Unfortunately for many, the end of the season doesn’t indicate the end of anything – more often than not, it is the beginning of a deeper analysis to understand what happened in the season and how it impacted the outcomes for the season, whether selling bushels of corn for profit or determining which hybrid seed performed better and, thus, should be produced at scale in the coming year.  

One of the challenges during this time of the season comes from getting the information required to make the decisions to inform the decision-making process. Every growing season has its own handful of variables – in 2022, it may have felt like they were unending: A slow start to the season due to cold and rain, followed by relentless heat; not to mention the implications from a supply chain perspective – creating challenges for finding a supply of key products to accelerated costs if they were found on the shelf.  

As we think about this analysis, it can feel like a bigger challenge at hand than any of the other issues from earlier in the season. How do you get the data to clearly understand what happened during the season? Can you trust in the accuracy? Should you use it to make future decisions?  

That’s where the latest technology comes into play – and there is a huge opportunity for machine learning in agriculture.  

What is Machine Learning?  

As a focus within artificial intelligence, machine learning is the capability for a machine to learn how to solve problems like how humans do. Using algorithms and statistical models, the machine “learns” how to respond to data based on the data it has been trained on, delivering responses in a consumable format.  

Machine learning exists in our everyday lives – from recommendations on what to watch next on our favorite streaming services to powering self-driving tractors – and it’s something that isn’t going away. In fact, a recent study found that two-thirds of companies are currently using machine learning and 97% predict that they’ll use it in the next year.  

That’s great, but how does it actually work? How can machines learn and essentially think?  

The most recent advancements to machine learning that allowed for the leap forward in accuracy and model complexity goes a layer deeper into a subset of machine learning called deep learning. Deep learning is taking artificial neural networks – a set of machine learning algorithms that have been around for decades – to the next level creating what is known as deep neural networks.  

This happens by showing the neural networks large amounts of data together with the response they should learn to extract, known as the training phase of the neural network. Once the network has been trained, it has “learned” to generate the desired result on data it has not seen before. The network’s ability to generalize well and generate more accurate insights on new data depends on its complexity and the variability of the data it has been trained on.  

Let’s break this down a bit more using an example near and dear to agriculture: weather.  

Weather is unpredictable, and often the most complex factor when it comes to agriculture. Like the Farmer’s Almanac has done for years, daily high temperatures for a particular location are loaded into a software platform for analysis via deep learning.  

So let’s say that a deep learning model is created to evaluate the temperature differences and  learn about key indicators that can be translated into predictions for future weather subsets. Perhaps the model tracks that every decade, temperatures increase by one-half degree. While this seems minuscule, this does have an impact on growing degree days, as well as timing for planting and harvest.  

As a result, the deep learning framework creates assumptions for ideal planting and harvest dates, given the tracked temperature increase as well as other key indicators. The resulting data set provides data-driven insight to optimize the growing season – more so than focusing on how the season has played out in seasons past.  

This is just one way that deep learning can work to provide predictive insights to agriculture – helping to transform data delivery and promote faster and more accurate analysis to make critical decisions about outcomes and performance.  

Machine Learning in Agriculture: What Does It Mean? 

Finding ways to be more efficient in agriculture – particularly when it comes to crop production – serves as paramount to delivering food to the world’s population.  

There is a finite amount of cropland available, but the world’s population continues to grow – the United Nations predicts that it will be 2 billion people by 2050. And, with that population increase, food production will need to increase by 60% to meet those demands.  

There is an urgency for agriscience researchers and leaders to uncover the next-best products and hybrid seeds to support this growth – all while battling other issues like rising temperatures, nutrient deficiencies, and resource constraints (like water).  

At the same time, growers and their advisors need access to the best data possible to make more precise decisions to optimize yields – meaning that they can produce more on the land that’s available.  

In each of these cases, deep learning has a place to support their needs – helping both get the data they need in the required timeframe to make the critical business decisions, both in- and post-season.  

Agriscience & Machine Learning  

Agriscience is tasked with one of the biggest challenges for our society: increasing food productivity to feed the world’s population.   

Whether it’s via hybrid seeds that produce a larger yield or new inputs that optimize how crops perform, uncovering agricultural advancements to drive crop production forward requires a dedication to precision and accuracy.  

Today, much of the data that’s captured during research trials and even during seed production rely on manual methods – or humans going out to the location, picking a place within the field, and counting the number of plants or measuring various characteristics of the selected group of plants. While this has been the way the industry has relied on measurements in the past, it creates a high rate of error – and when fueling such important work, it creates a huge amount of risk.  

Using ag drone systems to capture key measurements to characterize plant growth and performance has become popular in recent years. But often times, it ends there. The researcher can capture the data, get a visual indication of what’s happening in the field, and continues to capture more detailed data via manual methods.  

This is where deep learning has a place. The high-resolution aerial imagery captured by drones and sensors becomes a data set that the deep learning platform then translates into detailed measurements, providing insight from the number of plants that have emerged at a particular point in the season to counting the number of tassels on the corn. 

As a result, researchers have access to a detailed report with 10x more measurements than what can be collected via manual methods. While more data isn’t always beneficial, in this case, it’s necessary. It provides supplementary data that further validates research hypotheses, but also details additional measurements that provide crucial insight into crop health and performance throughout the season.  

In addition to eliminating the chance of human error, using a science-grade ag sensor can remove variability. This means that you don’t have to take into account the cloud cover between drone flights – resulting in accurate data that can be analyzed over a period of time.  

The Power of Machine Learning for Crop Production 

For those that advise growers, machine learning has its place, too.  

Market dynamics and weather challenges have made the role of the agronomic advisor – precision ag tech specialists and managers, seed and products salespeople, and retail ag leaders and managers – increasingly difficult. After all, how can you best advise what a grower should do when market conditions throw curveballs into ongoing plans?  

Like agriscience, data becomes essential at this point. Remote sensing techniques, like weather, soil, and satellite data, have become more popular in recent years. Easy access to this data via ag software platforms fuels instant insight and thus, critical decisions to effectively manage crop health and performance.  

It goes beyond that. Capturing high-resolution imagery via drones and sensors better informs boots-on-the-ground scouting, as the resulting data sets offer key visualizations to understand where low-performing areas may be in a field.  

Just like the data for agriscience, when these data sets are fed to a machine learning platform, it means that more data can be analyzed faster – delivering it in near real-time while providing an indication of full-field performance. And most importantly? This is delivered without sacrificing the accuracy or precision of the data.  

Stand Count, for instance, becomes a lot easier to manage when you have a crop scouting drone that can fly thousands of acres in a matter of hours; the amount of time that it may take to spot scout a few fields.  

The result? Detailed analysis, acre by acre, to understand exactly what’s happening within the field – and a clearer ability to inform replant decisions, as well as ongoing inputs throughout the season.  

Machine Learning in Agriculture: The Future 

Experts agree that agriculture is a key industry where artificial intelligence, machine learning, and deep learning have a transformative presence. 

From providing actionable data faster than with manual methods (not to mention promoting methods that collect data much faster than current methods) to giving visibility into key measurements that otherwise are not readily available, the adoption of machine learning across the industry will likely be the inflection point that finally spurts the fourth industrial revolution for agriculture forward.  

And, for those who are busy uncovering what can be done better for the season to come, machine learning becomes a helping hand to inform critical business decisions while supporting the broader goal of delivering food security for the world’s future population.