EcoSim

The World

Procedural generation with real climate science, not just Perlin noise.

Each simulation constructs a complete planetary surface from scratch. Seven stages in sequence, each building on the last:

1. Heightmap synthesis: layered Perlin noise produces continental masses with realistic coastlines and mountain ranges
2. Latitude banding: horizontal position sets base temperature, colder at poles, warmer at the equator
3. Elevation lapse rate: temperature drops with altitude at ~6.5°C per 1,000m, the real atmospheric lapse rate
4. Rain shadow modelling: prevailing winds deposit moisture on windward slopes, leaving leeward regions arid
5. Precipitation distribution: combined moisture from wind patterns and elevation creates varied rainfall zones
6. Whittaker classification: temperature + precipitation maps every tile to one of 22 biomes using real climate boundaries
7. Flora seeding: plants spawn with biome-appropriate traits, establishing the initial food web

The Genetics Engine

The crown jewel. Not hardcoded species, but emergent traits from continuous gene values.

Every organism carries a full genome of 90 genes organised across 8 chromosomes, each gene a continuous floating-point value between 0 and 1. There are no species definitions. No creature type enums. What an organism is emerges entirely from what its genes produce.

Diet type illustrates this perfectly. Rather than storing "herbivore" or "carnivore" as a flag, the simulation calculates diet from digestion gene values. A creature with high plant digestion and low meat digestion becomes functionally herbivorous, not because it was labelled that way, but because its genes make plant-eating efficient and meat-eating wasteful.

Four dominance patterns govern inheritance: complete dominance, incomplete dominance, codominance, and epistasis. When two organisms reproduce, their chromosomes undergo crossover and mutation, producing offspring that are genuinely novel combinations, not clones, not random.

Creature Intelligence

State-machine AI driven by needs, with gene-controlled pathfinding.

Each creature runs a hierarchical state machine that selects behaviour based on current needs: hunger, energy, reproductive urge. The priorities shift dynamically as internal states change.

Pathfinding uses A* search, but with a twist: the environmental_sensitivity gene controls how much terrain cost matters in route calculation. Creatures with high sensitivity stick to familiar, low-cost terrain. Creatures with low sensitivity take direct paths regardless of obstacles. This means pathfinding strategy itself evolves. Cautious navigators and bold chargers emerge from the same algorithm with different gene values.

The AI system was originally a 2,420-line monolithic class. Extracting it into a modular Strategy pattern, one class per behaviour dispatched by the state machine, was one of the project's most significant refactors. Each behaviour became independently testable and tuneable.

Combat & Predation

Rock-paper-scissors damage mechanics where shape determines type and size determines magnitude.

Shape genes determine damage type. Size genes determine damage amount. A small creature with piercing attacks and a large creature with blunt attacks occupy completely different combat niches, even at similar overall power levels.

Three damage types interact against three defence types in a rock-paper-scissors triangle:

This creates evolutionary arms races. When shell-armoured herbivores dominate, blunt-attacking predators gain an advantage. As blunt predators spread, fur-defended prey outcompete shelled ones. The cycle continues indefinitely, preventing any single body plan from permanently winning.

The Plant Kingdom

Often overlooked in ecosystem sims. Not here.

Plants are fully simulated organisms with their own genomes, not just static food sources. They compete for light and space, reproduce, and evolve dispersal strategies. Six emergent methods arise from gene combinations:

Zoochory is the most interesting. Fruit-dispersed seeds are consumed along with the fruit, pass through the creature's digestive tract, and are deposited elsewhere; genuinely simulated endozoochory. Burr-dispersed seeds attach to creatures with fur and detach as they move, favouring long-distance transport by active animals.

Plants also evolve chemical defences (toxins, thorns) and creatures evolve corresponding resistances. This produces coevolutionary arms races independent of the predator-prey dynamics, a second axis of evolutionary pressure running in parallel.

Emergent Behaviour

All systems combine. Predator-prey oscillations. Speciation. Extinction events.

When all the systems run simultaneously, complex ecological dynamics emerge without being programmed. Predator and prey populations oscillate in classic Lotka-Volterra cycles. Populations isolated by mountain ranges diverge genetically, producing allopatric speciation. Sudden climate shifts trigger mass extinction events followed by adaptive radiation as survivors fill empty niches.

The simulation generates scientific names for emergent species based on their phenotype: diet, size, defence, and distinguishing features:

Ten creature archetypes recurrently evolve across different simulation runs, convergent evolution emerging from selection pressure alone:

The Paper

An early version of EcoSim was published at the Computer Graphics and Visual Computing (CGVC) conference as part of my BSc dissertation at Bangor University. The paper covers the foundational genetics engine, world generation, and emergent dynamics.

The current version is vastly more complex than what was published. The genetics system has grown from a handful of traits to 90 genes across 13 categories, the combat system, plant coevolution, scent communication, and classification systems were all added since. The paper represents where it started; everything above shows where it went.