The adoption of precision agriculture technologies continues to grow on U.S. farms. Auto-guidance, variable-rate technology (VRT), and yield monitors have become standard options on modern farm machinery (i.e., tractors, sprayers, and combines). These technologies have enabled farmers to conduct on-farm research more easily. Common on-farm studies include evaluating input rates (i.e., fertilizer and seed) and management practices such as tillage. In these studies farmers are interested in understanding their impact on crop yield and profitability. Results from on-farm research have been used to improve input management and profitability. However, this type of research requires proper design of the study and quality data collection to obtain accurate results. Further, those involved in the research need to have a plan for storing collected data, backing it up, and sharing it. Understanding the following steps for conducting successful on-farm research is essential:
- Design the study
- Conduct the field study
- Analyze data
- Share/publish results
This fact sheet outlines the steps for designing and conducting on-farm studies. In addition, we explain farm management information system (FMIS) software’s role and use to support the design and analysis of on-farm studies (Table 1). Additional information about conducting on-farm research is available in the 2017 Ohio Agronomy Guide published by the Ohio State University. (Barker et al. 2017)
1. Design the Study
The first step in conducting on-farm research is designing the study (Figure 1). Designing an on-farm study begins with developing a concise and testable research question, such as “How do different downforce settings impact crop emergence and final yield?” or “What is the impact of foliar treatment compared to no treatment?” To ensure the research question can be tested effectively, available equipment, equipment size, technology control capabilities, field history and variation, and a determination of the data that needs to be collected, must be considered.
For instance, if testing different fertilizer rates within a pass or area, the time it takes to make a rate change must be considered to ensure ample distance exists for the rate treatment. Not considering the rate change time can cause the treatment area to receive the incorrect treatment rate if the plot length is not long enough to allow the technology to make a rate change to the target treatment rate. The time to make a rate change can be as long as three seconds or longer for some application technology. Understanding equipment response time helps ensure the correct rates are applied to each treatment area.
On-farm study layouts commonly used include grids, blocks, and strips of varying sizes (Figure 2). Most likely, the type of design used is determined by the person supervising the study and the software being used to support the study. Strip trials are the simplest design to implement and manage in the field without precision ag technology. The grid and block designs require a prescription (Rx) map with included treatments plus precision ag technology to install the treatments. In return, yield monitor data must be assigned to the study area to evaluate crop performance by treatment.
In the grid design, treatments are most often assigned to an individual grid cell through a randomization process (Figure 2a). This approach limits the area of the field used for the study. An Rx map is generated and is used to implement the study. If using test blocks, the user creates the management zones and positions the blocks within the proper zone. Figure 2b represents a seeding rate prescription with three blocks positioned across a field. Each block contains four different populations and one test per seeding rate zone. In this example, the final prescription map contains both the seeding rates and the blocks integrated, requiring a planter equipped with VRT. Strip trials are the simplest and do not require Rx maps (Figure 2c), although an Rx map could be used. The operator usually uses a printed map or study layout of the field. This map or layout is used as a reference for when to change rates or products manually if an Rx map is not created.
|Figure 2a||Figure 2b||Figure 2c|
|Common types of on-farm study designs include dividing a field into (a) equal-sized grids, (b) using blocks positioned within a zone, or (c) strip-trials. Graphics by The Ohio State University.|
After selecting a study layout, consider if buffer rows are appropriate. Installing buffer rows between treatments reduces the possibility of bleed over, drift, or contamination into other rows. For example, pest management trials typically require buffer rows where data is collected from only the middle rows to combat fast moving or reproducing pests.
Replications and Randomization
Replication allows for the same treatment to be tested multiple times. This helps to ensure any differences observed are caused by the treatment rather than field conditions such as low areas, soil type changes, or equipment errors. Five replications are typically recommended for on-farm studies, but it is understandable that field conditions or shape may not allow for five replications. At a minimum, three to four replications that each contain a set of the randomized treatments should be included (Figure 3). Randomization also helps to ensure differences are caused by treatments and reduces the chance of biases from variations within the field. Make sure there is a control treatment that represents the base treatment for comparisons. The control treatment will most likely be the typical management practice used by the farmer.
Software packages can be used to support the design of on-farm research, particularly for layout and for making replication decisions. Today, software packages recommend a range of sizes for grid, block, and strip study layouts. For example, the AgLeader software, SMS Advanced Field Trial Module, suggests a minimum length of 300 ft for strip trials. Other packages implement blocks ranging from 20 ft x 30 ft (Premier Crop Systems Enhanced Learning Blocks) to 0.25-acre sections (Beck’s Hybrids Farmserver). Furthermore, software packages may or may not require replications to be included in the study layout. In the AgLeader SMS Advanced Field Trial Module, up to 50 replications can be selected by the user. Typically, on-farm research does not use nearly that many replications due to field size and resource constraints. It is ultimately the user’s responsibility to select a layout that will work best given the study objectives and available resources. As of today, there are no software packages that automate the process of designing an on-farm study. It is expected that software will continue to be enhanced so a user can provide data layers such as the field boundary, guidance lines, and soil maps, and that the software will automatically create a design within a field, based on the treatments. However, many packages can be used as resources. Eight different software packages are presented in Table 1, highlighting different functionalities.
Table 1 (PDF): Selected farm management information software (FMIS) and their functionality for on-farm research. This information is for illustration purposes only.
2. Conduct the Field Study
After developing a suitable design, the study can then be implemented. Be sure to consider these important issues:
- calibrating equipment and technology for accurate operation and for collecting quality data
- reducing human errors
- taking detailed notes about the field operations’ production practices and crop health throughout the growing season
- collecting the appropriate data to answer the research question (Figure 4)
All equipment and technology should be in good operating condition and calibrated. Calibration should follow directions outlined in the operator manual. Different calibrations could be needed for different crop types, moisture ranges, field conditions, etc. Furthermore, all sensors should be clean, free of debris, and functioning. Failing to calibrate equipment and technology or having sensor malfunctions can result in inaccurate data. Hawkins et al. (2017) outline tips for calibrating grain yield monitors while Luck and Fulton (2014) note practices to ensure accurate yield data collection.
Error is not only associated with equipment and technology; it can be caused by unintentional human actions. For instance, mistakes during implementation of treatments can happen (e.g., a treatment within the specified replication is misplaced compared to the map or a rate was misassigned in the equipment controller). In such cases, it must be noted what happened on the map and adjusted in the layout for future data collection.
Inconsistency and unexpected accidents can occur during the implementation of an on-farm study. To account for this, take detailed notes throughout the study on any potential differences in field conditions, such as unusual weather conditions, potential problematic areas in the field that could negatively impact yield, application dates and rates, etc. Photos are one of the ways to help remember what happened months earlier, and satellite imagery helps to identify issues with crop health. Having detailed notes throughout the growing season and during data collection is critical. It can provide guidance and explanation for questions that arise when evaluating differences between treatments after the data analysis.
Managing the Study Site
In addition to implementing treatments in the correct location, management of production practices must be consistent throughout the study site. Only the variable being used to test treatments should be changed. For example, if a study is testing various nitrogen rates, ensure that the entire field receives the same tillage and seed. Adjusting other factors within the study area can influence results and impact treatment responses. Natural field variation will occur, but applying the same management strategies to the entire study will reduce unintentional variability.
Data collection can occur at different times throughout the growing season and even post-season/harvest. Yield data is the most common type of data used to evaluate treatments. Other data collected might include soil samples, stand counts, tissue samples, or moisture readings. Soil and tissue samples can be sent to a lab for analysis, and then the data can be used to evaluate treatments. If collecting samples in the field, make sure all sample bags and data sheets are clearly labeled and organized. When collecting data from different points across the field, ensure samples are representative and random. It is recommended that samples should be taken from several different areas within the field, and that data should not be gathered from the buffer rows. Creating a standardized sampling protocol helps to reduce errors resulting from biased sampling.
3. Analyze Data
Data analysis is necessary to determine if treatments are significantly different from one another. For example, a treatment may result in a higher yield by a few bushels but when considering field variability, is this improvement a statistically significant response due to treatments? Figure 5 outlines three important steps in the analysis process for on-farm research. Prior to analyzing data, make sure all study data has been saved and organized in a manner that can be easily accessed for analysis. In addition to organizing the data, make sure it is complete. Next, data should be cleaned, if needed, to remove any errors. Then, you can summarize and perform the needed statistical analysis to compare treatments. Chapter 11 of the 2017 Ohio Agronomy Guide is a good resource for understanding statistical analysis for on-farm research. (Barker et al. 2017)
It can be important to clean data, particularly yield data, to ensure it is accurate and can be used for sub-field analyses. A sub-field analysis is either a spatial analysis of an individual field, or zones or areas that are then analyzed individually. Data cleaning can be started by referring to the field notes and then removing any data that is questionable, erroneous, or collected from problematic areas. This process can be completed by manually editing the raw data or using a software package. Most software packages provide some level of data cleaning and statistical capabilities. For example, Corteva has the Yield Editor 2.0.7 tool embedded in some of their software packages. The USDA developed Yield Editor 2.0.7 to aid in the yield data cleaning process. It allows a user to set up filters and use other editing techniques. Setting maximum and minimum values that allow extremely low or high yield values to be removed is one example of a data cleaning filter. In addition, software packages commonly have post-harvest calibration tools that provide important data:
- yield data adjustments of the mass flow calibration curve (load calibration)
- overall harvested grain weight and grain moisture values adjustments within the software, based on the scale ticket values
A scale ticket commonly refers to the ticket noting the certified load weight of grain delivered. This post-harvest calibration helps to ensure that the yield map data accurately reflects the yield across a study site or field. Most software packages provide the ability to select and delete data manually, involving the user in each step. In this manual selection and deletion process, the user must have a good understanding of how yield monitors collect data and the origination of potential errors. Current software packages do not provide preliminary error checks or auto-calibration process capabilities.
Statistical analyses are important to understand if differences that exist between treatments are a result of the treatments tested and not just random chance. Data summary can include calculating the treatment means along with either standard deviation or coefficient of variation to understand variability across replications. Additional analysis can include comparing the means using the least significant differences test to evaluate if differences exist or not. Microsoft Excel provides the ability to not only summarize data by calculating treatment means, but also has statistical tools like ANOVA for data analysis. Mobile apps like the OSU Precision Led On-Farm Trial Support (OSU PLOTS) provide simple statistical analysis and study reports (Figure 6) (Ohio State PLOTS 2017 and 2020). Figure 6 indicates that no significant differences in yield existed at 26,000 and 30,000 seeds per acre treatments even though one yielded five more bushels per acre. Most farm management information systems (FMIS) provide capabilities for creating summaries with a few offering tools for conducting statistical analyses. However, the data summary and statical tools vary by software (Table 1). Premier Crop Systems Enhanced Learning Blocks provides automated, advanced statistical analysis conducted by an in-house statistician analyzing the data uploaded by a user. Lastly, software packages like Ag Leader SMS Advanced can conduct basic yield summaries, analyze yield maps by using different data layers (i.e., soil maps, variety maps, etc.), and perform multi-year yield analyses (Table 1).
4. Share/Publish Results
The final step for a successful on-farm research study is sharing results. Figure 7 outlines three steps for sharing results. This includes summarizing results to create a report with key findings to share with stakeholders. Ensure findings are accurate and supported with quality data. Finally, make sure you have identified the target audience and preferred method for sharing results.
In summary, on-farm research has become a valuable tool for farmers and others to test new ideas, technologies, and strategies. This on-farm work is important as farmers look to remain profitable, search for ways to improve input management, and understand field practices that provide the maximum return on investment for their farm operation. Successful on-farm research studies implement careful planning upfront to answer the question at hand. Poor design leads to flawed data and misinterpretation of results. A variety of FMIS software packages with different features are available to aid in designing and analyzing on-farm research. However, before purchasing software and paying for more than is needed, review the various packages carefully. Understanding these differences will allow these tools to be optimally used in supporting on-farm research studies.
If interested in on-farm research, visit the Ohio State University Digital Agriculture eFields website at digitalag.osu.edu/efields to learn about ongoing on-farm studies and their results. To review different study protocols and tips, visit digitalag.osu.edu/efields/get-involved/study-implementation.
AgLeader SMS Advanced Field Trial Module
Beck’s Hybrids Farmserver
Corteva Granular Agronomy
EFC Systems FieldAlytics
Premier Crop Systems Enhanced Learning Blocks
Fulton, J. P., and Port K. 2018. “Precision Agriculture Basics.” In Precision Agriculture Data Management, edited by D. K. Shannon, D. E. Clay, and N. R. Kitchen, 169–187. New York: John Wiley & Sons, Ltd.
Barker, David J., Steve Culman, Anne Dorrance, John Fulton, Ryan Haden, Edwin Lentz, Alex Lindsey, et al. Ohio Agronomy Guide. Columbus: Ohio State University Extension, 2017.
Hawkins, E., Fulton, J.P., and Port, K. 2017. “Tips for Calibrating Grain Yield Monitors-Maximizing Value of Your Yield Data” (ANR-81). Ohioline, The Ohio State University.
Luck, J., and Fulton, J. P. 2014. “Precision Agriculture—Best Management Practices for Collecting Accurate Yield Data and Avoiding Errors During Harvest.” Nebraska Extension Publications, University of Nebraska-Lincoln Extension.
Ohio State PLOTS, v. 2.1, (The Ohio State University, 2017), Android 4.4 or later. iPhone, iPad, and iPod touch ed., v. 1.1.1 (The Ohio State University, 2020), iOS 8.0 or later.
The authors would like to thank Ramarao Venkatesh (College of Food, Agricultural, and Environmental Sciences, The Ohio State University), John Barker (Ohio State University Extension), Matt Schmerge (Ohio State University Extension), Dr. Ajay Shah (Department of Food, Agricultural and Biological Engineering, The Ohio State University) for taking the time to review this fact sheet and provide constructive feedback. Thanks to Ryanna Tiejie and Kassidy Thompson for helping review and create illustrations.