Epsilon Open Archive

Abstract

As part of the "Monitoring of forest ecosystems" project, within the MISTRA
program Remote Sensing for the Environment (RESE), and also with funding
from the County Administration Board of Dalarna, a demonstration project
was undertaken to estimate forest stand parameters in Dalarna with the use
of satellite data. Using two full scenes of Landsat Thematic Mapper data and
sample plot data from the Swedish National Forest Inventory, estimations of
above ground tree biomass, age, total wood volume, and separate tree
species' volumes were made using the "k Nearest Neighbor'' method.
Accuracy assessment results at the single pixel level for total wood volume
are consistent with results from previous kNN estimations, with an overall
relative RMSE of 75% for the western scene, and 58% overall relative RMSE
for the eastern scene. Validation data show a bias of the estimate toward the
mean value of the estimation data. The pixel level estimates of above ground
tree biomass and age had similar validation results to those for total wood
volume. Biomass estimates had a 77% relative RMSE for the western scene,
and 69% for the eastern scene. Age estimates had a relative RMSE of 60%
in the western scene and 57% in the eastern scene. The results may suggest
the need to incorporate a geographic limitation on the plots used in the
estimation, and to further investigate the co-registration between the satellite
and plot data. While pixel level errors are high, an aggregation of the
estimates to larger (compartment-sized) areas could decrease the error significantly. Previous similar studies have shown that an RMSE of 10% for
total wood volume can be obtained for as small areas as 1 00 to 450 hectares.
The estimates from this study will be evaluated for use by the County
Administration Board of Dalarna to find areas of ecological interest and to
assist in planning.