primary 
geographic 
intermediate 
requested 
geographic 
data 
computation y 
Figure 3: A tree-form processing plan 
Regarding manipulations on the geographic data, two 
sorts of processing models can be differentiated: 
- The first sort of processing model is used for “jum 
ping across the branches” of the target taxonomy 
tree. That is, the input data and the output data 
concern different target attributes or target classes. 
Computing a vegetation index or performing a land- 
use classification from sensor data are typical exam 
ples of this kind of processing model. These models 
are all anchored in the target class taxonomy. 
- The second sort of processing model is used to 
change the properties of a single piece of geographic 
data. The operation of these processing models is 
independent of the target classes or target attributes 
to which the data are related. Equalizing data (for 
contrast enhancement) or resampling data onto a 
different grid are two examples of this kind of pro 
cessing model. 
Processing Plans and Computations 
Processing plans are dynamic objects that represent the 
results of a consultation dialog. They may be displayed 
(by the RESEDA Advisor prototype) or executed (by the 
final RES EDA Assistant expertsystem).Processingplans 
are trees of concrete geographic data and computations 
that are connected to one another in alternating sequence 
(figure 3). 
A computation is a concrete manifestation of some pro 
cessing model. It computes some concrete geographic 
data (as defined above) from other geographic data. The 
algorithmic aspect of a computation is represented by an 
instantiated call of a computer program (including the 
values of the arguments). That is, a computation is 
defined by an abstract processing model, a couple of 
input and output data, and a call of a computer program. 
The root of the tree represents the desired geographic 
data being computed by the last computation, and the 
leaves of the tree stand for the primary geographic data, 
being used as input to the first computations of the whole 
network of computations. Any other piece of geographic 
data in the graph serves both as the output of some 
computation and as the input to at least one other compu 
tation. A tree of processing plans will fork for one of two 
reasons: 
- The tree forks before some computation: This indi 
cates that this computation requires multiple input 
data that must be determined before starting the 
computation. An example of such a computation is 
the process of generating a color-composite from 
three files of geographic data or a statistical classifi 
cation of multiple sensor channels. 
- The tree forks before some geographic data item: 
This indicates that there are alternate computations 
leading to the same piece of data. There are multiple 
ways to solve the analysis task reflected by multiple 
processing plans. The expert system will try to elimi 
nate (or avoid generating) useless plans (e.g., uninte 
resting variations of an already known plan) and will 
present the remaining plans to the user, who may 
select a plan (s)he prefers. 
A rule base contains the strategic knowledge necessary to 
apply the knowledge represented in the target classes, 
target attributes, and processing models for building 
appropriate processing plans. Some rules select appro 
priate processing models from descriptions of the data to 
be determined and instantiate the corresponding compu 
tations. Other rules are used to check and satisfy the 
constraints between the input and output data of some 
computation. 
150