Agriculture and food
Fields, herds, ponds and supply chains, run forward to estimate whether supply will meet demand.
People modelled farming on paper long before computers: rotation plans, yield tables and breeding records go back centuries. Digital models now couple weather, soil chemistry, disease spread and market shocks in a single run. They can forecast a harvest, herd growth or a shortage in advance. The same kind of model works from irrigation timing in one field to famine early warning for a whole region. It also reaches past the farm gate, to fermenting, cooking, chilling and tracking food.
Crop growth and yield models
A field of a single crop reduced to pools of leaf area, biomass and grain, driven day by day by weather, sowing date and management.
Question it answers: How much will this field yield, and when should it be planted?
Leaves out: Treats the field as spatially uniform, so variation inside it such as waterlogged or under-seeded patches is not represented.
Runs on: about 1010 FLOPS rough a laptop, seconds per field-seasonSince: 1965 by 1965, de Wit's early crop growth models in the Netherlands
Soil water and nutrient transport
Water infiltrating through soil layers and nutrients moving, binding and leaching, tracked cell by cell down the profile.
Question it answers: How much water and fertiliser reaches the roots, and how much washes away?
Leaves out: Soil is never as neatly layered as the model's grid assumes.
Runs on: about 1010 FLOPS rough a desktop machine for a season-long profile runSince: 1972 by 1972, early soil water balance models for irrigation research
Pest and plant disease spread
Insects, fungi or bacteria spreading across a crop or region, carried by wind, rain, animals or trade, against weather that helps or hinders them.
Question it answers: How far and how fast will an outbreak spread, and where should it be contained?
Leaves out: Assumes reported outbreaks are the true extent, which they rarely are.
Runs on: about 1011 FLOPS rough a university cluster for a regional, multi-season runSince: 1980 by 1980, computer models of wheat rust and potato blight spread
Irrigation scheduling
Soil moisture and crop water demand tracked forward day by day to decide when and how much to irrigate.
Question it answers: When does this field need water, and how much?
Leaves out: Assumes the pump, the canal and the weather forecast all behave.
Runs on: about 109 FLOPS rough runs on a farm computer or phone appSince: 1975 by 1975, water-balance scheduling models for irrigated farms
Soil erosion and land degradation
Rainfall, runoff and tillage stripping soil from slopes, tracked as sediment moving downhill and out of the field.
Question it answers: How much topsoil will this field lose, and which practices slow it?
Leaves out: Calibrated mostly on temperate farmland, so it travels badly to other climates.
Runs on: about 1010 FLOPS rough a desktop machine for a catchment-scale runSince: 1965 by 1965, the Universal Soil Loss Equation and its computer descendants
Crop breeding simulation
Virtual crosses between plant varieties, run forward through simulated generations to predict which combination of genes yields the best variety.
Question it answers: Which cross is worth making in the field, out of millions of possible ones?
Leaves out: A gene's effect on paper is not always what it does in a real paddock.
Runs on: about 1012 FLOPS rough a university cluster screening thousands of virtual crossesSince: 1990 by 1990, genetic simulation used in plant breeding programmes
Precision agriculture optimisation
A field split into small patches, each with its own soil and yield history, to work out the best seed, water and fertiliser rate for each one.
Question it answers: How much input does this exact patch need, not the whole field?
Leaves out: Needs good sensor data, which most farms still do not have everywhere.
Runs on: about 1010 FLOPS rough a farm laptop processing one field's sensor mapsSince: 1995 by 1995, yield-mapping and variable-rate application models
Multi-model crop ensembles
Many crop models run with the same weather and climate inputs so the spread shows the uncertainty in yield projections.
Question it answers: How uncertain is a yield projection under future climate?
Leaves out: Pests, extreme events and management change are handled unevenly.
Runs on: about 1012 FLOPS rough Dozens of models over global gridsSince: 2013 The Agricultural Model Intercomparison and Improvement Project, AgMIP (2013) source
Soil organic carbon turnover
Carbon in soil moves between pools that decay at different rates under temperature, moisture and inputs from plants.
Question it answers: Is a field gaining or losing soil carbon, and what changes that?
Leaves out: Pools are conceptual and cannot be measured directly.
Runs on: about 105 FLOPS rough monthly pool simulations on a laptop.Since: 1977 Jenkinson and Rayner, the turnover of soil organic matter in some of the Rothamsted classical experiments. source
Agricultural catchment pollution models
Rainfall, crops, soil and farm practice in a river basin are simulated to give runoff of water, sediment and nutrients.
Question it answers: How much nutrient or sediment reaches the river under different farming practice?
Leaves out: Calibration needs many data and field-level management is guessed.
Runs on: about 109 FLOPS rough basin-scale daily simulations on a workstation.Since: 1998 Arnold and colleagues, large area hydrologic modelling and assessment, model development. source
Global gridded crop model ensembles
Many crop models run on the world grid under the same climate scenarios, and their spread shows the uncertainty in projected yield.
Question it answers: How much will world crop yield change under climate scenarios, and how uncertain is that?
Leaves out: Models share missing processes such as pests and extreme heat.
Runs on: about 1013 FLOPS rough many crop models run globally for decades on a cluster.Since: 2013 Rosenzweig and colleagues, the Agricultural Model Intercomparison and Improvement Project protocols. source
Farm agent-based models of land use and adoption
Farm households with their own land, labour and neighbours decide what to grow and which technology to adopt, with spatial spread of innovations.
Question it answers: How will farmers respond to a price, subsidy or technology?
Leaves out: Farmer decisions are modelled by simplified rules.
Runs on: about 108 FLOPS rough regional agent-based runs on a workstation.Since: 2001 Berger, agent-based spatial models applied to agriculture for technology diffusion and resource use. source
Pesticide resistance management models
A pest population under spraying evolves resistance, and strategies such as refuges, rotation and mixtures are compared.
Question it answers: Which spraying strategy delays resistance the longest?
Leaves out: Costs of resistance and pest movement are uncertain.
Runs on: about 106 FLOPS rough population genetic models on a laptop.Since: 1977 Comins, the management of pesticide resistance. source
Agrivoltaic light-sharing models
Solar panels above crops cast moving shade on them, so the layout, light reaching the leaves and crop growth are simulated together with the electricity produced.
Question it answers: How much food and energy can the same field give when both share the sunlight?
Leaves out: Crop response to partial shade is known for only a few species, and water and temperature effects under the panels are simplified.
Runs on: about 109 FLOPS rough hours on a workstation per layout and season.Since: 2011 Dupraz and colleagues' land-equivalent-ratio model of photovoltaic panels combined with food crops. source
Plant architecture growth models
A plant built up bud by bud from botanical rules for branching and ageing, giving the three-dimensional shape of a tree or crop as it grows.
Question it answers: How does the branching habit of a plant decide how it fills space and catches light?
Leaves out: Growth rules are fitted to one species and environment, and the response to wind, neighbours and damage is limited.
Runs on: about 108 FLOPS rough seconds to minutes per plant.Since: 1988 de Reffye and colleagues' plant models faithful to botanical structure and development. source
Degree-day pest phenology models
Heat accumulated each day above a threshold sets when insects emerge, breed and reach a stage where they harm a crop, so sprays can be timed.
Question it answers: On which day will the pest reach the stage at which control works?
Leaves out: Insects respond to temperature in a curve, not a straight line, and microclimate on the plant differs from the weather station.
Runs on: about 104 FLOPS rough trivial on a laptop; run daily for many sites.Since: 1976 Allen's modified sine-wave method for calculating degree days for insects. source
Sterile insect release models
Mass-reared sterile males released into a wild population mate with females that then produce no offspring, and the numbers are followed as they fall.
Question it answers: How many sterile insects must be released, and for how long, to wipe out a pest?
Leaves out: Wild females may refuse the sterile males, and insects arriving from outside can undo the release.
Runs on: about 105 FLOPS rough seconds to minutes on a laptop.Since: 1955 Knipling's analysis of insect control or eradication using sexually sterile males. source
Land-use change agent models
Landowners and farmers, each with their own goals and information, decide what to do with each parcel, and the land cover that results is mapped over decades.
Question it answers: How will land cover change under a policy, a road or a crop price shift?
Leaves out: Human decision rules are hard to measure, so they are often guessed, and the models are tuned to the past without evidence they continue.
Runs on: about 108 FLOPS rough hours on a workstation.Since: 2003 Parker and colleagues' review of multi-agent systems for simulating land-use and land-cover change. source
Greenhouse climate models
Heat, humidity and carbon dioxide balances of a greenhouse, with sun, vents, heating and the crop's own transpiration, to design and control the indoor climate.
Question it answers: How should vents and heating be run to keep a crop warm and dry at least cost?
Leaves out: Air is treated as well mixed, though real greenhouses have gradients, and crop feedback is simplified.
Runs on: about 107 FLOPS rough seconds per day simulated, so a season runs in minutes.Since: 1985 Garzoli's simple greenhouse climate model (Acta Horticulturae). source
Livestock herd and feed models
A herd's growth, milk or meat output tracked against feed intake, breed and management, animal by animal or as a whole group.
Question it answers: How much will this herd produce, and how much feed does it need to get there?
Leaves out: Day-to-day individual variation in animal health and condition is not captured by feed-based models.
Runs on: about 109 FLOPS rough runs on a farm computer per herdSince: 1970 by 1970, early beef and dairy production models
Grazing and pasture management
Grass growth and regrowth under grazing pressure, weather and soil fertility, used to plan how many animals a pasture can carry.
Question it answers: How many animals can this pasture support without wearing it out?
Leaves out: Pasture response to an early or unusually severe drought departs from the modelled growth curve.
Runs on: about 109 FLOPS rough a laptop run per paddock per seasonSince: 1973 by 1973, pasture growth models for grazing systems
Aquaculture and fish farming
Fish or shellfish raised in ponds, cages or tanks, tracked against feed, oxygen, temperature and stocking density.
Question it answers: How much stock should go in, and when should it come out, for the best yield?
Leaves out: Disease outbreaks in a real pond rarely coincide with the warning the model provides.
Runs on: about 109 FLOPS rough a laptop run per farm cycleSince: 1985 by 1985, bioenergetic growth models for farmed fish
Livestock disease outbreak models
A farm animal disease spreading between farms by animal transport, vehicles and wind, used to plan culling, movement bans and vaccination rings.
Question it answers: How far will this outbreak spread, and which control measure stops it fastest?
Leaves out: Farmers do not always report symptoms as fast as the model assumes.
Runs on: about 1011 FLOPS rough a university cluster for a national outbreak runSince: 2001 the UK's 2001 foot-and-mouth epidemic models
Livestock breeding and genetics
Virtual matings between animals, run forward through generations to predict milk yield, growth rate or disease resistance in the offspring.
Question it answers: Which sire or dam should be bred next, out of a whole breeding population?
Leaves out: Assumes the recorded pedigree is accurate, although errors in record-keeping and misattributed parentage occur in practice.
Runs on: about 1011 FLOPS rough a cluster run scoring a national breeding populationSince: 1990 by 1990, BLUP-based genetic evaluation of dairy cattle
Precision livestock monitoring
Individual animals tracked by sensor for movement, feeding and temperature, compared against a simulated healthy baseline to flag problems early.
Question it answers: Which individual animal, out of thousands, needs attention today?
Leaves out: A sensor glitch looks exactly like a sick animal to the model.
Runs on: about 1010 FLOPS rough a farm server processing one barn's sensor feedSince: 2005 by 2005, sensor-driven dairy cattle monitoring systems
Aquaculture pond bioenergetics
Each fish in a pond has energy from feed and oxygen turned into growth, and the water's oxygen and waste are tracked alongside.
Question it answers: How many fish, fed how much, grow fastest before the pond water goes bad?
Leaves out: Disease, behaviour and weather shocks are mostly left out, so the model suits steady conditions best.
Runs on: about 108 FLOPS rough minutes on a laptop.Since: 1985 Cuenco and colleagues' individual-fish bioenergetics and growth model for aquaculture ponds. source
Food supply chain simulation
Grain, produce and processed food moving from farm to warehouse to shop, tracked against transport, storage losses and demand shocks.
Question it answers: Where does the food system break first under a shortage or a shock?
Leaves out: Real supply chains hide informal trade the model never sees.
Runs on: about 1010 FLOPS rough a cluster run for a national or regional networkSince: 1985 by 1985, logistics simulation applied to food distribution
Famine early warning models
Rainfall, harvests, prices and conflict combined to flag which regions are heading towards food shortage months before it hits.
Question it answers: Which regions will run short of food, and when?
Leaves out: Depends on data from the places least able to report it.
Runs on: about 1010 FLOPS rough a cluster run scoring many regions each monthSince: 1985 by 1985, government-led famine early warning systems in the Sahel
Fermentation process modelling
Yeast, bacteria or mould growing inside a vat of dough, must or mash, tracked as they consume sugars and produce alcohol, acid or gas.
Question it answers: How long should this ferment, and at what temperature, for the result wanted?
Leaves out: Wild fermentations bring in microbes the model was never told about.
Runs on: about 109 FLOPS rough a laptop run per batch processSince: 1975 by 1975, kinetic models of brewing and baking fermentation
Baking and thermal food processing
Heat moving into dough, cans or packaged food, tracked to predict cooking, browning and how safely microbes are killed off.
Question it answers: Is the centre of this product cooked through, and is it safe to eat?
Leaves out: Ovens in the real world are never as even as the model's heat field.
Runs on: about 109 FLOPS rough a laptop run per product designSince: 1970 by 1970, heat transfer models for canning and baking
Cold chain and spoilage modelling
Temperature drift in trucks, warehouses and shop fridges tracked against microbial growth to predict how much shelf life is lost.
Question it answers: How much of this shipment will spoil before it reaches the shelf?
Leaves out: Brief, severe temperature excursions, such as a door left open for ten minutes, cause more damage than the model averages suggest.
Runs on: about 109 FLOPS rough a laptop run tracking one shipment's temperature logSince: 1990 by 1990, shelf-life prediction models for chilled food logistics
Food safety and traceability simulation
Contamination introduced at one point in the food system, traced forward through mixing, processing and distribution to see who ends up exposed.
Question it answers: If contamination starts here, how far does it spread before recall catches it?
Leaves out: Paper trails in the real supply chain have more gaps than the model assumes.
Runs on: about 109 FLOPS rough a laptop run per contamination scenarioSince: 1995 by 1995, traceability simulation following major contamination outbreaks
Predictive food microbiology
The growth and death of bacteria in food are modelled against temperature, acidity and water activity to estimate shelf life and safety.
Question it answers: How fast will a pathogen grow in this food at this temperature?
Leaves out: Food is more varied than a laboratory broth.
Runs on: about 105 FLOPS rough growth curve fitting on a laptop.Since: 1994 Baranyi and Roberts, a dynamic approach to predicting bacterial growth in food. source
Quantitative microbial risk assessment
Contamination, cooking, storage and eating are chained as probability distributions to estimate the chance of illness from a food.
Question it answers: How likely is a person to become ill from this food, and which step matters most?
Leaves out: Dose-response is poorly known for rare pathogens.
Runs on: about 108 FLOPS rough Monte Carlo over a farm-to-fork chain on a laptop.Since: 1998 Cassin and colleagues, quantitative risk assessment for E. coli O157:H7 in ground beef hamburgers. source
Digestion and gastric flow simulation
Stomach wall movement drives mixing and breakdown of food, and the flow and acid are computed to relate food structure to nutrient release.
Question it answers: How quickly is a food broken down and absorbed in the stomach?
Leaves out: Stomach wall motion and food rheology are simplified.
Runs on: about 1012 FLOPS rough moving-wall flow simulation on a cluster.Since: 2010 Kozu and colleagues, analysis of flow phenomena in gastric contents induced by human gastric peristalsis. source