.
The problem with too few field plots was solved using additional field data from the forest companies. However, the quality of NFI-plots and compartment information is very different. Thus new data editing methods were needed in order to remove outliers due to poor quality of field data, and to find out the prototypes for each class in the case of nearest neighbour classifiers.
The remote sensing methods developed are based on a new approach for classification of multitemporal satellite data sets, combining multispectral and change detection techniques.
This project is an example where the RESE-project has generated new constellations.