Machine Learning and Artificial Intelligence are speeding up the development of new crop varieties to adapt to climate change — to sustain heat!
Among the UN 17 Sustainable Development Goals (SDGs) is achieving food security and improved nutrition within a sustainability perspective, by the year 2030. These SDGs succeeding the previous UN Millennium Development Goals are all interconnected. Interestingly it is the food that connects them all as per the embedded model by the Stockholm Resilience Center.
Changing climate conditions
As we are in the midst of changing climate change, there is an urgent need to speed up operational processes to produce more food to achieve food security and agriculture sustainability. As illustration of current changing climate conditions we have used climate datasets of Canada to examine the hypothesized trends or shifts in some of the climate variables, such as changing trends in temperatures.
The figures above are illustrations of temperature trends, in Canada, for winter and summer of the year 1955, 1975, 1995 and 2015, respectively. These maps show that the changes are more noticeable during winter as monthly average temperatures of January continue to increase.
The trend of the increase in temperatures is a global phenomena with the extreme north registering more abnormal increase. The map above produced by NASA, using Robinson projection, displays global surface temperature anomalies for the period January 2016 through June 2016.
ML/AI to speed up adaptation and mitigation
To adapt to changing climate conditions we searched for adaptive traits, using different machine learning techniques where the different layers of climate maps or surfaces, such those above were used as predictors, including different surface maps. In this process, different sources of climate surface data were used including large data sets from the Canadian Centre for Climate Modelling and Analysis.
The other dataset used in the modeling process included recorded data from large number of plant samples that are stored in genebanks, worldwide. The machine learning techniques were then performed to virtually evaluate whether these plant samples have the traits needed to provide crops with resilience to sustain climate change, such as tolerance to heat.
There are more than 7 million plant samples stored in more than 1 750 genebanks, worldwide. Such samples can be the source of traits that can have impressive impacts on plant crop yields, such as the traits found in the Japanese wheat “Norin 10” plant samples.
The samples that were virtually screened to likely have the traits of adaptation to sustain heat were grown in the field to confirm whether they actually have the traits as predicted by the different ML models.
Interestingly the agreements, based on AUC curve and other metrics, were found between the ML predictions and the actual real field evaluation and observations. Those sample plants that have been predicted and selected by the models to sustain heat virtually are also those that sustained heat under the real heat conditions. The plants predicted by the models to have heat traits, they also have the ability to lower their surface temperature , under heat conditions, when compared to a random sample of plants (figure below).
The plants identified as likely to have the sought for traits of adaptation are currently used in the development of several crops, worldwide. Some of the traits found have been searched for in vain, in the past.
The results demonstrate in practice the potential of ML and on the overall the potential of AI in speeding up the adaptation to and mitigation to climate change.
Some other links on ML/AI and “big data” in agriculture:
In silico evaluation of plant genetic resources to search for traits for adaptation to climate change. Climatic Change 134(4): 667–680. http://dx.doi.org/10.1007/s10584-015-1541-9