Ecology, evolution and animal behaviour

Ant colony foraging

Individual ants laying and following pheromone trails, with no leader and no map.

Question it answers: How does a colony find the shortest path to food when no ant knows the way?

Leaves out: Real ants also use sight, sound and taste, which most trail models ignore.

Runs on: about 10¹¹ FLOPS rough a desktop workstation simulating thousands of foraging antsSince: 1990 Deneubourg and colleagues' model of Argentine ant trail formation

Run the interactive model

Honeybee colonies

Foragers, nurses and scouts sharing work, and swarms choosing a new home by a kind of vote.

Question it answers: How does a swarm choose the best nest site without a leader?

Leaves out: Assumes bees follow fixed simple rules, though real bees vary and can be wrong.

Runs on: about 10¹¹ FLOPS rough a desktop workstation simulating a swarm's nest-site choiceSince: 2003 Passino and Seeley's model of honeybee nest-site selection

Flocking and schooling

Each bird or fish steering by three local rules: separation, alignment with neighbours, and cohesion.

Question it answers: How do thousands of animals turn together as one?

Leaves out: Real flocks respond to predators and terrain the three simple rules leave out.

Runs on: about 10¹³ FLOPS rough a gaming GPU simulating thousands of steered agentsSince: 1987 Craig Reynolds's boids model of flocking

Run the interactive model

Termite mound building

Termites dropping and moving soil pellets, and the air flowing through the structure they build.

Question it answers: How do insects with tiny brains build ventilated towers?

Leaves out: Real termite behaviour depends on chemical and structural cues only partly captured.

Runs on: about 10¹² FLOPS rough a university cluster simulating pellet-dropping termites and airflowSince: 1999 Bonabeau and Theraulaz's stigmergic models of termite construction

Slime mould aggregation

Free-living amoebae releasing and following a chemical signal to stream together into a single slug-like body when food runs out.

Question it answers: How do independent cells with no nervous system agree to become one body?

Leaves out: Lab conditions are far cleaner than the soil the real organism lives in.

Runs on: about 10¹² FLOPS rough a desktop workstation running a chemotaxis aggregation modelSince: 1970 Keller and Segel's model of slime mould chemotactic aggregation

Bacterial swarming and quorum sensing

Millions of bacteria sensing each other's chemical signals to switch, together, into a coordinated behaviour such as swarming across a surface or lighting up.

Question it answers: How does a bacterial population know when there are enough of it to act together?

Leaves out: Lab strains grown in a dish behave more uniformly than a real mixed microbial community.

Runs on: about 10¹² FLOPS rough a desktop workstation running an agent-based colony modelSince: 1994 early simulations of quorum-sensing circuits coordinating bacterial behaviour

Locust marching bands

Individual insects align with neighbours at high density, and bands begin to march together in a common direction.

Question it answers: At what density do locusts switch from random wandering to marching?

Leaves out: Habitat structure and food are not included, and real swarms involve flight.

Runs on: about 10⁷ FLOPS rough self-propelled particle simulation on a workstation.Since: 2006 Buhl and colleagues, from disorder to order in marching locusts. source

Movement and foraging

Animals searching a landscape for food, modelled as random walks and decision rules.

Question it answers: Is this animal searching efficiently, and what does it know about where food is?

Leaves out: Fitting a walk statistic does not establish what the animal is actually sensing.

Runs on: about 10¹¹ FLOPS rough a desktop workstation fitting a random-walk foraging modelSince: 1996 Viswanathan and colleagues' Levy flight model of albatross foraging

Migration and navigation

Birds, turtles or whales travelling thousands of kilometres using sun, stars, smell or the magnetic field.

Question it answers: Which cues could get a first-year bird to the right continent?

Leaves out: Which cues real animals rely on, and how they combine them, is still debated.

Runs on: about 10¹¹ FLOPS rough a desktop workstation simulating cue-based bird navigationSince: 1996 early simulations of magnetic and celestial cue use in bird navigation

Animal locomotion

Fish swimming, insects flying and animals running, with the fluid or ground pushing back.

Question it answers: How does a bumblebee fly, and why do fish swim in formation?

Leaves out: Simplified wing and body shapes miss details a real animal's surfaces provide.

Runs on: about 10¹⁴ FLOPS rough a university cluster running a fluid-body flight or swimming simulationSince: 1994 early fluid dynamics models of insect flight

Evolution of behaviour

Strategies such as fight, share or cheat, played against each other over many generations.

Question it answers: Why would an animal ever help an unrelated neighbour?

Leaves out: Strategies come from a menu the modeller picked, not the full space of behaviour.

Runs on: about 10¹¹ FLOPS rough a desktop workstation running an evolutionary game tournamentSince: 1981 Axelrod and Hamilton's simulated evolution of cooperation strategies

Birdsong and vocal learning models

A young bird's brain circuit for copying a tutor's song, modelled as a sensorimotor loop that compares its own babbling to a memorised template.

Question it answers: How does a young bird turn babbling into an accurate copy of an adult's song?

Leaves out: Models built to explain songbirds do not obviously generalise to vocalisation in other animals.

Runs on: about 10¹² FLOPS rough a desktop workstation training a sensorimotor learning modelSince: 2002 early computational models of song learning in the zebra finch

Dynamic energy budget modelling

One animal's whole energy budget followed over its life: what it eats, and how it splits between staying alive, growing and reproducing.

Question it answers: How will this animal grow and reproduce under a given food supply and temperature?

Leaves out: A small set of universal parameters cannot capture the specific features of every species.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation solving a small set of energy-flow equationsSince: 1986 Kooijman's dynamic energy budget theory for individual organisms

Predator–prey dynamics

Numbers of hunters and hunted rising and falling in linked cycles, the Lotka–Volterra equations and their descendants.

Question it answers: Why do lynx and hare populations oscillate cyclically?

Leaves out: Real populations have age structure, space and multiple prey the model ignores.

Runs on: about 10⁹ FLOPS rough a laptop solving the Lotka-Volterra equationsSince: 1925 Lotka and Volterra's predator-prey equations

Run the interactive model

Food webs

Who eats whom across a whole ecosystem, and how energy moves through it.

Question it answers: What happens to the rest of the system if one species disappears?

Leaves out: Real feeding links are harder to measure than the model networks assume.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation analysing a model food web's stabilitySince: 1972 Robert May's stability analysis of model food webs

Forest dynamics

Individual trees germinating, competing for light and dying across centuries.

Question it answers: What will this forest look like in two hundred years?

Leaves out: How a tree species will respond to a changing climate is still hard to calibrate.

Runs on: about 10¹³ FLOPS rough a university cluster tracking individual trees over centuriesSince: 1970 Botkin, Janak and Wallis's JABOWA gap model of forest succession

Fisheries and wildlife management

Fish or animal stocks under harvest, with birth, growth and uncertainty.

Question it answers: How many can be caught each year without the population collapsing?

Leaves out: Stock assessments depend on catch data that can be incomplete or misreported.

Runs on: about 10¹¹ FLOPS rough a desktop workstation running a stock assessment modelSince: 1957 Beverton and Holt's dynamic model of exploited fish populations

Species ranges

Where species can live, and how their ranges shift as climate changes.

Question it answers: Where will this species be able to survive in fifty years?

Leaves out: Assumes climate is the main limit on range, understating competition and dispersal.

Runs on: about 10¹¹ FLOPS rough a desktop workstation fitting a bioclimatic range modelSince: 1986 early bioclimatic envelope models predicting species distributions

Wildlife disease

Infection moving through animal populations, and sometimes spilling over into people.

Question it answers: Could this disease wipe out the population, and can vaccination bait stop it?

Leaves out: Contact rates between wild animals are hard to observe directly.

Runs on: about 10¹² FLOPS rough a university cluster simulating disease spread through a wild populationSince: 1991 early simulation models of rabies spread through fox populations

General ecosystem models

A mechanistic model in which every organism type, from plankton to large animals, eats, grows, reproduces and disperses by shared rules across the globe.

Question it answers: Can global patterns of ecosystem structure emerge from rules for individual organisms?

Leaves out: Individual species are not tracked, only functional groups.

Runs on: about 10¹⁴ FLOPS rough Cohort-based global runs for many simulated yearsSince: 2014 Emergent global patterns of ecosystem structure and function from a mechanistic general ecosystem model (2014) source

Metapopulation dynamics

A species lives in many habitat patches in a landscape, going extinct in some and recolonising others, and the model asks whether the whole persists.

Question it answers: Will a species survive in a fragmented landscape?

Leaves out: Dispersal is represented by a simple kernel and local habitat quality is held fixed.

Runs on: about 10⁸ FLOPS rough patch occupancy simulation on a laptop.Since: 1994 Hanski, a practical model of metapopulation dynamics. source

Individual-based ecological models

Each plant or animal is represented separately with its own state, so that population patterns arise from many individual lives.

Question it answers: What population patterns arise from the interactions of individuals?

Leaves out: Parameters for each individual are hard to measure and validation is difficult.

Runs on: about 10⁹ FLOPS rough individual-based simulation of a landscape on a workstation.Since: 1988 Huston, DeAngelis and Post, new computer models unify ecological theory. source

Coral reef resilience models

Corals, algae and grazing fish compete on a reef through growth, grazing and disturbance, and the model asks whether a reef can flip from coral to algae.

Question it answers: How much grazing keeps a reef from tipping to algae?

Leaves out: Larval supply and the effects of warming are modelled in simplified ways.

Runs on: about 10⁸ FLOPS rough spatial reef simulations on a workstation.Since: 2007 Mumby, Hastings and Edwards, thresholds and the resilience of Caribbean coral reefs. source

Spatial host-parasitoid and pest lattices

Hosts and parasitoids live on a grid with local movement, and complex spiral and chaotic patterns emerge that stabilise otherwise unstable interactions.

Question it answers: Can space let predators and prey coexist when they could not otherwise?

Leaves out: Real landscapes are not uniform grids.

Runs on: about 10⁶ FLOPS rough coupled-map lattices on a laptop.Since: 1991 Hassell, Comins and May, spatial structure and chaos in insect population dynamics. source

Fisheries management strategy evaluation

A simulated fish stock and its fishery are run under different harvest rules, with noisy data and uncertainty fed back into each rule, to find rules that work despite error.

Question it answers: Which harvest rule keeps a stock healthy when we do not know its size?

Leaves out: Operating models rest on assumptions about stock biology and fishers that may be wrong.

Runs on: about 10⁹ FLOPS rough thousands of simulated management projections on a cluster.Since: 1999 Butterworth and Punt, experiences in evaluating and implementing management procedures. source

Trophic mass-balance ecosystem models

A marine food web is described as balanced flows of biomass between groups, then run through time to see the effect of fishing or changes of predators.

Question it answers: What happens to the whole ecosystem when one species is fished harder?

Leaves out: Groups are lumped and diet data are patchy.

Runs on: about 10⁷ FLOPS rough balance and time-dynamic runs on a laptop.Since: 1997 Walters, Christensen and Pauly, structuring dynamic models of exploited ecosystems from trophic mass-balance assessments. source

Plankton ecosystem models

Nutrients, phytoplankton, zooplankton and detritus in the ocean surface layer are linked by growth, grazing and sinking.

Question it answers: How does plankton production respond to light, nutrient and mixing?

Leaves out: Many species are lumped into a few groups and parameters differ by region.

Runs on: about 10¹⁰ FLOPS rough coupled to ocean circulation models on a cluster.Since: 1990 Fasham, Ducklow and McKelvie, a nitrogen-based model of plankton dynamics in the oceanic mixed layer. source

Lake eutrophication models

Nutrients, algae, plants and fish in a shallow lake are modelled to ask when clear water flips to murky and how it can be restored.

Question it answers: How much nutrient can a lake take before turning murky, and will restoration reverse it?

Leaves out: Each lake needs its own calibration and sediment dynamics are uncertain.

Runs on: about 10⁷ FLOPS rough ordinary differential equations on a laptop.Since: 1995 Janse and Van Liere, PCLake, a modelling tool for evaluating lake restoration scenarios. source

Marine larval dispersal and connectivity

Virtual larvae are released from reefs and carried by modelled ocean currents, building a map of which populations supply which others.

Question it answers: Which reefs seed which, and how large should a protected area be?

Leaves out: Larval behaviour and mortality are poorly observed.

Runs on: about 10¹² FLOPS rough particle tracking in ocean model output on a cluster.Since: 2006 Cowen and colleagues, scaling of connectivity in marine populations. source

Honeybee colony and landscape models

A whole bee colony's brood, workers and stores simulated day by day, with foragers drawing from flowers across a landscape and facing mites, pesticide and weather.

Question it answers: Will this colony survive the winter in this landscape?

Leaves out: The landscape's flower supply and the bees' decisions are simplified, and queen failure and disease are only crudely included.

Runs on: about 10⁹ FLOPS rough minutes per colony-year on a laptop.Since: 2014 Becher and colleagues' BEEHAVE model of honeybee colony dynamics and foraging. source

Pollination service across landscapes

Wild bee nesting sites and flower resources are mapped across a landscape, and each crop field is given an estimate of the pollinators visiting it.

Question it answers: How much crop pollination does the surrounding habitat provide, and what if it is removed?

Leaves out: Bee foraging is reduced to distance decay from nests, and bee abundance is a weak guide to actual crop yield.

Runs on: about 10⁸ FLOPS rough seconds to minutes per region on a laptop.Since: 2009 Lonsdorf and colleagues' model of pollination services across agricultural landscapes. source

Wind seed and pollen dispersal models

Seeds and pollen released into turbulent air are carried downwind and fall at random, building up a distribution of where they land.

Question it answers: How far do the seeds of a plant travel, and how fast can it spread?

Leaves out: Rare long jumps matter most to spread but are the hardest to measure, so the tail of the distribution is poorly known.

Runs on: about 10⁸ FLOPS rough minutes on a laptop.Since: 1989 Okubo and Levin's theoretical framework for data analysis of wind dispersal of seeds and pollen. source

Near-term ecological forecasting

An ecosystem model is run forward days to years and updated each time new observations arrive, as weather forecasting is done.

Question it answers: Can the state of a lake, forest or fishery be forecast next season, with honest error bars?

Leaves out: Ecological data are sparse compared with weather data, and processes are less well known, so forecasts often only just beat a simple baseline.

Runs on: about 10¹² FLOPS rough ensembles of thousands of runs refreshed on a cluster.Since: 2018 Dietze and colleagues' framework for iterative near-term ecological forecasting. source

Size-spectrum ecosystem models

All organisms in an ecosystem grouped by body size, with energy moving from small to large as they eat and are eaten, and growth and mortality set by size alone.

Question it answers: How does fishing pressure reshape the balance of small and large fish in the sea?

Leaves out: Species identity is ignored, so predators with the same size but different diets are treated as the same.

Runs on: about 10⁹ FLOPS rough minutes to hours on a workstation.Since: 1978 Silvert and Platt's time-dependent equation for energy flux through the pelagic ecosystem by size. source

Neutral community models

A community of species treated as identical in birth, death and dispersal, with only chance and migration deciding which are common.

Question it answers: How much of the pattern of species abundance needs no differences among them at all?

Leaves out: Real species clearly differ, so the model works as a null baseline rather than a description of any one community.

Runs on: about 10⁹ FLOPS rough minutes on a laptop.Since: 1976 Caswell's neutral model analysis of community structure. source

Population genetics

Gene variants drifting, mutating and being selected through generations.

Question it answers: How much of the variation we see is chance, and how much is selection?

Leaves out: Real populations violate the simplifying assumptions of random mating and constant size.

Runs on: about 10¹² FLOPS rough a university cluster simulating drift and selection across a genomeSince: 1964 early computer simulations of genetic drift under the neutral theory

Coalescent and phylogenetic simulation

Family trees of genes or species, simulated backwards in time to test methods for reading real DNA.

Question it answers: Can the method recover the true history from the DNA?

Leaves out: Assumes a simple population history; real demography is more complex.

Runs on: about 10¹¹ FLOPS rough a desktop workstation simulating gene trees backward in timeSince: 1983 Hudson's early coalescent simulation software for population genetics

Origin of life

Simple chemicals forming networks that copy themselves and compete.

Question it answers: How could chemistry cross the line into biology?

Leaves out: No simulation has yet produced anything that would count as alive.

Runs on: about 10¹² FLOPS rough a desktop workstation simulating self-replicating chemical networksSince: 1971 Eigen and Schuster's hypercycle models of self-replicating chemistry

Host-parasite coevolution

Hosts developing defences and parasites evolving ways around them, each generation's simulated population responding to the other in a modelled arms race with no fixed winner.

Question it answers: Does a host-parasite arms race reach equilibrium, or does it continue indefinitely with neither side gaining a lasting advantage?

Leaves out: Real hosts and parasites coevolve with everything else in their environment at the same time.

Runs on: about 10¹¹ FLOPS rough a desktop workstation running a multi-generation coevolution modelSince: 1990 early simulations testing the Red Queen hypothesis of coevolution

Speciation and macroevolution models

Lineages splitting, persisting and going extinct across millions of simulated years, testing what patterns of diversity and family-tree shape simple rules would produce.

Question it answers: Could the pattern of species we see have come from simple birth and death of lineages alone?

Leaves out: Fossil and genetic records are patchy, so many different rules can fit the same pattern.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation simulating birth-death branching processesSince: 1969 early birth-death models of diversification used to study evolutionary trees

Sexual selection and mate choice evolution

A simulated population in which an ornament and a preference for it evolve together, testing whether female choice alone can run away to an extreme.

Question it answers: Can mate choice alone drive a trait to an extravagant, costly extreme?

Leaves out: Assumes preference and trait are simply heritable, which is simpler than most real mate choice.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation running a genetic model of coupled traitsSince: 1981 Lande's population-genetic model of runaway sexual selection

Adaptive dynamics of coevolution

A trait evolves by mutations of small effect that are tested against the current population, leading to evolutionary stable states or splitting into branches.

Question it answers: Where does evolution settle, and when does a lineage split?

Leaves out: Assumes rare mutations and simple trait spaces.

Runs on: about 10⁶ FLOPS rough analysis and simulation on a laptop.Since: 1996 Dieckmann and Law, the dynamical theory of coevolution derived from stochastic ecological processes. source

Gene drive population models

A genetic element that spreads faster than ordinary inheritance is released into a modelled population to ask whether it could suppress or alter a pest species.

Question it answers: Would a gene drive spread through a wild population, and could resistance stop it?

Leaves out: Real dispersal and the evolution of resistance are poorly known before a release.

Runs on: about 10⁸ FLOPS rough spatial population simulations with genetics on a workstation.Since: 2003 Burt, site-specific selfish genes as tools for the control and genetic engineering of natural populations. source

Forward-time genome simulation

Every individual and its genome is followed through generations, with selection, linkage and varied demography, to see how real DNA patterns come about.

Question it answers: What do selection and population history leave behind in the genomes of a species?

Leaves out: Population sizes are scaled down to keep the run feasible, which distorts the balance of selection and drift.

Runs on: about 10¹⁴ FLOPS rough days on a cluster for a large population over many generations.Since: 2013 Messer's SLiM, a forward-time simulator of evolution with selection and linkage. source

Animal contest strategies

Individuals meeting at random to contest a resource choose among fixed strategies, and the mix settles at one no mutant can improve on.

Question it answers: When is fighting or backing down the better policy, given what others do?

Leaves out: Animals are reduced to a short list of strategies with no learning, memory or individual differences.

Runs on: about 10⁶ FLOPS rough seconds on a laptop.Since: 1973 Maynard Smith and Price's 'The logic of animal conflict'. source

Population viability analysis

A small, threatened population's births, deaths and catastrophes rolled forward many times to estimate the chance it survives, and for how long.

Question it answers: How likely is this population to go extinct in the next hundred years?

Leaves out: Small samples of a rare species leave the input numbers themselves highly uncertain.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation running thousands of stochastic trajectoriesSince: 1980 early stochastic models estimating extinction risk for endangered species

Reserve and corridor design

Candidate sites and the links between them scored against cost and the species they would protect, searched for the smallest network of reserves that covers the most.

Question it answers: Which set of land parcels protects the most species for the least cost?

Leaves out: Optimises for the species data that exist, which is a small fraction of what actually lives there.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation solving a reserve-selection optimisationSince: 1988 early systematic reserve-selection algorithms for conservation planning

Reintroduction and translocation planning

A small founder group released into new or former habitat, followed forward to see whether it establishes, given the area's carrying capacity and the founders' genetics.

Question it answers: Will releasing these animals here actually establish a lasting population?

Leaves out: Habitat quality and disease risk at the release site are often estimated roughly rather than measured.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation running a founder-population modelSince: 1990 early simulation models used to plan the reintroduction of a large mammal

Invasive species spread

An introduced species spreading across a landscape cell by cell, and the effect of barriers, culling or biological control on how far it gets.

Question it answers: How fast will this invader spread, and where should control effort go first?

Leaves out: New invaders often spread in ways no past invasion prepared the model for.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation running a landscape spread modelSince: 1951 Skellam's diffusion model of a spreading population

Poaching and anti-poaching patrol games

Rangers and poachers as players in a repeated game over a map of a protected area, used to decide where patrols should go when they cannot be everywhere.

Question it answers: Where should limited rangers patrol to do the most against a strategic poacher?

Leaves out: Assumes a rational, patient poacher, when much of it is opportunistic.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation solving a security game over a patrol mapSince: 2014 early security-game models applied to wildlife ranger patrol scheduling

Rewilding and trophic cascade modelling

A predator or other keystone species added back to a food web, followed through its knock-on effects on prey, vegetation and even rivers.

Question it answers: What happens to the rest of the ecosystem if we bring this species back?

Leaves out: The famous cascades took decades to unfold and are easier to model after the fact than to predict.

Runs on: about 10¹⁰ FLOPS rough a desktop workstation running a multi-species food-web modelSince: 1995 wolves reintroduced to Yellowstone, later modelled for cascading effects