Improving N management in processing carrots using drone-based remote sensing for more sustainable production

Progress report for GNC19-283

Project Type: Graduate Student
Funds awarded in 2019: $14,563.00
Projected End Date: 09/01/2021
Grant Recipients: Michigan State University; Michigan State University
Region: North Central
State: Michigan
Graduate Student:
Faculty Advisor:
Dr. Zachary Hayden
Horticulture, Michigan State University
Expand All

Project Information


Effective nitrogen management is an area of major concern to all farmers, and its impacts extend far beyond final yield. Excess nitrogen fertilization can cause losses in quality and even yield for a variety of crops, and N leaching is associated with environmental damage and contamination. Still, many farmers tend to over-apply to avoid yield losses. Both the environmental and plant nutrition problems are exacerbated in sandy soils, where leaching can occur rapidly. Processing carrots are grown in such soils in the North Central region, and due to their long season many growers apply three or more topdresses during the summer and fall. The proper timing of these applications is uncertain. Some carrot growers have adopted petiole nitrate sampling as a means of timing these applications, applying N when the sampled nitrate values fall below a threshold. This method is labor intensive, however, and has poor spatial and temporal resolution relative to what is required for decision-making. Further, the thresholds at which fertilization should occur are not well-established. Remote sensing using drones comes with the advantages of being relatively low-cost and rapid to implement, yielding spatially continuous image data which can be collected on-demand. The objective of this study was to evaluate the impact of topdress application timing on carrot yield, and to compare the cost, accuracy, and reliability of drone-based remote sensing methods to petiole nitrate sampling. A two-year on-farm trial (2019-2020) was established with a grower collaborator near Hart, MI to complete this objective. Treatments included a starter-only control and a ramp of season-total N rates (67, 135, and 202 kg ha-1 N) applied as starter fertilizer plus topdressed urea, as well as two treatments in which the recommended rate (135 kg ha-1 N) was applied on a split schedule offset from the other treatments by two weeks (early, late). Three additional treatments had the topdresses applied entirely at the first topdress date rather than splitting. These treatments allowed analysis of the impact of rate and timing on carrot yield and the evolution of petiole nitrate and carrot imagery over the course of the season. Results are planned to be communicated through publications in peer-reviewed journals, and have been communicated through presentations at scientific conferences and grower meetings. Outcomes will include an improved understanding of carrot nutrient management, as well as drone technology and its potential uses for vegetable farmers in the region.

Project Objectives:

The main goal of this study was to evaluate how drone-based remote sensing technology can be used for fertilizer application timing decision support in processing carrots. Concomitant with this goal was an evaluation of how application timing and rate impact yield. The results of this study can give a realistic picture of the potential of this new technology for carrot nitrogen management, as well as information regarding the importance of timing as a management consideration. The specific outcomes we sought are as follows:

Learning Outcomes

  1. Carrot farmers will have a more complete understanding of the importance of nitrogen topdress application timing as a management consideration, and the potential utility of drone-based decision-support relative to current practices (petiole sampling).
  2. Vegetable farmers and researchers will have both an improved awareness and greater understanding of drone technology and its strengths and limitations.
  3. Farmers and agricultural researchers will be better equipped to critically evaluate scientific or sales publications regarding the use of drones in agriculture.

Action Outcomes

  1. Carrot farmers will improve their nitrogen management practices through more robust decision support relative to current practices.
  2. Some carrot farmers will be able to integrate drones into their scouting and monitoring programs for nitrogen management.
  3. Carrot farmers and researchers will be better able to experiment with drone-based remote sensing for different applications more specific to their needs.


Materials and methods:

Field Trials

From June to October 2019 an extensive field campaign was initiated in an active commercial agricultural field during carrot production in Oceana County, Michigan, USA (43.5552° N, 86.0958° W; “Site 1”). The experiment was repeated at another commercial agricultural field in 2020, located northwest of Site 1 in the adjacent Mason County (43.8267° N, 86.3827° W; “Site 2”).

Both sites were prepared for carrot planting using a custom strip tiller consisting of a shank, bed forming disks, and rotary cultivator. At Site 1, processing carrots of the cultivar Cupar were seeded in 3-row beds on 20 April 2019. Carrot rows were spaced 0.46 m apart within each bed with 1.63 m between bed centers. The same spacing was used at Site 2 with cultivar Canberra seeded on 24 April 2020. A total of 29 kg ha-1 N was applied with the planter as liquid starter fertilizer in both years. Experimental plots 4.88 m (three beds) across and 12.19 m deep were arranged in a randomized complete block design with four replications.

In addition to a low N control which used starter fertilizer only, treatments included a ramp of season-total N rates (67, 135, and 202 kg ha-1 N) corresponding to 50, 100, and 150% of the recommended season-total N rate for processing carrots on mineral soils in Michigan (Warncke et al., 2004). These rates were applied as starter fertilizer plus topdressed urea. Due to projections of a hot and dry summer in 2020, topdresses in that year were applied as urea mixed with Agrotain at the manufacturer-recommended rate of 2.09 L t-1 to mitigate volatilization losses. Topdresses occurred as either a single large frontloaded application or three equally sized split applications. Frontloaded applications occurred entirely on the dates corresponding to the first split applications, with split applications continuing on at roughly 4 wk intervals. Two additional treatments were included that follow Michigan State University’s recommended grower practice (135 kg ha-1 N applied as starter plus split topdress applications) with the exception that split topdress applications were initiated approximately 2 wk earlier and 2 wk later than the other split treatments. The 2019 and 2020 topdressing schedules for these different groups of treatments are shown in Table 2.

Table 2. Timeline of topdress events for different treatment groups.

Data Collection

Site visits occurred at approximately 2 wk intervals to accommodate the fertilizer application schedule and collect plant data. During each site visit, 5-10 carrot plants were collected per plot. The youngest fully-expanded leaf was removed from sampled plants, and the petiole sap was expressed and analyzed for nitrate concentration. To estimate final yield, carrots were harvested shortly before commercial harvest (29 October 2019, 13 October 2020). Roots and shoots were separated and the shoots weighed fresh. Before being weighed, harvested carrots were graded into either marketable or cull categories based on market specifications.

A DJI Phantom 4 Pro quadcopter equipped with a 20-megapixel natural color camera was used to collect red-green-blue (RGB) composite imagery of the experimental sites. A MicaSense RedEdge-MX camera was also affixed to the drone to collect multispectral image data at the 475 nm (blue), 560 nm (green), 668 nm (red), 717 nm (red-edge), and 842 nm (near-infrared) spectral bands.


Vegetation and Sufficiency Indices

The normalized RGB and multispectral bands were used to calculate two vegetation indices (VIs) which are associated with plant N status: the visible Green Leaf Index (GLI) and the multispectral Normalized Difference Vegetation Index (NDVI) (Louhaichi et al., 2001; Rouse et al., 1973). GLI was calculated using only the RGB camera data while NDVI used multispectral camera data in order to assess their relative performance given the differing equipment required. Mean values of these VIs were calculated over the plot areas.

In order to convert plot mean VI values to sufficiency indices (SIs), a reference value was first calculated for each sampling date by averaging the VI values of the plots which received the 202 kg N ha-1, frontloaded topdress treatment. This treatment (the “reference treatment”) was assumed to be the best approximation of a non-N limited reference area due to the high N rate and single topdress required. Plot mean VI values were then effectively normalized by dividing by the reference value for the corresponding date, yielding SI values (GLI-SI; NDVI-SI) for each experimental plot on each sampling date from 20 June through 9 October 2019 and 18 June through 8 October 2019.


Statistical Analysis

Statistical analysis was also performed using R statistical software. The fixed effects of total season N rate and topdress division on defect-free and total yield, shoot fresh weight, root:shoot ratio were evaluated for each year using type 3 ANOVAs. The data evaluated using this model excluded the starter-only treatment as well as the early and late topdress treatments to maintain balanced data. Dunnett’s test was then used to compare the individual treatments within this data subset to the control treatment, selected as the low-N starter-only treatment, to further develop the interpretation of topdress rate and division effects. The fixed effects of split topdress timing on the same response variables were evaluated using a subset of data which included the early and late topdress treatments as well as the split 135 kg ha-1 N rate treatment on the intermediate timing. Replicate was included as a random blocking factor in both models.

In order to make hypothetical topdressing decisions in the same manner as the petiole sap nitrate test, i.e. by comparing sampled values to a minimum threshold, minimum SI thresholds for triggering topdress applications were calculated for the GLI-SI and NDVI-SI. Using simple linear regression equations between total yield and the SIs for each date, SI thresholds were calculated as the SIs corresponding to the mean total yield of the reference treatment in each year. Only highly significant (p < 0.01) regression equations were used to maintain accuracy in yield predictions. The average of these calculated thresholds for each year was then compared to plot SI values, with SI values below the threshold corresponding to a hypothetical topdress event.



Louhaichi, M., Borman, M. M., & Johnson, D. E. (2001). Spatially located platform and aerial photography for documentation of grazing impacts on wheat. Geocarto International, 16(1), 65–70.

MSU Extension. (2011). Petiole sap nitrate guidelines – MSU Extension.

R Core Team. (2020). R: A Language and Environment for Statistical Computing (3.5.1). R Foundation for Statistical Computing.

Rouse, J. W., Haas, R. H., Schell, J. A., & Deering, D. W. (1973). Monitoring the vernal advancement and retrogradation (green wave effect) of natural vegetation.

Warncke, D., Dahl, J., & Zandstra, B. (2004). Nutrient Recommendations for Vegetable Crops in Michigan (E2904). In Michigan State University Extension Bulletin (Issue E2934).


Research results and discussion:

Progress report

Participation Summary
2 Farmers participating in research

Educational & Outreach Activities

1 Curricula, factsheets or educational tools
4 Webinars / talks / presentations
1 Workshop field days

Participation Summary

40 Farmers
25 Ag professionals participated
Education/outreach description:

4 Sept 2019: An on-farm demonstration entitled “Drone Imagery and Carrot Topdress Timing” took place as part of the Oceana Research Tour organized by Dr. Ben Werling, West Michigan Vegetable Educator for MSU Extension. Area growers and other industry professionals were invited to observe the research plots for this experiment, see a drone flight demonstration, and ask questions related to the costs, uses, limitations, and technical aspects of drone-based remote sensing.

19 Feb 2020: An oral presentation entitled “N Topdress Strategies in Processing Carrots” was given at the winter Michigan Carrot Committee meeting. The presentation served as a summary of this project’s first-year results. Several MSU researchers, approximately 10 Michigan carrot growers, and one or two industry professionals attended.

16 Sept 2020: A “virtual field day” video presentation took place as part of the virtual Oceana Research Tour organized by Dr. Ben Werling, West Michigan Vegetable Educator for MSU Extension. Area growers and other industry professionals were invited to view prepared videos about research updates, as well as to participate in Q&A sessions with the researchers involved in each video. First-year results of this project were shared along with remote sensing background information for growers.

10 Nov 2020: A pre-recorded oral presentation entitled “Remote Vs Proximal Monitoring for N Topdress in Processing Carrots” was given at the ASA-CSSA-SSSA (Tri-Societies) 2020 Annual Meeting. The presentation detailed the motivation, methods, and first-year results of this project.

24 Feb 2021: A summary of trial agronomic results along with drone-based remote sensing work was given in pre-recorded video format at the winter Michigan Carrot Committee meeting. Several other MSU researchers attended along with major Michigan carrot growers. The video created for this meeting may also be used for ongoing MSU Extension purposes.

Project Outcomes

Project outcomes:

Progress report

Knowledge Gained:

Progress report

Success stories:

Progress report


Progress report

Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or SARE.