Using AI to Reduce Time-to-Market

Syngenta help improve global food security by enabling millions of farmers to make better use of biotechnologies.

Challenge

Today, the average farmer feeds about 6 times the number of people they did in 1960. And by 2050, they’ll feed more than two-thirds as many again! These amazing efficiencies are only possible with the high yields that technology-driven breeding can offer. Syngenta is a leading agriculture company helping to improve global food security by enabling farmers to make better use seed genomics. Developing a new, high-yielding seed variety is highly technical and time-consuming. A single variety can take about 9 years and lots of testing resources. For this reason, Syngenta wanted to develop methods for quickly selecting high-yielding soybean varieties.

Strategy and Solution

I worked with another research scientist to deliver a solution to Syngenta. The approach was to organize the data that Syngenta collected, supplement it with additional remote sensing data, and build a high-accuracy ML model.

Digging Deep into the Data

The first challenge of this project was bringing together a large amount of data in a usable way. I cleaned and organized the data that Syngenta was collected from sources as diverse as weather reports, soil characteristics, solar radiation, genetic single nucleotide polymorphisms (SNPs), and yield from over 50 test farms. Then, I merged it with remote sensing data collected from the NASA Moderate Resolution Imaging Spectroradiometer (MODIS) database. Vegetation indices derived from this database have been shown to be good estimators of crop yield and biomass.

syngenta-graph

Building a ML Model that Reduces Development Time

I used ML methods to identify soybean varieties that were consistently high-yielding under a range of environmental, geographical, and soil conditions. The ensemble model that was built was able to predict high-yielding varieties with 99 % accuracy. Most importantly, I used the model to identify 15 soybean varieties that were expected to perform well once commercialized. Using this method allows Syngenta to increase the yields of their commercial soybean varieties by ~5 % and shorten the time it takes to develop new varieties.

Transformation

Syngenta can help farmers grow their profitability by providing them with consistently high-yielding soybean varieties. Moreover, this method for seed selection drastically cuts down on Syngenta’s time-to-market. As part of Syngenta’s Good Growth Plan, they have committed to increasing the average productivity of the world’s major crops to 20% without using more land, water, or input, and data-driven processes like this are helping them reach this goal.