For the past year, the team at Soil Action has been working toward building an artificial intelligence driven product to sense soil nutrition in real-time. Whereas other companies have attempted to revolutionize soil testing before, co-founders Jack Oslan and Nate Storey say the AI tools available today are making what was once difficult or nearly impossible, possible.
“Soils are unknown and misunderstood,” Storey says. “We can use AI to understand soil better, and our goal is to come up with the instruments to solve the problem.”
Soil Action’s solution in progress includes building models and training models pairing near infrared spectroscopy with AI. Its goal is to reengineer the traditional process of sampling, shipping, agronomic recommendations, prescription files and applications while making it all in real-time. They are doing on-farm demonstrations this fall.
Before founding Soil Action, these two businessmen first met 12 years ago co-founded indoor agriculture startup Plenty. Storey’s time at Plenty was applying his laser focus on yield with innovation in algorithmic nutrition.
“I went into indoor ag because it was an area with the largest opportunity to drive yield. I have a lot of interest in yield,” he says. “In indoor, you can control everything and measure it–everything can be known in those systems and control every part of the process: root zone temperature, gas composition, and more.”
Now, Storey and Oslan want to bring those learnings outside and into the field.
“We got really good at understanding how to take an algorithmic approach to yield. It’s about understanding the yield equation, breaking it apart, optimizing individual aspects, and restacking them,” Storey says. “In row crops, the soil is the most important part, and to solve the yield equation we have know the variables that correlate and then begin to manage.”
What Does The System Look Like?
Currently, the beta version product is housed in a 3”x6” steel tube which can be mounted on any style of implement or equipment to automatically take measurements 4” to 6” deep every 50’.
“The real end goal is to have every equipment cab be mounted with an AI enabled agent to give you real-time measurements of what’s going on in your field,” Storey says. “It’s an AI agent focused on optimizing yield.”
The first testing was conducted in northern Iowa.
“We’re building our models on data collected from the field, and we’re using deep learning to ingest all of the information and help understand correlations,” Oslan says. “We can see everything that’s there, but we don’t understand everything that is there. That’s a focus for our work right now.”
“Holy Grail of Soil Sampling”
When it’s ready to be commercially available, Soil Action aims to provide results measuring two forms of nitrogen, phosphorus and potassium. Other crop nutrients will be added in the future.
“Every expert we talked to said we couldn’t use NIRS in soil sampling, but the physics said we could,” Oslan says. “We took two intensive weeks using sand and manipulating it for measurements with NIRS, and our deep learning models can untangle data in a way classical statistical methods cannot. Now, it’s about how fast we can solve for soil nutrients with these newer tools.”
Soil Action says it aims to provide the equipment to farmers for a hardware fee of $10,000 paired with a subscription for the analysis on an annual fee basis.
