How does the Traction Enterprise Yield Model work?

Overview:

Knowing the long term expected yield of an area is important for many field-level decisions, such as determining how much rent you can afford to pay or deciding what crop to grow. However, getting good yield estimates is tough, especially at the field level. That's why we developed the Traction Enterprise Yield Model.

Requirements:

To see our yield predictions, you will first need:

• To send in at least 5 years of yield data to your Customer Success Manager or Data Specialist.

Answer:

To give the best estimate possible, Traction Enterprise Business uses all of the historical data that you provide. (Some farms have provided records going back a full 30 years or more).

We trend-adjust the data, as needed, to take into account the fact that yields have increased over time due to improvements in management practices and seed quality. Traction Enterprise Business then uses your data to make field-specific yield estimates.

To build the Traction Enterprise model, we start with enterprise-wide yields: first calculating de-trended farm-wide averages (across the entire enterprise), and fitting that with a distribution to determine likely yield ranges across your entire operation.

Figure 1. Farm average commercial corn yields over a 20 year period from a farm in the northeastern U.S. Each vertical bar is a single year's yield. Traction Enterprise fits this data to estimate the odds of any particular outcome. On this case you can see that a farm average above 200 bu/acre is unlikely and the most likely outcome is a little under 150 bu/acre. 

Traction Enterprise scores each field's performance: for each year in which there is data on a field, Traction Enterprise compares field-level performance to the farm average to determine how each field compares to the rest of your operation.

Note: This is important, because a low corn yield in 2012, for example, doesn't necessarily mean you have a bad field. It may just mean you were growing corn during a drought.

Figure 2. Normalized field performance (field-level yield divided by that year's farm-wide yield median) for a single field from the farm shown above. Each vertical bar is performance in a single year. You can see that this field often beats the farm median (its score is greater than 1), and in one year it yielded 2x the farm median.

Traction Enterprise then combines Steps #1 & #2 to get a field level yield estimate.

By combining both farm-wide and field-specific data, Traction Enterprise can use all of your farm's available information to make a statistical estimate of both your expected yield and your associated yield risk.

Figure 3. Final yield profile for the field shown above. Note that since this field was above average compared to the rest of the farm (from Figure 2), its most likely yield outcome is over 150 bu/acre.

How the Traction Enterprise Yield Model compares to other estimating methods

When using actual farm data to measure the accuracy of yield estimates, we find the Traction Enterprise Yield Model is significantly better than many other methods of estimating the long-term expected yield of individual fields.  

Below you can see the root mean square error (RMSE) of 4 different yield estimation strategies. We find the Traction Enterprise model is almost twice as accurate as the second most accurate model (a field-specific 5-year Olympic average). 

Figure 4. Comparison of RMSE of 4 different yield estimation strategies for a dataset with 10 years of historical yield. The Traction Enterprise Yield Model significantly outperforms the other 3 methods.

How to access to the Traction Enterprise Yield Model

To get access to the Traction Enterprise Yield Model, simply contact your Customer Success Manager and provide them with 5 or more years of historical yield data in whatever format is convenient, such as spreadsheets or insurance records. We'll do all the rest!

 

Related Articles:
Getting Started: Landowner Yield Report