AG AI stands for Artificial Intelligence for Agriculture, and that’s what we’re doing.
Farming and bringing crops to market is a difficult business, which involves, among other things, managing complex biological systems and understanding how they respond to their environment. One project we’re currently working on at Pavo helps farmers develop a better understanding of what drives crop yield for hazelnut orchards. We’re using big data and machine learning techniques to expand the possibilities in prediction and proactive crop management.
Biological Complexity Captured
Trees are complex and each is a unique biological organism – no two are the same. The way a tree responds to its environmental variables and their individual impact on its yield is convoluted and intertwined; our models must reflect this hidden complexity. Capturing this diversity is an important part in understanding yield and an orchard’s response to its environment. We’re gathering information and, through the power of big data, developing models at an individual tree level using decades of historical information. A tree’s response to its local conditions is modeled, leveraging a combination of methods like neural nets (a.k.a. deep learning) and classical time series methods. We then scale these tree-level models to understand and predict the behavior of an entire orchard with all its biological diversity of possible outcomes. After our models are built, all that is required for a prediction of yield is a reasonable estimate of the expected future conditions.
Using Pavo’s network of sensors we’re able to measure the environmental conditions affecting crops closer to their point of impact – as localized as at an individual tree. This precision helps us reduce uncertainty in our yield predictions due to errors in the model inputs. Combining automated sensor data with decades of historical information on nut yields, collected with the same degree of precision, has created an extremely powerful dataset from which to construct models. We’re able to more accurately predict and understand the impact of varying environmental conditions and weather scenarios on orchards through highly localized and precise measurements of both the input conditions and output nut yield.
We’re leveraging a combination of methods like neural nets (a.k.a. deep learning) and classical time series analysis to capture a tree’s response to its local conditions. We then scale these tree-level models to understand and predict the behavior of an entire orchard with all its biological diversity of possibilities and outcomes.
Making Use of Community Feedback
Of course, predictions are only as good as the data that drive them. Bad data leads to the classic “garbage-in equals garbage-out” nightmare scenario.
In addition to the power of data harvested from our network of sensors, we’re collecting feedback and information from our community of experts and local farmers. This information is used to refine our understanding of not only what affects an orchard but to what degree it will impact a particular crop. It’s good to know when frost will occur, but even better if you can tell when it will have a minor or severe impact on your yield. We use the expertise and feedback from our community of users to continually evolve and enhance the power of our predictions.
We’re combining state-of-the-art data science techniques with our ability to collect precision data, and developing models that allow farmers to prepare for whatever Mother Nature may bring. And we’re taking these models even further, applying and adapting them to indoor agriculture for produce crops, like spinach, lettuce and strawberries, to ensure that the world’s farmers can continue to feed a growing global population. We believe that technology has an increasingly important role to play in meeting the rapidly increasing caloric demand, and our IoT blockchain solution is up to the task.
The Future – Bringing it Indoors
For an outdoor orchard, our ability to control and deliver optimum conditions can be somewhat limited. Indoor farming, however, provides an opportunity to fully exploit the power of precision crop measurements and the analytical capacity to determine optimal growth paths. Indoors, we can not only measure but ensure complete control over growing conditions. It’s possible to not only discover a crop’s optimum conditions path but to also carefully guide it along a journey that maximizes output. By combining precision measurements of actual growing conditions with knowledge of optimal crop development paths we can customize a plant’s journey from seed to harvest.