Ted to
ie-like
)OX of
ing to
with
estrial
le, an
reased
enetic
11/
1/
algorithm, and would be monitored by satellite UV
sensors. The rules would allow the system to be scaled
for local or regional applications, and might then be
integrated into global models.
The approach outlined in Table 1 conceptualizes how the
internal complexities of natural genomes might be
integrated with observable environmental attributes to
simulate dynamical systems. It represents a logically
structured "bit map" for coding and storing complex,
interactive knowledge about an ecosystem as strings of
binary data that could then be machine-processed in a
virtual reality environment to monitor adaptive reactions.
As new, or better knowledge became available, the
cybergenes could be changed or lengthened; or, whole
new cyberchromes could be added as data from different
scientific areas became available. Perhaps the most
important feature of such a genetic algorithm is that the
rules and neural networks governing self-adaptive
behavior would reduce the level of human bias now
common in ecosystem models.
Current ecosystem models are based largely on an
assumption of linearity that accounts for neither
environmental perturbations nor identifiable periodicities.
Instead, they project evolutionary trajectories based on the
straight-line portion of time functions representing
"today", and which are assumed will continue far into the
future. Indeed, there are many practicing professionals
who are convinced that resource “management” equates
to keeping everything as it is.
In short, current models are not designed to self-adapt.
They do not ride the fuzzy edge of Order and Chaos to
maintain high levels of structure and function without
sacrificing flexibility (Kauffman, 1993). Evolutionary
trajectories of species and ecosystems are not random, but
neither are they linear. Evolution is normally such a slow
process that adaptive reactions are only observable after
the fact. Some critical questions therefore are..."how can
the science community gather and process data fast enough
to drive a genetic algorithm?"; "how can fuzzy rules and
artificial neural networks be incorporated to simulate an
ecosystem's internal "learning" process?"; and, "how can
routinely collected satellite environmental data be used in
a time-dimension to visualize dynamic processes?"
This last question is especially interesting for the remote
sensing and GIS communities because modelers will have
neither the variety nor the spatial continuity of in-situ data
necessary to drive dynamical models over large areas. As
an alternative to discontinuous, in-situ measurements,
scientists can search remotely acquired sensor data for
ecological and genetic information, and thus approach the
exercise from satellite to ground. Pixel digital numbers
(DNs) from time-series data would become "windows"
through which genetic and ecologic systems are linked..
ROLE OF REMOTE SENSING AND GIS
Genome maps and genetic networks provide the
foundation for understanding how species respond and
adapt to environmental stimuli. Another part of natural
complexity involves measuring the attributes that
constitute those stimuli, and understanding how they vary
In time and space. Remote sensing is perhaps the best
means for obtaining many of these data because satellite
507
sensors can collect uniform global data that are applicable
at local scales and can thereby provide time series
comparability. Similarly, airborne sensors collect fine
resolution data at the local level but can be flown over
analogous sites around the world to obtain comparable
geographic data. For developing self-adaptive models,
however, the greatest asset of satellite measurements is
that they represent repetitive, synoptic, calibrated data
sets that satisfy the criterion for "continuity" in rule-based
models.
Remotely acquired data come in both raster (image) and
vector (sounder) formats depending upon sensor design
and measurement technique. Both kinds are directly
employable in Geographic Information Systems (GIS),
which have proven their value as a means for managing
and displaying relationships between data sets.
Traditional data collected in the field will not be replaced
by remotely acquired data because these, too, are directly
employable as point, line, and polygon attributes. The
combination of remote and ground-based data types
constitute the universe of data that could reside in a
relational database sandwiched between the object
databases. All of these, in turn, would be employed to
form a genetic algorithm that would describe an
ecosystem. Figure 2 illustrates this design.
Cultural & Physical Relational
Milieu Databases
Figure 2. Object and Relational Databases
A critical requirement of any relational GIS is that all data
sets be accurately co-registered. In past, this has been
accomplished by geocorrecting images using ground
control points and by improving the pointing accuracy and
pixel-to-pixel registration of sensors. Another technique
is to use Global Positioning Systems (GPS) capable of
geolocating raster, point, line, and polygon features with
accuracies in the centimeter to meter range. GPS is the
most recent addition to the spatial data tool kit. It
represents the enabling technology for merging data sets
into spatially coherent systems, and offers the hope that at
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996