Neural networks: a computerized crystal ball
A new type of computer software called neural networks is helping to make better predictions about the behavior of agricultural systems.
These brain-like networks can generate computer simulations of biological systems.
Because they learn from past experiences, neural networks are a form of artificial intelligence. They can sort out the complex relationships in research data.
For example, a landowner may wish to know if a site is suitable for the release of an insect that, under favorable conditions, feeds on a certain weed and controls it.
Read Also
VIDEO: Green Lightning and Nytro Ag win sustainability innovation award
Nytro Ag Corp and Green Lightning recieved an innovation award at Ag in Motion 2025 for the Green Lightning Nitrogen Machine, which converts atmospheric nitrogen into a plant-usable form.
The insect is affected by such things as the direction the land faces, the slope and depressions in the land, the amount of bare ground, the amount of shade, the presence of certain plants and soil moisture conditions.
A neural network can be trained to give the answer on site suitability. Past experiences are presented to the network in numerical data and from these the network learns which combinations of factors result in successful releases. The trained neural network is then fed data on the release site in question and it predicts a result.
This technique has been used at the Agriculture Canada research centre in Lethbridge, Alta., to select release sites for the root-feeding beetle which controls leafy spurge.
An accuracy rate of 85 percent has been achieved in predicting successful releases. Without the neural network, costly unsuccessful releases occur about half the time.
Study pesticide dissipation
Neural networks are also being used to study pesticide dissipation under different weather scenarios. Time, temperature, amount and pattern of moisture events, dry periods and evaporation rates all affect pesticide degradation. More accurate predictions of pesticide residues will help prevent crop injury from the carryover of residues into the next crop year.
Neural networks are also being explored for their ability to predict beef carcass quality soon after calves enter a feedlot. Eventually, neural networks may be used to adjust diet to produce specific carcass types for specific markets and to predict seeding dates based on spring weather patterns. Such predictions will allow other scientists to better study global warming scenarios.
Neural networks are not bound by traditional laws of statistics and they don’t need an external expert to formulate a set of rules. They find their own rules in the data by adjusting the weightings given to the different input factors.
– Agriculture Canada
