Four Mistakes to Avoid When Estimating Yields
Winter means planning and preparing for the coming growing season for many farmers, and here on the data science team we’re working with many of our farmer customers as they plan inputs and create financial forecasts based upon their yields and profits from previous years. In these times of low commodity prices especially, taking time to analyze and learn from prior production years is more important than ever. It’s also a great risk prevention tool. Craig Linneman of Linneman Farms, a Granular customer from Missouri, summed it up recently when he said, “My philosophy on dealing with the uncertainty on both the revenue and input sides is to identify the analytics that will best help me to be profitable over many years leading to long-term success.”
While unpredictability abounds in farming and there are many variables that are hard to plan for, one of our key goals at Granular is to help you understand the best way to set yourself and your operation up for success, both in the short and long-term. With that in mind, I want to focus on analytics around perhaps the biggest and most uncontrollable factor in profitability uncertainty — yield.
The Surprising Costs of Poor Yield Estimates
We find that most farms keep detailed records of their historical yields, often with field-level yield data that goes back ten, sometimes even twenty, years. However, most farms aren’t taking full advantage of this invaluable information when estimating their future yields.
Before going into techniques for estimating yield, let’s first establish why estimating yield is so important.
How poor estimates can influence your bottom-line by more than $100/ac
Poor yield estimates can lead to poor decision-making. Let’s look at two common examples where it is critical to have good yield forecasts: crop choice and land rents.
Crop Choice Decisions: We see that growers’ yield estimates typically differ from more accurate statistical estimates by 10 to 20 bu/acre, but even smaller differences can have a big impact on your bottom-line. For example, one grower we work with was overestimating corn yields by only 4 bu/acre and underestimating soybean yields by 7 bu/acre. When calculating the relative profitability of those two crops, these estimates led to underestimating the profitability advantage of soybeans by about $60/acre.
Land Rent Decisions: In another example, a customer overestimated their non-irrigated yields by 49 bu/acre for corn and 5 bu/acre for soybeans. That means that they overestimated the profitability of their fields somewhere between $45 to $195/acre. With better yield estimates, you have better data to dial-in optimal land rents, or drop fields that no longer make financial sense to farm. The Granular Land Cost Benchmarking tool is an excellent tool that can be useful when making these tough decisions. A Granular Business customer who used the Land Cost Benchmarking tool to decide on dropping an unprofitable field for this production cycle said, “I hate to lose ground, but hate losing money more.”
Common Mistakes When Estimating Crop Yield
I think we can all agree on the importance of getting forecasts as accurate as possible. In order to get them closer, let’s talk about a few common mistakes we see growers make when estimating their yields:
While it can be tempting to put in yield estimates that reveal attractive bottom-line revenue numbers, the first step in making good estimates is to base your numbers on real historical data—not emotion or gut instincts. Unlike grain prices, the best predictor of future yield performance is past performance.
Many calculate their yield estimates on the conservative side. While this strategy is more likely to lead to good feelings come harvest, why not put your best yield estimates forward? Conservative yield estimates can lead to suboptimal input decisions and won’t allow for accurate planning when it comes to grain inventory storage, or how to best execute on your grain marketing strategy.
We find that many growers are hesitant to use all of their historical yield data when forecasting next year’s yield. We often hear that varieties and practices have changed enough that data from ten years ago is no longer a good predictor of future yields. We disagree. Data from ten years ago is still very relevant when estimating next year’s yield. You want to make sure that you are capturing the risk of “unusual events” and that can’t be done if you are only looking at the past several years.
ARC payments have made popular the idea of “Olympic averages”, which means throwing out the highest and lowest data points. While this may make sense when judging figure skating, it is not a good practice when trying to forecast long-term expected yields. Consider the following 5-year yield scenario:
|Year 1||Year 2||Year 3||Year 4||Year 5|
|150 bu/ac||50 bu/ac||150 bu/ac||150 bu/ac||150 bu/ac|
An Olympic average would be equal to 150 bu/acre. If you take into account the bad years, which have a 20% chance of happening, you expect a yield closer to 130 bu/acre per year in the long run. Using the tips below, we think you can do even better!
Tips for Making More Accurate Yield Forecasts
By applying some relatively straightforward techniques, it is possible to build robust long-term expected yield estimates. Often overlooked methods to use when creating crop yield estimates:
- Use all your historical data: Take into account a farm’s full history and don’t “throw” out unusually good or bad years. Remember, we’re trying to capture long-term yield estimates. The good and the bad occur over the long-term, so let’s capture these outcomes in our estimates.
- Don’t isolate weather. Don’t penalize a field for the vagaries of nature. Just because you happened to grow corn on that field during a drought year doesn’t mean it is a bad field. This only means that you should take historical weather effects into account when estimating future yield.
- Quantify the yield risk: This can be done for both individual fields and your entire farm. Knowing the probability of specific yield outcomes is incredibly powerful.
Using these principles, there’s an opportunity to dial-in your long-term expected yield estimates. In addition to the best practices just mentioned, the Granular long-term expected yield model, built on these same principles, has been shown to be 40% more accurate for estimating long-term expected yield than an Olympic Average.
To learn more about Granular products and tools that can help you best estimate yields in 2019, speak to someone on our team.
Ky Kiefer is a data scientist at Granular, where he builds machine learning models based on farm data and satellite imagery to help Granular’s customers make more informed decisions to drive profitability. He holds a Chemical Engineering degree from the University of Michigan. Ky can be reached at firstname.lastname@example.org with any questions.
Stay in the know!
Subscribe to get new blog posts sent straight to your email