Reese, Heather
ILLUSTRATIVE USES OF CONTINUOUS ESTIMATES OF FOREST PARAMETERS DERIVED
FROM SATELLITE DATA
Heather REESE, Per SANDSTRÖM, Mats NILSSON, and Häkan OLSSON
Remote Sensing Laboratory
Department of Forest Resource Management and Geomatics
Swedish University of Agricultural Sciences (SLU)
Umeä, Sweden
Heather.Reese @resgeom.slu.se
KEY WORDS: applications, satellite data, forest inventory, continuous estimates, kNN, planning, habitat
ABSTRACT
Continuous estimates of forest parameters such as total wood volume, wood volume by species, stand age and
above-ground tree biomass have been made for several areas within Sweden by combining data from the
sample plot based Swedish national forest inventory, satellite imagery (both Landsat and SPOT), and map
data. Four different projects are discussed here with study areas ranging from central to northern Sweden
where the vegetation is typified by managed boreal forest.
Field data from the forest inventory plots ranging from a five to nine year time span were used to aid in
making the estimates. The method for estimation was a "k-Nearest Neighbor" algorithm, in which forest
variables were calculated as weighted means of k spectrally nearby samples. The output grids contain pixel
level continuous estimates for each of the variables, which were volume, volume by species, stand age and
above-ground tree biomass. Validation results show that accuracy of the estimates for all parameters is low at
the pixel level (approximately 55-7596 RMSE), with a tendency toward the mean, and an underestimation of
higher values while overestimating lower values. However, when the estimates were aggregated to larger
areas (e.g., 20 ha and larger), accuracy has been improved to give lower errors, ranging from 3096-1090
RMSE depending on aggregation area and method of estimation.
These estimations represent a database which can give a landscape view of parameters normally collected as
spatially intermittent plot data. Due to the results of these projects, estimates for other areas in Sweden have
been requested by agencies such as county forestry boards, county administrative boards, university
researchers, and others. As interest grows, it is useful to examine how these data can be used. Examples of
ongoing and potential applications for these estimates are presented, such as a moose habitat use study, county
level planning activities, use as input to prognostic programs, and for computation of statistics within drainage
basins and smaller land holdings. These examples illustrate how other end users in Sweden might apply
estimation data such as these.
1252 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.