■ to extract site variables such as ground moisture and sub-surface waterflow from a DEM, from a soil 
map and from a cadastral map. 
- to test the digitally-extracted site variables against manual measurements in the field. 
- to produce a site index image over the study area. 
MATERIAL 
The study area was a 5x5 km square of mainly forest land situated about 10 km outside the city of UmeS 
in northern Sweden (64° N), see map in FIGURE 4. The site conditions are extremely variable for this 
climatic region with a maximum forest production potential of 8 irr^/ha*year \ corresponding to site 
index G28. 
The following map materials were used: 
- The cadastral base map on scale 1:10.000. Streams and roads were digitized in raster format with 10 
meter resolution. 
- A soil map on scale 1:100.000. The map was enlarged to 1:25.000 and digitized in raster format. 
- A standard DEM from the Land Survey with 50 meter point distance. The DEM was resampled by bilinear 
interpolation to 10 meter point distance. 
Thus the map material consisted of three digital raster images with 500x500 pixels and 10 meter 
resolution. 
A field reference data set was obtained from observations made on 1000 ten meter radius plots. The site 
variables were manually estimated on each plot using the existing SCS. The plots were also located in 
the Swedish National Grid (coordinate system) for comparisons with the digital images. 
METHODS AND RESULTS 
Processing the DEM 
The DEM was used to determine several basic topographic data for each pixel in the image: 
1. The slope in each pixel was defined as the highest altitude difference between the pixel and its 8 
neighbors, divided by the horizontal distance. 
2. The ground water flow direction from a pixel was defined by the minimum range of directions that 
would just include the lowest 2 pixels out of its eight neighbours - provided that the two pixels 
were lower in altitude than the pixel itself. A "downhill pointer" was set toward each of these two 
pixels. The reason to pick 2 downhill neighbors instead of 1, was to allow the water flow to spread 
out when it reaches a convex part of the ground surface. See also JENSON & DOMINGUE 1988. 
The flow direction was completed by merging it with the streams in the cadastral map. In pixels 
coinciding with a stream, the flow direction was erased, implying that (a) water that hits a stream 
is lost for the forest land and (b) the ground water flow can not cross the stream and continue on 
the other side. 
3. Given the downhill pointers above, the groundwater flow was simulated. The net precipitation was set 
to one unit in each pixel. The water flow was then simulated, starting in pixels with no flow 
directions pointing towards them (i.e. hilltops). The precipitation in these pixels was added to 
their downhill neighbors, with the assumption that the water is shared equally between the two 
872