Skip to content

Utilizing drones, machine learning and artificial intelligence to optimize nitrogen credits from a red clover cover crop

Principal Investigator: Jordan Sinclair

Research Institution: Veritas Farm Management

Timeline: April 2019 – December 2022   


  • Determine how to utilize Remote Piloted Aerial System (RPAS) / Unmanned aerial system (UAS) to measure biomass at the field level and incorporating spatial variability.
  • Determine if clover plant counts and biomass can be categorized into production groups that correlate to consistent nitrogen (N) credits for the following corn crop.
  • Determine if the N credit for the following crop is consistent enough to utilize for variable rate N applications.


  • By understanding how N accumulation correlates to clover count and/or biomass we will be able to inform policy and best practices at a provincial level.
  • The development of machine learning (ML) and artificial intelligence (AI) algorithms that can determine N credits from clover will allow farmers to confidently estimate and fully realize N credits on a case-by-case basis.
  • UAS technology allows for a geospatial attribute to the N credit which can be translated into variable rate N prescriptions, optimizing the 4R philosophy.

Scientific Summary:

For 35+ years double cut red clover (DCRC) has been considered the “gold standard” cover crop in winter wheat. There has been extensive research conducted that demonstrates the benefit of adding DCRC into Ontario crop rotations, including calculations of the N credit DCRC can supply as a legume to the following corn crop. Additionally, research has determined that the amount of time that the DCRC is allowed to grow prior to termination has a strong correlation to the amount of N that can be credited. What is not fully understood is if the value of the N credit should be related to DCRC biomass, the number of DCRC plants, or a combination of these variables and how to effectively incorporate spatial variability of DCRC growth into the credit process. As a result, current practices necessitate a conservative approach when assigning N credits from DCRC.

Over the past several years, Veritas has utilized UAS technology to determine crop characteristics. In 2018, we combined UAS technology with artificial intelligence and machine learning to identify and count weeds and crops at the species level. This drone technology now has the potential to classify clover plants and measure biomass at the zone level within a field. Nitrogen credits can then be geospatially validated using a pre-side dress nitrogen test (PSNT). The results of PSNT can then be used to create a variable rate N prescription for the corn crop. The data generated from the N prescription application equipment and the yield data from combine yield monitors can then be used to validate any response from the N credit.


We have now completed all activities associated with this project. For three years we monitored clover cover crops after wheat harvest. We flew a drone with an RGB camera over each field taking high resolution photos in a grid pattern. These images we manually classified by clover density and used to inform sample location. We then went to each field and manually counted clover plants at each sample site and took a soil sample to measure baseline nitrate levels. Finally, we flew each field with a drone and multispectral camera to obtain an NDVI image of the entire field. This image was used to estimate clover biomass spatially across the field. The following spring, we resampled each field to measure nitrate levels and determine the change over time. We used the clover NDVI data layer as the baseline for a variable rate N application prescription, to which we added check boxes or strips in each application region. Finally, we flew the field with a drone and multispectral camera to get an NDVI image of the corn crop, and after harvest we collected the corn yield data.

What we found was that clover NDVI correlated well with the manual classification of clover cover. Neither of these metrics correlated with clover count, and none (clover count, NDVI or classification) correlated with change in soil nitrate levels. Following the clover zones through corn yields revealed yield stability across reduced N rates in areas that had high clover growth. One limitation of this study was that we did not reduce overall applied N by enough to see a consistent yield drag, and thus we were unable to determine actual N rate values associated with different clover classes. Since we were not offering farmers compensation for potential loss yield, most were unwilling to lower total N too much.

The results of this project provide evidence that a successful catch of red clover after wheat can significantly reduce the amount of N fertilizer required for the following year’s corn crop. The benefits of this include fertilizer cost reduction, which, if synthetic N costs remain high can be significant. The use of an NDVI map that quantifies the biomass of clover across the field allows the grower to map clover quality and capitalize on smaller scale success. Instead of applying N credits only when a whole field has had a successful clover catch, growers can utilize the NDVI map as a base layer for variable rate N application and thus reduce N use anywhere a quality clover catch occurred.  

External Funding Partners:

Veritas Farm Management

Green Eye Technologies

Project Related Publications: