Commodity Crop Verification

Leveraging remote sensing to verify producer reported commodity Crop.

🛰️ How are commodity crops detected ? 

Monitor API uses machine learning to identify the crop growing in a field, leveraging satellite imagery and historical data. By analyzing multi-spectral Sentinel-2 satellite data from throughout the season, the model can detect crops and predict the crop type. CropID is trained to predict the crop near the end of the commodity growing season. This means that the model can detect the crop type near harvest even if the crop is still in the ground, or any time after. 

Trained on datasets from government reports and farmer-submitted data, the model recognizes distinct reflectance patterns for different crops. Commodity crop type detection is the core of all Monitor practice detection, as tillage and cover crop practices are determined based on the commodity.

Conflict detection summary table

Some programs may choose to verify the type of commodity crop that was planted during the intervention year, in addition to the tillage and cover cropping practice. You may want to consider including a commodity crop check if:

• Your program pays farmers for a specific type of commodity

• An eligible intervention for your program is diversifying the crop rotation

• Your program’s protocol requires it

Producer-reported commodity crop during the program reporting period are compared with Monitor API observed commodity crop. A prioritization framework is applied to categorize conflicts based on their impact, severity, and confidence level. 

Step by Step process to flag conflicts

1. Align farmer-reported crop(s) with Monitor’s crop determination.

Tips to help align farmer data with Monitor data:

• It can be helpful to assign crops to a year. We recommend using the year of the harvest date.

2. Identify fields where Monitor had low confidence in the crop type determination.

Monitor may not be able to predict the main commodity crop on a field when there is not enough high-quality data available to make a determination (ex: high frequency of cloud cover). This can show up in the Monitor field results in two ways:

1. Main crop confidence is low. Confidence is reported on a scale of 0-100. Regrow recommends using Monitor’s crop prediction when confidence is 75 or higher.

2. There is no commodity crop reported for a cultivation cycle. In this case, Monitor will report no data, indicating that there was not enough remote sensing data available to make a determination.

In these cases, you may choose to use an alternative method of practice verification.

3. Flag practices to review

Background on Monitor crop detection:

Monitor crop detection leverages a Machine Learning model to evaluate remote sensing imagery over a field and predict the crop type. There are cases where the Monitor model 'confuses' similar crops. This can happen when for crops that have a similar growing pattern or spectral signature, making them appear similar or the same via remote sensing.

We refer to these as ‘common confusions’ with Monitor’s ML Model, and we don’t typically consider them a conflict. Here’s a general guide of common confusions that don’t indicate conflict.

• Winter cereals: winter wheat, barley, oats, and rye are often confused with each other.

  • If Monitor reports one of these and the farmer reports a winter cereal, you can consider that verified.

• Root vegetables: Potatoes and sugar-beets are often confused, due to their similarities.

  • If the farmer reports a root vegetable, and one of these two crops is provided from Monitor, consider the crop verified.

Pro tip: API customers should work with an agronomist to determine if there are other crops in your program that would fit into this category.

Decision tree for flagging practices: 

1. First, identify fields where Monitor and the farmer agree. If the farmer-reported crop matches the Monitor detected crop, the practice can be considered verified.
   
   Pro tip: It can be helpful to review the crop types in your program first to identify potential duplicates, or crops that may have different names, spellings, etc. but represent the same thing (ex: soy vs. soybean vs, soybeans). Additionally, crops that are used for different purposes but are the same crop fall into this category too (ex: corn, sweet corn, corn silage).
   
   
2. Check for 'common confusion'. If the Monitor-detected crop is a 'common confusion' with the farmer reported crop type, the practice can be considered verified.
   
   
3. If there is a disagreement between the farmer-reported practice and Monitor:

   1. Determine if the two crops fall into the same season. For example, if the farmer reports corn and Monitor reports winter wheat during the same year, the crops fall in different seasons. This scenario represents a big enough discrepancy that we recommend contacting the farmer to clarify the field practices. ❌

   2. If the crops are in the same season, and are not similar types of crops, then it’s considered a conflict. If your program includes crop rotation as an intervention practice, you may need to consider an additional method of verification. If you are verifying crop type for non-crediting reasons, you may want to reach out to the farmer to clarify the reported practices. ❌