296 
gradient and aspect, ignoring all surface 
variability within the hillslope. This 
variance, and hence the variance of received 
insolation, rises with the degree of surface 
dissection relative to the number of 
hillslopes used in the watershed 
representation. In this manner, there may be 
model sensitivity to the watershed 
representation as expressed by the cartridge 
file. While we are currently investigating 
this problem, it is not discussed here. 
Estimation of LAI and SWC Fields: LAI 
over the vaiershed was estimated at a 30 
meter scale using Thematic Mapper imagery 
that was registered to the DEM. Peterson, et 
al (1988) have previously noted the good 
correlation of Thematic Mapper band 4 to 
band 3 ratio with the stand level LAI over 
this area of western Montana. This 
relationship is most applicable under 
conditions of closed forest canopies, but as 
the canopy opens up, understory radiance 
contributions to the ratio degrade the 
correlation. Using the same field plot data 
as Peterson, et al (1988) we have found that 
normalizing the ratio by multiplication with 
a ratio of (1.0-b5)/(1.0+b5) significantly 
improves the relation over the range of 
observed canopy conditions. This appears to 
work in our study area as the higher b5 
response from grass in comparison to conifer 
canopy produces lower values of the ratio, 
which in turn reduces the normalized b4/b3 
ratio, improving the ability to correlate the 
normalized ratio with observed canopy LAI. 
We do not present this as a general solution 
as it depends on the spectral reflectance and 
homogeneity of the understory. Inspection 
of the cartridge file (table 1) shows that 
north facing slopes tend to have higher mean 
LAI than south facing slopes, except where 
disturbance has been significant, as in the 
burned area (low LAI) in the watershed 
headwaters. 
SWC is retrieved by assigning typical 
values to mapped soil series found in the 
area and digitizing and registering the soil 
series map to the DEM. Rasterization of the 
soil polygons and overlay allows retrieval of 
the SWC encountered in each hillslope. 
A basic problem with this method should be 
noted. While LAI, or at least the TM 
imagery, is sampled with the same frequency 
as the DEM, and therefore slope and aspect, 
the soils information is sampled and mapped 
in a very different manner. The process of 
compiling a soil series map involves a 
significant amount of intuition, interpretation 
and generalization. Although soil properties 
are known to vary significantly over very 
small distances even within mapped soil 
series, the cartographic product of the soil 
map does not represent that behaviour. 
Instead, each soil polygon is represented as 
having zero variance in its digitized form. 
While the approximate field variance may be 
known or estimated by the soil scientist, that 
information cannot be carried through the 
geographic information processing by the 
overlay analysis. The effect of this is that 
SWC appears to have little variance over an 
area when there may be very large variance 
in the field. This leads to the potential of 
having significant bias in the ensuing 
simulations due to model nonlinearity as 
previously discussed. This will be most 
important in water limited environments, 
under which FOREST-BGC is very sensitive 
to SWC. This is discussed here not because 
we have a solution to the problem, but 
because it is much more general than our 
specific application, and we raise it as it is 
propagated and perhaps exasperated by the 
common overlay process in GIS. 
DISTRIBUTED SIMULATION OF ELK 
CREEK 
FOREST-BGC was run for the 
fourteen hillslopes described in table 1. 
Model results for ET and PSN are given in 
table 2 and ET on a hillslope by hillslope 
basis is draped over the DEM in figure 3. 
The range of ET is 38.9-51.9 cm./yr. while 
PSN ranged from 7.4-8.6 Mg./yr. Through 
most of the watershed, the major differences 
in ET and PSN are associated with the