If there's one thing I love to cook, it's soup. There's nothing quite like a homemade stock, hearty vegetables and savory beef, chicken or pork. I can't count the variety of soups I've made over the years, and even when I follow the same recipe time and time again, I almost never measure my seasonings. I do it by eye and by taste. No two batches are ever exactly the same.
Just like every batch of soup is a little different, no two growing seasons are exactly the same. Analysts may arrive at similar estimates but take different routes to get there. The same is true for algorithmic and statistical models that rely on satellite data, historical crop condition, weather data, etc. Just because the two models incorporate large amounts of identical data, it does not mean they'll arrive at a similar answer.
Two companies that generate such models -- Gro Intelligence and Indigo Agriculture -- are a case in point.
Both draw heavily from publicly available satellite data and employ machine-learning methods to arrive at their estimates. Machine learning is a statistics technique where an algorithm assesses large amounts of historical data to make a prediction about future.
Gro's estimate, 171.8 bushels per acre, was above than USDA's by 3.4 bpa and higher than any analyst included in a Dow Jones survey. Indigo capped the opposite end of the spectrum at 161.2 bpa, 7.2 bpa below USDA.
"The recipe is in the soup," DTN lead analyst Todd Hultman said. (My stomach growled, and then I debated which soup to make this weekend when the weather in Tennessee will finally feel like fall. I've settled on beef vegetable, but I digress.)
It's been a high-drama growing season, said Barclay Rogers, VP of Business Development for Indigo's GeoInnovation team, and it's putting all statistical models to the test.
"Our projections are well below market expectations and well below where the USDA was in September. We believe in our model," he told DTN. "This is an unusual year, and we will see how it plays out over time, but we have we have confidence in our model because our model has performed very well historically."
Both Gro and Indigo's models are constantly updating as they incorporate new data, a feature that both organization's customers can access in real time.
Last August, the DTN/Progressive Farmer 2019 Digital Yield Tour was conducted and combined Gro Intelligence's yield forecasts with on-the-ground perspective from farmers. Over the course of that tour, Gro's Vice President of Agribusiness James Heneghan said this year is going to put these machine-learning models to the test.
For perspective, when DTN conducted the digital yield tour, Gro's national average yield was 163.2 bpa, more than 6 bpa below USDA at the time. A month later, Gro's yields had climbed to around 170 bpa, while USDA lowered its yield estimate by 1.3 bpa to 168.2 bpa.
Ultimately, the question will be whether a warmer-than-usual September and early October was enough to help crops reach maturity, and how much this late-week snowstorm across the upper Midwest takes off those late-developing crops.
In the meantime, the challenge for growers is what to do with this information. Like Hultman said, the recipe is in the soup. You can't really live on soup. Make sure you incorporate some warm, crusty rolls into your decision making and perhaps a salad.
And remember: it doesn't matter which model gets closest to the actual yield at the end of the day. What matters is the market's perception of the supply and demand situation. So look at a wide array of sources, consider DTN's Six Factors Market Strategies and make the best marketing decisions for you.
And please, make some soup. Just don't try to live on it alone.
Katie Dehlinger can be reached at Katie.firstname.lastname@example.org
Follow her on Twitter @KatieD_DTN
© Copyright 2019 DTN/The Progressive Farmer. All rights reserved.