"Deep learning" as a phrase doesn't hold much interest to me. But applying machine learning to industry problems catches my eye - especially when public welfare is at stake. That's how a pitch from The Climate Corporation rose to the top of my inbox.
Soon I was talking to Steven Ward, Director of Geospatial Sciences at The Climate Corporation, about how they help farmers improve their yields - and sustain their soil - with deep learning.
The population - agriculture gap poses a threat
We don't think about the population-and-agriculture problem every day, but it's serious. As Ward told me:
This is really at the root of our mission statement as an organization. By 2050, we are going to increase the global population by a significant volume - in the order of 3 billion people.
But today's farms couldn't support that:
Currently, while that trend is going up, we see a downward trend in the number of production scale acres in the agricultural industry. You can see there's obvious gaps related to food security and food supply. Some of this can be seen in some parts of the world right now.
That fuels The Climate Company's ambitions:
We feel really strongly that it's our job to help protect yields on the agricultural fields, as well as help farmers figure out a way to produce more on less land, with less resources and input. That's the reality of the world moving forward.
Closing the farming gap with data
Data presents a new frontier for closing that gap:
In order to do that we consider sustainability, and we have to consider optimization of agricultural practices. We sincerely believe that data is the way to get there; innovation is the way to get there. It's going to help us get ahead of this curve.
But when I brought up the problem of corporate agri-business, Ward pushed back. In his experience, farms large and small are already tech-savvy, and eager to try new approaches:
There's a misconception that farmers are not willing to change or are stuck in a brute force mentality. What history shows is that farmers are innovators. They are people that will adapt and change, when presented with the opportunity... So whether it is the farmer we're supporting on a fifty acre or a 10,000-acre field, in the agricultural industry is farmers are very open to adopting new tech.
Now, we believe the next inflection point is data. Farmers are voracious data collectors; data wants to be analyzed. We are presenting them with an opportunity to do so.
The Climate Corporate's efforts center around their flagship product, FieldView. FieldView addresses three data issues farmers struggle with:
- Compiling data that's spread across numerous data sources, from laptops to flash drives to IoT/equipment sensors
- Extracting insights that busy farmers can use to improve their yields, without time consuming analysis
- Giving them access to data/insights in a form they can easily use in the field (e.g. easily accessed via tablets and phones on foot, or while driving equipment)
Forget about machine learning without a data platform
It might sound obvious, but Ward says other tools haven't addressed these problems in a cohesive way. Disparate data is an issue for any industry, but with farming, it's an absolute deal-breaker:
The farm is a living system. Just like living systems will tell you, a change in any one component will change the others. To look at any of these data sets in isolation is the not the appropriate way to approach problems on the farm. We have to list them all in the context of one another.
But there's another huge benefit to bringing all the data onto one platform. From there, you can enhance with emerging "AI" tools like machine learning and spatial modeling. Ward:
This is where advanced technology like AI is bringing us that opportunity. Whether it's a fertility decision, or a feed decision, or looking at imagery to help you understand changes in crop physiology and health - all of these things are in the background of our application.
AI in action - fertility management and recommendations
The Climate Corporation is researching new AI models as we speak. The best ones are a convergence of mobile consumption and AI tech, such as mobile image recognition of plant diseases. Another example: the Nitrogen Monitoring Tool.
Fertility recommendations are [embedded] in our monitoring tools, such as the Nitrogen Monitoring Tool. The technology is allowing us to pull in information the grower provides himself. These are very complex and well validated fertility models.
These models also pull in external data sources such as weather information and satellite data:
All of these different sources of information leads to a single recommendation for a grower. This is a great example of blending remote sensing with traditional agronomic modeling. It's all under the umbrella of AI.
Sounds useful, but what do farmers think? The Climate Corporation sent me a video of Greg Deim, a farmer from Iowa:
Climate FieldView has identified the variability in our fields. It makes us ask the question why - looking at why is a certain area high yielding, and what can we do to keep that a high yielding area.
Deim spoke to the before-and-after:
Before Climate FieldView, we would have to sit down and compile all the data... Sitting at an iPad or a laptop, we can sit down and within minutes, have a seeding prescription filled.
He's using The Climate Corporation for nitrogen management:
The Nitrogen Management Tool has definitely helped us to improve our nitrogen applications. With the Nitrogen Monitoring Tool, we've been able to determine that there are areas of our farms that we need to take a look at.
Knowing why we have to apply more to a certain area will help us justify our ultimate goal: to increase our yield and increase our bottom line.
The wrap - AI isn't a cure-all
After our talk, The Climate Corporation clarified that while the Nitrogen Management Tool has been on the market for two years, they've been giving farmers field recommendations on soil optimization for years. Their 150 data and research scientists continue to make on-site visits to keep the field views coming.
Two other crucial ingredients: an open platform for partners (and even competitors) to extend/build on, and the design chops to make apps farmers love. But 150 data scientists would be useless if they weren't industry-savvy:
AI tech and computing has been a huge force multiplier for us... But the key is having the data to run in those environments, and having scientists who know what questions to ask, and what methods to apply to solve those questions.
With 700 employees across the world, The Climate Corporation is reaching a scale where the benefits of deep learning on farming can be measured. In a recent study, they compared 66,290 acres in 653 fields where customers should have managed to 15lbs surplus, and applied their Nitrogen in a variable rate application.
After looking back at their 2017 Nitrogen Beta, The Climate Corporation determined that users could have experienced an average of $12.31/acre profit (through yield increase and nitrogen savings) using the Nitrogen Management Tool.
As the benefits gain traction, Ward cautions not to view AI as an agricultural cure-all:
It's not as simple as what a lot people might perceive it to be.
The Climate Corporation is spreading its tools and mission to new areas. The company started in the "corn belt" of the U.S. and expanded into the north and south. Last year, they expanded to Canada and Brazil, as well as a handful of countries in Europe.
Our agricultural problems won't go away with a sexy algorithm. But for The Climate Corporation, their relationships with farmers drives them forward:
It's an industry that's been hungry for this. It's fun to work with them. That's really motivating for the staff here - it's awesome to see your research applied in the real world.