simulations in agronomy have used “mechanistic models” for two to three decades. These allow the functioning of crops to be described by means of equations; these, for instance, represent the production of biomass as a function of solar radiation, or the growth of plants as a function of soil temperature or humidity. Simulations make it possible to account for the physiological and biological dynamics of plants, on the scale of a plot or a set of cultivated lands. They can represent different agricultural practices and help to assess their impact on yields.
An example of modeling applicable to biological phenomena? The interaction model between hosts and biological control auxiliaries. It describes how biological populations, such as predators and their prey (aphids and ladybirds for example), evolve in an ecosystem. For example, it explains the development cycles of many species (animal, plant, etc.) in many environments, such as the marine phytoplankton (Figure 1.7).
The model consists of two differential equations, proposed independently by two mathematicians, the Austrian Alfred-James Lotka (1880–1949) and the Italian Vito Volterra (1860–1940), in 1925–1926. These equations are written as:
The equations describe the evolution of the prey and predator population, represented by variables x(t) and y(t). The first equation states that prey, having access to an unlimited source of food, would grow exponentially (this is rendered by the first term in the right-hand side of the equation αx(t)), and are sometime faced with predators at certain occurrence (this is rendered by the second term of in the right hand side of the equation –βx(t)y(t)). The second equation states that predators perish from natural death (this is rendered by the first term in the right-hand side of the equation –α′x(t)), but grow by hunting prey (this is rendered by the second term of in the right-hand side of the equation +β′y(t)x(t)).
Figure 1.7. Satellite image showing algae growth in a North Atlantic region (source: www.nasa.gov). For a color version of this figure, see www.iste.co.uk/sigrist/simulation2.zip
COMMENT ON FIGURE 1.7.– In the waters of the North Atlantic, a large amount of phytoplankton, microscopic algae that play a role in the food chain of the marine ecosystem and contribute to the ocean carbon cycle, develops each spring and fall. The image is a photograph taken by NASA’s “Suomi” satellite on September 23, 2015. Blue spirals represent high concentrations of algae, waters loaded with microscopic creatures that contribute to the production of part of the planet’s oxygen.
The Lotka–Volterra equations have as unknown the populations of the competing species, the coefficients describe their survival and mortality rates. They predict a cyclical evolution of populations that is consistent with the observations (Figure 1.8). Many other equation-based models are available for studies of organic and agricultural systems. In recent years, data-based simulations have been developed to complement these models – nowadays, they use methods such as automatic learning techniques, discussed in Chapter 4 of the first volume. Statistical models are based on an adjustment of equations to data and allow an empirical relationship between different quantities to be established.
Figure 1.8. Typical evolution of the prey/predator populations as predicted by the Lotka-Volterra equations
Satellites, drones (Figure 1.9), sensors, field surveys: statistical models are based on a large number of data and it is the variety and diversity of the latter that gives credibility to predictions. Data used in statistical models are diverse: they describe the crop environment (e.g. topography and meteorology), farmers’ practices (e.g. frequency of watering or spreading), animal species behavior or plant species growth.
Figure 1.9. Agricultural drones are used to monitor crops and collect useful data to develop or validate certain simulations
(source: © Christophe Maitre/INRA/ www.mediatheque.inra.fr/ )
The models of each family complement each other, each providing information whose multimodel analysis makes it possible to identify trends [MAK 15]:
“Modeling is a synthetic expertise of the knowledge available to a community at a given time. One of the most promising approaches is to reconcile these classes of models with the statistical processing of simulation data. The efficiency and reliability of ‘Big-Data’ techniques is increasing, and it is not excluded that they may eventually replace models based on equations…”.
Simulations involve equations coupling different scales (plant, plot, farm or agricultural region). They require a large amount of data and still require very long computational times. Despite these current limitations, which artificial intelligence algorithms help to push back, modeling in agriculture is becoming more widespread and a tool for scientific debate and political decisions. Let us listen to the researchers involved in the development of models through a few examples.
1.3. Decision-making support
Farmers, political and economic decision makers, and consumers shape the landscape of agricultural practices to varying degrees, each acting at its own level: by guiding a continental agricultural policy, by deciding to invest in a new machine tool, or simply by doing one’s shopping. Different agents, actors and practices influence it and the behaviors of each have societal and environmental consequences.
Some agricultural practices have potentially negative impacts on climate and biodiversity, for example [BEL 19]. The disappearance of many insect species is attributed to the destruction of their habitat by intensive agriculture and their poisoning by the widespread use of pesticides [HAL 17, SAN 19]. How can these harmful effects be anticipated and limited? How can climate change be taken into account in current and future practices [BAS 14]? What is the best combination of agricultural practices that makes it possible to produce while guaranteeing a country’s food sovereignty? How can we legislate and cultivate for the benefit of the greatest number of people? Hélène Raynal, researcher at INRA and project manager of a digital platform dedicated to agrosystems [BER 13] provides an initial answer:
“Different models aggregated within a shared computer system make it possible to represent agricultural systems taking into account their complexity. Modeling should be able to represent biological processes, such as plant growth. They must also make it possible to account for physical processes related, for example, to water (evaporation of water from the soil, drainage of water to the deep layers of the soil, etc.), carbon and nitrogen, which are the factors that determine agricultural production. They must also integrate the climate dimension or farmers’ practices (such as irrigation levels), and socio-economic aspects. These processes fit into different scale levels and, depending on the issue, the model is used to simulate a cultivated field, an agricultural operation – or even an entire region. The aim is to integrate as much relevant information as possible into the model, from soil chemistry to local climate factors, agricultural practices and assumed market trends or climate changes. Simulations make it possible to play on various scenarios and estimate the impact of a decision taken by stakeholders in the sector on agricultural production. They can also help to design new agricultural systems adapted to different issues, such as climate change and the reduction of chemical inputs to the crop”.
This