A number of simulators in recent years, including the predecessor to this one, have used spatially explicit, individual-based modeling to explore ecosystem behavior [6,15]. This approach offers great promise in several areas. One is that aggregate population-level modeling can become intractable with only a few species, whereas individual-based modeling is not inherently more difficult in many species than in a few. Thus more complex ecological management questions can potentially be asked with an individual-based model.
Another promising area is in the viable range of the model, sometimes called the ``nanofox problem'', in reference to the classic ordinary differential equations describing rabbit and fox population swings. A differential equation model of two populations is only valid so long as there are two populations carrying on normally. It can describe, but not extrapolate. Yet in environmental management, it's often the breakdown portion of ecosystem behavior, when self-correcting dynamics no longer function, that we wish to understand. Individual-based modeling does not presuppose understanding of how the aggregate behaviors break down, but rather plays out the scenario, and observes the overall breakdown.
To what degree playing out a mock scenario is predictive of natural behavior depends on the realism of the mockup. For experiments with perfect realism, only the living model will do [8]. But even with a living system, we need a theoretical framework within which to pose the questions---what must be held constant, what can safely be overlooked, what data to collect, what constitutes a ``significant'' answer. Computer simulation is potentially ideal for working on such frameworks. One abstracts a certain subset of biological reality into the model, plays out the scenario, and observes whether that subset is sufficient to explain (generate) some natural phenomena.
The subset of biological reality that spatially explicit models attempt to abstract is that populations don't interact, individuals do, and they do so locally. Population-level events are an aggregate of local events, but not simply as a sum. Some local events may remain local in their effects, such as a species introduction or mutation that dies out. Some may snowball, as with a species introduction that thrives and spreads, as with the rabbits in Australia. Others can have indirect strong effects, as with forest fragmentation increasing the edge habitat so stimulating to the deer population in the northeastern United States. Many environmental management issues, such as disease spread and containment, species re-introduction, and habitat conservation, pivot on this localization.
This version of Gecko's innovations lie in spatial modeling and another crucial aspect of biological reality, energetics. There have been a number of ecology models based on discrete lattices [10,9]. Swarm Gecko's prototype, Echo Gecko [15,1,3,4], was spatially explicit in one discrete dimension---individuals had position and range on an elastic one-dimensional ring. These spatially explicit models demonstrate the effect of local interactions on the aggregate dynamics. But although agents in these discrete models have position and range in one or two dimensions, they do not have extent. The size of the agents bears little or no reflection in the ``space'' used to model them. Neighborhood definition and crowding competition rules are specified, but specified independently from the space the individual agents occupy and interact within. The consequences of agents differing in size are also divorced from their spatial representation. Yet these very same models indicate how crucial localities are in driving the ecosystem dynamics. Whether or not the lattice or list abstractions of space are appropriate depends on the sophistication of these independent neighborhood rules, which is difficult to evaluate save by empirical calibration.
Gecko uses no lattice. Agents have free range in two dimensions. They have extent and compete directly for space. Crowding is an endogenous result of resource availability and neighbors competing for those resources. This enables Gecko to show subtle community-level effects of spatial competition which would not emerge from a lattice representation.
Energetics are also modeled explicitly. To survive in Gecko, an agent must find food, assimilate it with realistic inefficiency, and use the energy to pay metabolic costs and grow. Realistic assimilation rates engender a realistic pyramid of numbers that agents must search through to find food. Agents must grow to a certain size before they can reproduce. Size and metabolism are tied through allometry, the empirical laws of scaling fundamental throughout nature. Allometry refers to the fact that organisms across size classes do not scale isometrically, as with isometric triangles of the same angles, but in a proportion dictated by physical constraints. Thus, for instance, a mouse and an elephant, though both of the mammal body plan, cannot have isometric body proportions or metabolism [13]. Via allometry, Gecko's spatial representation of agents is integrally tied to energetics.
This paper describes the Gecko simulator and its rules. The power of the approach is demonstrated with stable artificial ecosystems of not only multiple trophic levels, but multiple species within a trophic level. An interesting feature of one ecosystem is that multiple sessile species coexist indefinitely in the same environment, without specialization. Simple spatial properties are sufficient for this coexistence.