Deriving nationwide estimates of forest variables for Sweden using Landsat

The use of satellite data in forest inventories makes it possible to do a wall-to-wall mapping of forest variables. This type of raster databases with estimated forest variables provides a great potential for planning how the forest resources should be utilized, and can be used as input ecological models. There are different remote sensing methods that can be used to map forest resources. One is the k-nearest neighbours (kNN) method that has been successfully tested and evaluated in different countries. In this method, multiple forest variables are simultaneously calculated for individual pixels as weighted mean values of spectrally nearby field samples. The objective of this project is to derive raster databases with estimates of forest variables, for example volume and tree height, for the whole of Sweden. This is done by combing Landsat ETM+ data and field data from the Swedish National Forest Inventory (NFI) using the kNN algorithm. Approximately 40 Landsat ETM+ scenes will be used in the project. The accuracy obtained for estimated variables can be considered poor on the pixel level. However, the accuracy will improve quickly when estimated variables are aggregated into larger areas. In Sweden, the accuracy for stem volume per hectare, for example, will be in the order of 15 % (RMSE), when calculated for an area of approximately 150 ha. The project is carried out in co-operation with the Swedish National Board of Forestry, the Regional Forestry Boards, and the Swedish Environmental Protection Agency.