Ireland, unlike most of the world, has a very sustainable diary farming sector as it based on a grass-fed or pasture-based system, where animals are roaming free in a field rather than being housed in sheds. Herds are rotated over the fields on the farm and
calving needs to be timed with natural cycle of grass growth throughout the year.
However, the success of this system depends on pasture management; on a weekly basis the dairy farmer needs to plan ahead,work out their grass budget (how much grass in is their fields to feed their herd) and make sure that fields, that are currently
recovering from being grazed, are progressing to be used in the next rotation.
If grass if not managed well, then the farmer will need to buy supplementary feeds (e.g., soya or corn), which are carbon intensive because of the transport they entail. They also impact the bottom line as they are more expensive than grass.
A key step in grass budgeting is estimating the growth rate for grass in the coming week. Predictions of grass growth in the coming weeks are provided by Teagasc for Irish Dairy Farmers.
Insight and the VistaMilk SFI research centres have collaborated used Artificial Intelligence models to improve grass growth prediction. There have developed a range of models that produce better and more accurate predictions to support dairy farmers.
However, these predictive models face a fundamental problem created by climate change. Their success is based on using historical data, but as climate changes, the weather in the past looks less and less like the weather we are currently experiencing.
So, the models become less accurate and less good at predicting the growth that might occur.
Insight’s Prof Mark Keane has been working with PhD students, Eoin Kenny and Mohammed Temraz, to try to solve these new problems created by climate change. They have developed completely-novel ground-breaking AI techniques that try to create artificial data (based on adapting historical data in very specific ways) that will allow the models to continue to perform well, even when climate-disruptive events occur.
These methods use a counterfactual techniques to do this task. Basically, their AI method poses the question: “if only the weather was more rainy or hot or both then we would expect this sot of grass growth to occur”.
They have tested this method against what happened in the drought of 2018, where there was a forage crisis in Ireland due to a very hot summer, that burnt off grass stocks on most Irish farms. They showed that using these artificial data-points the models could continue to accurately predict grass growth over the coming weeks.
These AI solutions are quite general, meaning that they should provide solutions for any prediction problem in which climate change is influencing the prediction of natural and human phenomena.