In this paper, we review learnings gained from early On-Farm Experiments (OFE) conducted in the broadacre Australian grain industry from the 1990s to the present day. Although the initiative was originally centered around the possibilities of new data and analytics in precision agriculture, we discovered that OFEs could represent a platform for engaging farmers around digital technologies and innovation. Insight from interacting closely with farmers and advisors leads us to argue for a change in the ways we approach OFE research. Acknowledging that conditions have changed and drawing from business and social sciences, we suggest that OFE approaches today should develop aspects related to skill development, value generation and value sharing, the social dimension of change, and a renewed focus on farmer-centric research to better bridge industry requirements and scientist inputs.

Agronomic researchers have recently begun running large-scale, on-farm field trials that employ new technologies that enable us to conduct hundreds of farm trials all over the world and, by extension, rigorous quantitative and data-centered analysis.  The large-scale, on-farm trials follow traditional small-plot trials where the fields are divided into plots, and different treatments are randomly assigned to each plot. Over the past two years, researchers have been designing trials with plots approximately 90 m in length, following recommendations provided in the agricultural engineering literature in the 1990s. However, in this type of research, smaller plots are preferred for the benefit of more repetitions of the treatments on one field. This is important because advice given to producers is based on the experimental results, and such advice would be of greater value with more repetitions. With the minimum width of experimental plots being fixed due to the size of the farming equipment, the purpose of this research is to investigate optimal plot length. While shorter plot length in the experimental design results in additional plots, there is a tradeoff between the richness of the data from a single plot and the number of plots possible in a field. In order to weigh the tradeoff between the richness of the data from one repetition and the number of repetitions possible in one field with different plot length, Monte-Carlo simulations are conducted to compare the Economic Optimum Rates of fertilization (EOR) derived from the estimated yield on the experiment field with variable plot length and determine the optimal plot length.

On-farm research has been traditionally used to provide local, field-scale information about agronomic practices. Farmers tend to have more confidence in on-farm research results because they are perceived to be more relevant to their farm operations compared to small plot research results. In recent years, more farmers have been conducting on-farm studies to help evaluate practices and input decisions.  Recent advances in precision agriculture technologies have stream-lined the on-farm research process, allowing data to be collected and analyzed on a sub-field level. By aggregating this data into large on-farm research datasets, it can be used to mine valuable agronomic information regardless of productivity level variations in the field. Challenges exist when determining what data should be collected and how it is aggregated, managed, analyzed, and shared. However, once standardized, this data could be used to create or improve current decision-making tools and processes. 

eFields is an on-farm research network that focuses on building local knowledge for Ohio producers. In 2017, two standardized research protocols were replicated at 22 locations across 11 counties in Ohio. Yield data collected from these locations were combined with site-specific information about agronomic management practices and publicly available data layers in order to classify the results by potential yield influencing factors. This pilot study made it possible to explore the amount, types, and quality of data that is necessary to accurately aggregate on-farm research results. Timely recording of field and crop notes is often overlooked in-season; this makes the accurate classification of results more challenging. This year’s testing provided insight on the need for data collection and management strategies that optimize the transfer and sharing of agronomic data. Inconsistencies in technology adoption and understanding from farm-to-farm resulted in the need for personal contact to obtain data manually. Looking forward, a strategy for the 2018 season has been developed to improve the collection, aggregation, analysis, and reporting of future results. 

Implementing better management practices in corn and soybeans that increase profitability and reduce pollution caused by the practices requires large numbers of field-scale, replicated trials. Numerous complex and often unmeasurable interactions among the environment, genetics and management at the field scale require large numbers of trials completed at the field scale in a systematic and uniform manner to enable calculation of probabilities that a practice will be an improvement compared with current practices. Farmers have been completing such trials in large numbers since yield monitors on combines became widely used in the early 2000s. Upwards of 3000 field-scale trials have been completed on corn and 1500 on soybeans in the US, with the likelihood of many more trials completed in other countries with these crops and other crops like wheat. Creating a curated, anonymized data repository for these types of data that is available for use by the agricultural community, especially farmers, farm advisors and scientists, would enable farmers to grow more food at lower cost with much reduced pollution. This paper contains three parts, all in draft form: 1) an outline of a proposal to establish a Field-Scale Trial Data Repository (FSTDR, pronounced faster) for results from replicated field-scale trials harvested by combines with calibrated yield monitors, 2) a set of guidelines for the privacy and security of the farmer data, and 3) a list of the minimum data collection requirements for trials to be included in FSTDR. We welcome edits and comments to improve this paper. We also welcome collaboration with anyone who is interested to join the Working Group entitled “On-Farm Data Sharing” (https://www.rd-alliance.org/groups/farm-data-sharing-ofds-wg) in the Research Data Alliance organization that has the goal of providing a venue to make these types of data available. 

On-farm experiments are used to evaluate a wide variety of products ranging from pesticide and fertilizer rates to the installation of tile drainage. The experimental design for these experiments is usually replicated strip trials.  Replication of strip trials is used to estimate experimental error, which is the basis for judging statistical significance of treatment effects. Another consideration for using strip trials is greater within-field variability than smaller fields used for small-plot research. Data from strip trials also differ in their statistical properties from data collected in small-plot trials.  The larger scale experiments often result in thousands of correlated observations, violating the assumption of independence. When conducting on-farm experiments, it is not always possible to replicate strip trials. Since there is no replication, there is no estimate of error and it is not possible to determine statistically whether or not there is a real treatment effect.  Spatial statistical modeling is used to describe spatial heterogeneity and the relationship of neighboring observations.  In spatial analysis, characteristics of neighboring observations are incorporated into the statistical model.  A method is proposed to exploit the underlying spatial relationship of a strip trial with adjacent control strips to estimate a treatment effect for unreplicated strip trials and improve management decisions. Specifically, ordinary kriging based on a spherical semivariogram model with nugget is used to predict yield for treatment points.  These predicted yields are matched with the corresponding observed yields for the treatment points and a paired   test is performed.  The proposed method is applicable to many on-farm experiments to improve the quality and value of farmer’s site specific management decisions.

Nitrogen (N) fertilisation affects both rice yield and quality. In order to improve grain yield while limiting N losses, providing N fertilisers during the critical growth stages is essential. NDRE is considered a reliable crop N status indicator, suitable to drive topdressing N fertilisation in rice. A multi-year experiment on different rice varieties (Gladio, Centauro, and Carnaroli) was conducted between 2011 and 2017 in Castello d’Agogna (PV), northwest Italy, with the aim of i) establishing the best N fertilisation management to maximise crop yield, iii) defining a statistical method to obtain calibration functions establishing the NPI as a function of NDRE measured at PI, iv) comparing the best N management based on NDRE readings. A statistical model was developed to define dose/response curve. Results suggested different fertilisation strategies depending on the rice variety. For Gladio variety, grain yield is optimised equally splitting total N amount between pre-sowing plus tillering and PI application. Conversely, considering Centauro and Carnaroli varieties, best N management suggest reducing N application at the early growth stages, increasing topdressing N fertilisation. The algorithm also allowed obtaining calibration functions for each variety, establishing NPI as a function of NDRE values measured just before N application, with the aim of maximising grain yield. The obtained calibration functions can be implemented in variable rate application fertiliser spreader, providing an interpretative model capable of translating vigour maps into N fertiliser applications. However, calibration functions have been obtained only for Gladio, Centauro, and Carnaroli varieties, under specific environmental conditions. Further extension to other rice varieties and agro-environments is needed, and can be performed using the methodology proposed in this study, with the aim of promoting a widespread application of precision N fertilisation in rice.