Could more accurate yield predictions help secondary and tertiary agricultural industry players better asses business risk? The founders of a Agrograph, a Wisconsin ag-software company, certainly think so.
Agrograph, a start-up based in Madison, Wisconsin, recently announced that it raised $ 500,000 in seed funding from the Idea Fund of La Crosse – a locally-focused investment group – to fund the expansion of its yield prediction software.
Predict crop yields at individual-field level
According to Agrograph’s official press release, the software combines high-resolution satellite imagery and field data with machine learning algorithms to predict crop yields at the individual-field level. This, they say, results in more granular and accurate bushel per acre predictions.
The company sells to 3 main customer groups:
- Crop insurers and lenders to better assess their risk portfolio;
- Grain distributors to improve their supply chain logistics, and
- Agricultural tech providers to enhance their precision agriculture solutions.
The press release quotes Mutlu Ozdogan, co-founder and CEO of Agrograph, as defining the agricultural industry by “volatility and variability.”
Agrograph is turning data into solutions at the individual field level, which is a level of insight the agricultural industry hasn’t had
“In a drought year, for example, one field might produce great results, while another just a handful of miles away won’t”, says Ozdogan. “When you are looking at data at the county or regional level, you simply can’t take into account that variability. […] Agrograph is turning data into solutions at the individual field level, which is a level of insight the agricultural industry hasn’t had.”
In a later email exchange, Jim O’Brien, one of Agrograph’s co-founders, describes the software as bundling satellite imagery along with soil, weather and cropping models to create a crop type identification map and derivative products for both historical and in-season yield predictions. This field-scale yield data, he says, builds greater accuracy into the company’s models; in other words, the measurement method becomes more accurate with greater volumes of aggregate data.
Text continues underneath image
Predictive yield data has value for the individual farmer, Agrograph says, from in-season crop assessment to grain marketing and damage assessment (in the case of hail or wind, for example). - Photo: AFP
The assumption he and his colleagues are making, says O’Brien, is that data generated by their software will be offered to farmers “likely as a white-label solution” from a partner company. Predictive yield data has value for the individual farmer, he says, from in-season crop assessment to grain marketing and damage assessment (in the case of hail or wind, for example).
Operational decisions are still decided on a field-scale
“Our focus on the field scale is the same reason a farmer’s focus is on the field scale – because that is their benchmark for the smallest unit of measurement”, O’Brien says. “Yes, precision ag allows you to ‘see’ variability to the square meter, but operational decisions are still decided on a field-scale.”
On the question of data ownership and transparency, O’Brien says the Agrograph policy focuses on the idea that gaining new insights starts with sharing current insights. “I see this sharing economy continuing to mature worldwide and we can see this in our own spaces with companies like PayPal, Google, WeChat, etc.”, he says.
Data for machine learning algorithms
“From our standpoint, our technology utilizes data for machine learning algorithms and it’s not a direct source back to the contributor. Just like Google uses your cell phone location in their traffic algorithm, they are not pin-pointing and sharing your location to the wider community, unless you want to share it.”
Agrograph’s software is purported to have a prediction error rate between 5 and 15 per cent. In response to questions on the significance of that rate, O’Brien says his company is ahead of the industry and are unique in their field-scale yield prediction.
“We see billion-dollar decisions being made with anecdotal data across these sectors. It’s not that these companies are make poor decisions today”, he says. “It’s that they have limited information and a short timeline to make billion-dollar decisions and getting it wrong is very expensive.”