How AI Can Help Agribusiness Address Climate Change
By Arun Ravindran, Hamid Maher, and Rata Jacquemart (alumnus)
Agriculture is at the forefront of climate change. As temperatures, humidity, and rainfall patterns shift, agricultural businesses around the globe — from family farms to multinational enterprises — will be tremendously affected, whether through soil chemistry, insect migration, or other factors threatening crop quality and yields. (See Exhibit 1.)
To prepare, agribusinesses need to understand the potential impact on their land and crops. For every agricultural enterprise, from farmers to vineyards to producers of seeds, animal feed, or biofuels, knowing the potential changes in yield coming down the road — along with the drivers of those changes — could allow them to optimize their agricultural practices, making them more adaptable and resilient. In addition, knowing the drivers of crop quality would help them improve profits and increase their ability to anticipate the need for crop conversion.
Small farms are particularly at risk of losing their livelihood from climate change; many governments are therefore supporting and leading adaptation and resilience efforts to mitigate the impact on small farms.
Yet analyzing agriculture in the context of climate change is no trivial pursuit, given the geospatial spread of the change, the variety of physical properties involved, the number of climate determinants, the lengthy time horizon, and the variance in observable quantities of rainfall, solar radiation, and other factors. Weather forecasts alone are difficult enough to get right, much less the breadth and depth of widespread climate shifts.
Creating a yield model through AI
Despite the complexity they face, agriculture businesses need to put a stake in the ground, developing an informed perspective on, and a degree of confidence about, what’s to come, and taking their first steps towards understanding and reacting to this complex problem.
One large agriculture business with crops spread over many different areas decided to address this problem sooner rather than later. The first step of its solution was to build an AI yield-prediction model that used historical data on typical drivers of crop yields — including temperatures, rainfall, solar radiation, and other key variables. The second step was to overlay a climate change model with a representative concentration pathway (RCP) emissions scenario — one that was not too stringent but not too lax — onto its yield forecasting model to project crop yields for 2040 and 2050. (See Exhibit 2.)
Looking at three example regions, the company found that, relative to 2018 production, these three regions could see an approximate 14% reduction, 20% reduction, and 14% increase in yields by 2040, respectively. Where yields are projected to increase, the business needs no response; where they may decline, action is clearly required — whether buying more land, changing production methods, or transforming the regions facing declines to emulate those projecting an increase.
To further understand the impact of climate change, the company deployed AI models to project shifts in crop characteristics, such as chemical composition. These projections enabled the business to anticipate additional risks and opportunities associated with each qualitative metric, propose solutions for its existing territories, identify opportunities to procure new land, and understand where to extend or maintain production.
Of course, the company has not “solved” climate change — but it has made a start.
Beginning with baby steps
Climate change is clearly a difficult problem, and not all businesses can build extensive AI models right away. Instead, they should think big — and start small.
To begin, they should look at the data they have in hand to understand the additional data they need. They should then collect, process, and normalize that data. Next, they should build simple, physics- and agricultural-science based linear models to get a sense of the primary determinants of yield for their crops.
Of course, any crop yield forecast will not be driven by just a few variables: the business will need to consider soil and fertilizer properties, key weather variables, and many other factors. It can then use this information to gradually build out a more complex AI model, adding variables one by one (within physics and agricultural science constraints), then overlaying different climate change scenarios to generate robust yield projections and, eventually, projections for other qualitative metrics. As a result, it can make educated decisions about production on current land and whether new land is needed.
