Pre-2002 Research Projects

My main focus is on the application of evolutionary computational techniques to ecological modelling, particularly of evolving populations and ecosystems. Below, the first two projects cover this topic, while the third, MIGMOD, looks at the use of individual-based techniques for migration modelling.

EUZONE: Evolution of Aquatic Ecosystems

Ideal Free Distributions: Their Emergence via Genetic Algorithms

MIGMOD: A General Migration Model for Biological Impact Assessment

EUZONE

A typical EUZONE session, complete with a graphic display of the aquatic environment, dynamic time series of probed variables, and an inspector for plankton phenotypes.

This research involves the simulated emergence and evolution of aquatic microorganism communities, with a special focus upon the bidirectional relationship between these populations and their physicochemical environments. The EUZONE system enables investigations into both (a) the effects of local habitats upon the nature of their biota, in terms of the emerging ethologies and population dynamics, and (b) the reverse influence, that of organisms upon their habitats. These feedback loops between biota and their environments embody Gaia theory (Lovelock, 1991), which essentially broadens the scope of evolution from organisms and populations to entire ecosystems.

EUZONE is an acronym for euphotic zone, the upper layer of the ocean where net phytoplankton (i.e., algae) growth occurs. Below this layer, which can range from 5 to 100 meters deep (depending upon the prevailing atmospheric and and oceanographic conditions), the attenuated solar irradiance is too weak to support significant photosynthesis. Aquatic life originates in the euphotic zone in the sense that algae harness solar energy and consume inorganic compounds (carbon dioxide, nitrates and phosphates) to produce carbohydrates and proteins - thus forming the lowest level of the food web and the basis for most other aquatic biomass. The growth or blooming of phytoplankton is therefore labelled primary production.

EUZONE goes beyond conventional plankton simulation by introducing evolution, via genetic algorithms and genetic programming. If Gaia theory is correct, then as plankton evolve, inorganic aspects of the ecosystem, such as chemical concentrations, temperatures, and even climate, should also change. In addition, plankton are the lowest level of the aquatic food web, so their evolution will surely affect higher trophic levels. In short, plankton appear to be "lever points" (Holland, 1995) in several important complex systems, and EUZONE positions them at the center of its modelling paradigm.

In the spirit of contemporary artificial life (alife) research, this work provides a virtual laboratory for the emergence of complex ecosystems from the interactions of simple organisms. However, whereas most alife systems abstract away many real-world environmental constraints, EUZONE employs detailed physical and chemical models in combination with evolutionary computational constructs (i.e. genetic programming and genetic algorithms) to support the emergence of carbon-based aquatic ecosystems.

Using Euzone

The Simulation of Gaian phenomena in EUZONE

Goals of the EUZONE project:

To simulate the emergence and evolution of low-level aquatic food webs.

To gain a better understanding of Gaian interactions between primitive organisms and complex physical phenomena, such as climate, via simulations in which evolving organisms lead to evolving ecosystems and changes to the physical environment.

To simulate the emergence of different life-history strategies in organisms as a function of the environment.

To develop information-theoretic metrics (e.g. entropy models) to assess the emergence of order in complex ecosystems.

EUZONE publications:

Downing, K. (1997) EUZONE: Simulating the Evolution of Aquatic Ecosystems. Artificial Life, vol 3(4), pp. 307-333.

Downing, K. (1998), Combining genetic programming and genetic algorithms for ecological simulation. Proceedings of the 3^rd International Conference on Genetic Programming, Madison, Wisconsin.

The Evolution of Ideal Free Distributions

EUZONE has also spawned a general interest in the evolution of interacting agent/organism populations. Of particular interest is the emergence of spatial and temporal ideal free distributions (IFDs) in nature. In a nutshell, an ideal- free situation occurs when a population distributes itself over time and/or space to match the temporal/spatial resource distribution.

For example, if the larvae of a particular insect can emerge from dormancy on any of k days (they often mate and die within a day or 2), then their survival will depend primarily on the availability of resources on their emergence day. Since all emerging larvae must share the available resources, there is no single optimal day on which all larvae should emerge (assuming that non-zero amounts of resource are available on most of the k days). Ideal-free distribution theory predicts that the proportion of larvae that emerge on day d is proportional to the relative amount of resource available on day d. Hence, the emergence patterns will reflect the resource distribution curves.

Preliminary tests with simple genetic algorithm (GA) simulations indicate that temporal ideal free distributions of larvae emergence times can be generated in less than 20 generations (1000 individuals). The emergence-time distributions will match a wide variety of resource curves, from linear slopes to sine curves of varying amplitudes and periods. A slightly more complex model incorporates sex differences to simulate the evolution of protandry, wherein females emerge in response to resource curves, while males emerge in response to available females. Protandry, a well-known concept in insect ecology, results when the male curves rise and fall several timesteps before the female curves. Once again, simple GA models were able to simulate the emergence of the protandric ideal free distribution in a couple dozen generations.

IFD Publications

Downing, K. (1997) The emergence of insect protrandry: a "natural" evolutionary computation application. Proceedings of the 4th IEEE International Conference on Evolutionary Computation, Indianapolis, Indiana.

Downing, K. (1997) The emergence of emergence distributions: using genetic algorithms to test biological theories. Proceedings of the 7th International Conference on Genetic Algorithms, East Lansing, Michigan.

MIGMOD

The General Migration Model (MIGMOD) exploits the object-oriented paradigm to capture the idiosyncratic daily and seasonal migratory movements of birds, polar bears and sea mammals. The migratory patterns generated by MIGMOD are compared to the simulated trajectories of oil and chemical spills to provide a quantitative measure of the biological impacts of acute aquatic pollution.

For each species of interest, MIGMOD employs biological field data concerning (a) traditional site locations and activities, (b) ranges of arrival times and visitation durations at these locations, and (c) general and age/sex specific migratory behaviors. This data drives stochastic movement routines that are tailored for each species and age/sex class. Gridded habitat, ice, tenacity and bathymetric data are also used to govern the simulated animals' choices of sites and migrational paths. This data-based approach avoids many of the difficulties of first-principled (i.e. causally-based) migratory modelling, wherein facilities such as gradient sensing and following and animal memory must be modelled. In short, MIGMOD employs migratory data concerning "where" and "when" to circumvent the problems of modelling "how" and "why".

Goals of the MIGMOD project:

A general-purpose migration model

The ability to perform biological impact assessment of oil/chemical spills without making simplifying assumptions concerning uniform or random distributions of animals within a region.

Status of the MIGMOD project

MIGMOD has been fully implemented and tested in Microsoft's Visual C++. To date, birds, seals and polar bears have been modelled. In 1994, Norsk Hydro, one of Norway's largest oil companies, used MIGMOD to investigate the biological consequences of oil drilling and shipping in the Barents Sea.

In the future, we plan to couple MIGMOD to a population dynamics model in order to assess the long-term effects of acute pollution episodes upon animal populations. Another future goal entails a multiple interacting species version of MIGMOD ( In the present model, each species must be simulated under separate runs). This ecosystem model would enable, for example, fish schools to follow plankton blooms; birds and seals would then follow the fish, while polar bears would in turn follow the seals.

MIGMOD publications

Downing, Keith and Mark Reed (1996), Object-Oriented Migration Modelling for Biological Impact Assessment. Ecological Modelling, Elsevier Publishers, vol. 93, pp. 203-219.

Downing, Keith (1996) . A data-driven, object-oriented approach to migration modelling. Pacific Symposium on Biocomputing, Kohala Coast, Hawaii.