Reference: Alexis Hoffman et al 2020 Environ. Res. Lett. in press https://doi.org/10.1088/1748-9326/ab7b22
Imagine you are a farmer who has been growing the same crop in your hometown for generations. Then, someone tells you that a changing climate will soon make your area no longer the ideal place for this crop to flourish. Farmers already deal with dynamic weather that can either bolster or destroy a harvest each year. Now, climate shifts may soon present another hurdle if global warming continues without drastic intervention. Farmers will be increasingly reliant on technological advances to inform them how climate change will affect their crops. Machine learning, such as artificial intelligence (AI), can quickly analyze multiple climate variables over many areas at once to provide large scale trends. With its automated processes and huge sample size, AI can more precisely and accuracy analyze data than humans. Farmers can use this information to adapt their farms and crops.
What kind of data can AI offer?
Researchers at Penn State University recently published a paper online in Environmental Research Letters in which they implemented AI to analyze the dependence of summer crop yields on climate between 1980-2016. Using this model, they then predicted future trends for the year 2064. Overall, they suggest that if warming continues, the best climate for growing maize and soybean (currently Iowa and Illinois) may shift northwest to Minnesota and the Dakotas.
For me, trying to correlate one variable with one plant (how many hours of sunlight my rattlesnake plant needs to survive) is a recipe for disaster. These researchers correlated six fundamental climate variables with the grain yield of three crops (maize, sorghum, and soybean) in 18 states over 36 years – all performed with computer algorithms. They included analysis for the entire growing season, but also divided each season into specific growing phases: “establishment” (early germination), “critical window” (reproduction – determines yield potential), and “grain filling” (actual growth).
Teaching a computer to inform human productivity
Artificial intelligence combines massive datasets with scientist-built algorithms. Researchers “teach” a computer to learn patterns in a small set of the data. The computer remembers what it was taught and uses that map to sort and analyze new data from massive datasets automatically. It’s important to note that much of the data that will fuel artificial intelligence is already available – it just needs to be formatted into an appropriately built algorithm.
For their crop study, the scientists at Penn State gathered crop yield data from the United States Department of Agriculture and climate data from a commercially available meteorological product. Along with their estimated planting dates and identified growing phases for the crops, the researchers used the commercially available data to train artificial intelligence to determine phase-specific yield responses to climate variables. They used an algorithm called “Random Forest” composed of many individual “decision trees”. Although Random Forest provides most of the advantages of AI, it is also one of the more complex algorithms and requires a longer computer training period than a singular decision tree.
A new view on patterns and detailed correlations
What makes this study unique from similar ones is the level of detail the more complex Random Forest algorithm allowed the researchers to analyze. In addition to whole season analysis, the researchers broke down the growing phases of the crop. They also performed individual analyses on three summer crops in a large geographical location in the central United States. And, they expanded their climate assessment to six variables:
- Maximum and minimum temperature
- Extreme degree days (calculated)
- Solar radiation
- Vapor pressure deficit (dryness) (calculated)
The researchers found that temperature and precipitation had the largest effects on crop yields. This result makes sense based on previous records and what farmers already know about crop growth. AI allowed them to go even deeper. Specifically, they noticed that ideal temperature had the greatest influence during the grain filling phase over either the establishment (germination) or the critical window phases. This information could be helpful to scientists or farmers who try to predict good growing seasons. While the researchers found some correlations with other variables like extreme degree days and dryness, they didn’t exhibit clear thresholds like temperature and precipitation.
When the researchers plotted the variables for optimal conditions against each year, they found that they could correctly predict which years had low (2012) and average (2016) maize yields. This ‘validation’ helped increase confidence in their findings. To model a futuristic scenario, the researchers generated an optimal growing region figure for 2064 that took into account the current climate change projections. The results demonstrate a northwestward shift in optimal growing conditions for maize from Iowa and Illinois to Minnesota and the Dakotas. In fact, the optimal growing conditions predicted for 2064 appear remarkably similar to the conditions of low crop yield observed in 2012.
As of 2016, the best climate for maize, soybean, and sorghum production is in the geographic region (Iowa, Illinois) that has some of the most productive, or nutrient rich, soil. If the models are correct, the ideal climate for maize and soybean will shift northwest by 2064. Sorghum seems to have a greater tolerance to climate change.
How we might experience crops in a shifting climate
If the climate does shift, northwest farmers might find it difficult to produce as many food staples as Iowa/Illinois historically has, given the less productive soil. While it is not possible to predict how much such a shift will affect the general food population, one can imagine a scenario where these crops are less abundant than they are now. A lot of variables go into how consumers will be affected from climate change, making such futuristics predictions extremely difficult. However, it will be important for producers to know if a disruption to their current way of doing business is likely. With better computer models to predict future agricultural realities, communities have the chance to prepare and adapt.