295 
these conditions are generally the situations 
we are most interested in, it is important to 
efficiently parameterize the model to limit 
numerical inaccuracy and to allow detailed 
investigation of the resulting system 
behaviour. Because we are aggregating 
continuous parameter fields into the 
parameter sets for each discrete landscape 
unit (hillslope in this case) it is important to 
avoid aggregation over significantly different 
environments (such as north and south 
facing slopes). Besides filtering out the 
extreme responses of the system, the use of 
mean parameter values may yield 
significantly biased model results relative to 
the true mean system response in the 
presence of strong nonlinearity. While more 
linearized models would appear to avoid this 
problem, we point out that natural systems 
tend to be very nonlinear, hence the 
nonlinear form of FOREST-BGC and the 
important constraints on model 
parameterization in terms of the admissible 
variance or range of parameter values within 
a landscape unit. 
DISTRIBUTED PARAMETERIZATION - 
RESSYS 
Geographic Information Processing 
The prepartion of the landscape unit 
template and estimation of the model 
parameters for each hillslope involves a 
combination of digital terrain analysis, 
remote sensing, and geographic information 
processing. RESSYS is an integrated data 
and simulation system that is currently under 
construction to accomplish the data 
preprocessing, model parameterization, 
model execution and output production 
(figure 1). As can be seen a substantial 
amount of the computation and analysis is 
devoted to raw data processing and 
combination. The results of these processes 
are the assembly of the fixed site 
information for each landscape unit. This 
information is stored in a cartridge file (table 
1, figure 1), which in combination with the 
seasonal base station meteorological 
information is fed info MT-CLIM which 
derives additional micrometeorlogical 
parameters for FOREST-BGC. 
The cartridge file acts as the 
parameter representation of the landscape. 
As alternative landscape objects can be 
defined to describe the landscape, such as 
composite watersheds, larger or smaller 
hillslopes, any cartridge file can be pulled 
out and another cartridge file inserted to 
model another representation of the 
landscape. This provides additional 
flexibility for exploration of efficient 
parameterization strategies. 
Digital Terrain Analysis 
A 30 meter resolution DEM (figure 2) 
was used to compute surface gradient and 
aspect, and to automatically extract the 
stream network and the set of hillslopes 
comprising the North Fork of Elk Creek. 
Band (1986, 1989) and Lammers and Band 
(1990) have described these steps in detail 
and they are not repeated here for the sake 
of brevity. At present we use a 
representation of the North Fork by seven 
stream links (the stretches of channel 
between stream junctions or a stream source 
and a junction) and fourteen hillslopes (two 
hillslopes draining into the two banks of 
each stream link, see figure 3.) Our 
methods allow the flexible decomposition of 
the watershed into a wide number of 
hillslopes, from just the mainstream up 
through the smallest tributaries. These 
different watershed representations would be 
stored as different cartridge files, but we will 
not discuss the effects of different watershed 
representation here as we use the fourteen 
hillslope watershed just for illustrative 
purposes. 
The gradient and aspect of each 
hillslope is retrieved by computing the 
resultant vector from the set of surface unit 
normal vectors of each DEM pixel in each 
hillslope. This effectively models each 
hillslope as a facet with the spherical mean