Humans have physically reconfigured half of the world’s land to grow just eight staple crops: maize (maize), soybeans, wheat, rice, cassava, sorghum, sweet potato, and potato. They represent the vast majority of calories consumed worldwide. As the world’s population grows, there is pressure to increase production even more.
Many experts say the expansion of modern industrialized agriculture – which relies heavily on synthetic fertilizers, chemical pesticides and high-yielding seeds – is the wrong way to feed a growing world population. In their view, this approach is not ecologically or economically sustainable, and both farmers and scientists feel trapped in this system.
How can societies develop a food system that meets their needs and is also healthier and more diverse? It has proven difficult to generalize alternative methods, such as organic farming, as widely as industrial farming.
In a recent study, we looked at this problem from our perspective as computer scientists and agronomists. Together with our colleagues Bryan Runck, Adam Streed, Diane R. Wang and Patrick M. Ewing, we have proposed a way to rethink the design and implementation of agricultural systems, using a central idea of computer science – abstraction – which summarizes data and concepts and organizes IT, so that we can analyze and act on them without having to constantly examine their internal details.
Big yield, big impact
Modern agriculture intensified in just a few decades in the middle of the 20th century, a blink of an eye in the history of mankind. Technological improvements led the way, including the development of synthetic fertilizers and statistical methods that improved plant breeding.
These advances have allowed farms to produce much larger quantities of food, but at the expense of the environment. Large-scale agriculture has contributed to climate change, polluting lakes and bays with nutrient runoff, and accelerating species loss by turning natural landscapes into monoculture fields.
Many American farmers and agricultural researchers would like to grow a wider range of crops and use more sustainable farming methods. But it is difficult for them to determine which new systems might work well, especially in a changing climate. Low-impact farming systems often require deep local knowledge, as well as an encyclopedic understanding of plants, weather and climate modeling, geology, and more.
This is where our new approach comes in.
Farms as state spaces
When computer scientists think about complex problems, they often use a concept called state space. This approach mathematically represents all possible ways of configuring a system. Moving through space involves making choices, and those choices change the state of the system, for better or for worse.
Let’s take the example of a game of chess with a chessboard and two players. Each board configuration at any given time is a unique state of the game. When a player makes a move, they move the game to another state.
The entire game can be described by its “state space” – all the possible states the game could be in through valid moves made by players. During the game, each player searches for the states that suit him best.
We can think of an agricultural system as a state space within a particular ecosystem. A farm and its arrangement of plant species at any point in time represent a state in that state space. The farmer is looking for better states and trying to avoid bad ones.
Humans and nature move the farm from one state to another. Every day the farmer can do a dozen different things on the land, like plowing, planting, weeding, harvesting or adding fertilizer. Nature causes minor state transitions, such as plant growth and falling rain, and much more dramatic state transitions during natural disasters such as floods or forest fires.
Viewing an agricultural system as a state space expands choices for farmers beyond the limited options offered by today’s agricultural systems.
Individual farmers don’t have the time or the ability to do years of trial and error on their land. But a computer system can draw on agricultural knowledge from many different environments and schools of thought to play a metaphorical game of chess with nature that helps farmers identify the best options for their land.
Conventional farming limits farmers to a few choices of plant species, farming methods and inputs. Our framework allows for consideration of higher-level strategies, such as growing multiple crops together or finding the best management techniques for a particular plot of land. Users can search the state space to determine what combination of methods, species, and locations might achieve these goals.
For example, if a scientist wants to test five crop rotations – raising planned sequences of crops on the same fields – that each last four years, growing seven species of plants, that’s 721 potential rotations. Our approach could use information from long-term ecological research to help find the best potential systems to test.
One area where we see great potential is intercropping, i.e. growing different plants in a mixture or close together. It has long been known that many specific plant combinations grow well together, with each plant helping the others in one way or another.
The best-known example is that of the “three sisters” (corn, squash and beans) developed by indigenous farmers in the Americas. Cornstalks act as trellises for climbing bean vines, while squash leaves shade the soil, keeping it moist and preventing weeds from sprouting. Bacteria on the roots of bean plants provide nitrogen, an essential nutrient, to all three plants.
Crops throughout human history have had their own preferred intercropping systems with similar synergies, such as turmeric and mango or millet, cowpea and ziziphus, commonly known as red date. And new work on agrivoltaics shows that combining solar panels and agriculture can work surprisingly well: the panels partially shade the crops growing below, and farmers earn extra income by producing renewable energy. on their land.
Modeling of alternative agricultural strategies
We are working to turn our framework into software that people can use to model agriculture as state spaces. The aim is to allow users to consider alternative designs based on their intuition, minimizing the costly trial and error that is now required to test new ideas in agriculture.
Today’s approaches largely model and pursue optimizations of existing, often unsustainable farming systems. Our framework enables discovery of new farming systems and then optimization within those new systems.
Users will also be able to specify their goals to an AI-based agent that can search the farm state space, much like it might search the state space of a chessboard to choose the winning shots.
Modern societies have access to many more plant species and much more information about how different species and different environments interact than a century ago. In our view, farming systems are not doing enough to take advantage of all this knowledge. Combining it computationally could help make agriculture more productive, healthy and sustainable in a rapidly changing world.
Barath Raghavan is an Associate Professor of Computer Science and Electrical and Computer Engineering at the University of Southern California.
Michael Kantar is Associate Professor of Tropical Plants and Soil Science at the University of Hawaii.
This article is republished from The conversation under Creative Commons license. Read the original article.