At DXD Ag Insights, we are working with a type of data that you don't hear much about in ag data analytics: in-field scouting observations.
We all know that in-field observations are critical for making pest management and other in-season recommendations. Once an issue is identified, agronomists add in their knowledge of the science, their experience with the field, and local weather patterns to identify the best course of action. Even with so much to consider, decisions need to be made quickly to stay ahead of the curve.
However, trying to look back and review those observations in an organized fashion is difficult, and far too often involves hours struggling to wrangle data in an Excel spreadsheet. We can and will do better.
One of our goals at DXD Ag Insights is to make the process of using scouting data faster and more effective, by building products that help agronomists explore and analyze their in-field observations combined with operation data such as planting and yield data, in-field weather, and soil data.
Why Hasn't AgTech Analyzed Scouting Data?
There are two reasons that you don’t hear about using scouting observations for analytics very often.
First, scouting observations are generally less structured as compared to something like operation data being recorded by the equipment. In general, unstructured data is much more difficult to deal with. The data needs to be standardized in order to gain insights across fields and to create useful comparisons, which can become difficult when multiple people are collecting the data. Click HERE for some tips on collecting useful data.
For example, pest populations or pressure is often described by a rating such as high, medium, or low. Is a 'Low' rating to one person the same as 'Low' to another? If the growth stage is important, is everyone recording that information on every visit to a field?
A second reason that scouting observations have not been in analyses as compared to equipment data is the perception that you must have lots of data to apply machine learning. This is often the case, depending on the particular algorithm or with visualization methods to learn from data.
More important than big data is having the “right” data, that is, data that is relevant to your situation. If you took a statistics class, you may have heard something like “make sure the sample you are analyzing is representative of the population you are studying.” We need to use this same concept with machine learning. Is the data representative of the fields you work with? Could analyses that consider relationships between the planted variety and yield be ignoring the impacts of local pest pressures, and local soil types?
The ‘right” data is also reliable. Or in other words, does the data mean what you think it does? Is it in a format that can be worked with? When you have collected or if you know that the data was collected with the appropriate procedure and in cropping situations similar to yours, then you know them you know the data is right – both relevant to you and reliable.
Despite its challenges, in-field scouting observations very frequently are both the right data, and the most reliable data for the questions agronomists need to answer. Just consider the significant effort and cost required to manage for local pest pressure and local soil types- all of these impacts can be 'aggregated away' when only using equipment data pooled across a region.
The Future of Ag Data Analysis
The future of agtech will include easier-to-use software to explore and visualize data, combined with emerging methods to handle and clean unstructured data. When combined with improving methods to obtain field-level weather data, it will be exciting to see what is learned from on-the-ground observations. And of course, having the right expert, the agronomist, involved with data analytics, will be a critical factor for the successful application of AI to improve crop management decisions over time.
Stay tuned, we are working on building that future for agronomists!
To receive these posts in your email, subscribe below: