Olsson, Häkan
2.3 Field Data
Field data were collected as a part of the normal operations of the Swedish National Forest Inventory. Sample plots of
7m (temporary) or 10m(permanent) radius are arranged in clusters on a regular grid. Variables are recorded by field
teams on the sample plots according to an established protocol. Observations are made about the soil and hydrology
conditions, ground vegetation, tree size and species composition. Several variables are derived or calculated for the plot
level, based on sample trees. The database was queried for all plots measured within the study area during 1985 to
1996. Due to the cyclic nature of the inventory, and the large proportion of temporary plots, very few in this set were
actually re-measured during the period. In total 1545 plots were available in the study area after screening for clouds.
30 600
20 1
HEIGHT S
VOL TOT
0 100 200
AGE BA AGE BA
Figure 2a) Stand height v.s. age shows significant variation due to site productivity. b)
Stem volume (per hectare) as a function of age
Unfortunately the position information for the plots is rather variable. Plots have nominal coordinates according to the
inventory design. They may have digitized coordinates from 1:10000 scale maps, and the most recently surveyed plots
are positioned with differential GPS. In all cases the best positional information available was used to extract pixel
values from the image for comparison with the field data.
3 METHODS
3.1 Image Normalization
Differences in atmospheric clarity during the different dates of image acquisition mean pixel values are not directly
comparable between images in the temporal sequence. In general the atmosphere affects the recorded radiance through
attenuation (a wavelength-dependent multiplicative effect) and path radiance (a wavelength-dependent additive effect).
One approach to calibrating for these differences is to first convert the recorded digital counts to radiance through
published calibration factors, then model the atmospheric radiative transfer to convert to surface reflectance factors with
can be compared. This however requires in situ measurements of atmospheric parameters during the time of image
acquisition. For a recent discussion of procedures, see Ouaidrari and Vermote, 1999. In the absence of in situ
measurements, the atmospheric parameters may be inferred from radiance over dark targets such as clear lakes (Teillet
and Fedosejevs, 1995). This approach to calibration has the advantage of being based on physical considerations, and
results could be compared to model outputs. However with Landsat-5 at least, the main difficulty may be retrieving
realistic radiance values from the recorded digital counts. After several years of operation, the on-board calibration of
Landsat-5 was essentially unknown, and different receiving stations applied different calibration factors at different
times for producing data. For a discussion of the issues, see the appendix in Teillet and Fedosejevs (1995). Without
sensible starting values for radiance, the atmospheric correction procedures have almost no chance of producing
1084
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.