The need for trials in crop production is high; new developments as well as promises in breeding, fertilisation, crop protection, technology or production methods require constant, objective testing.
The following applies to such trials: “The special thing about them is that nature itself gives us the answers.” (Quote from Dr Gerhard Hartmann, Head of Department 22 “Regional Field Trials, Variety Testing” at LLG Saxony-Anhalt, Bauernzeitung 24th week 2017, page 15). The challenge is to be able to read and interpret nature’s answers correctly! This is only possible by conducting experiments on a scientific basis.
The questions and hypotheses must be formulated correctly before the answers are given. This is done by defining several test elements (variants, experimental elements) and designing them according to the question. The results (yield, success, impact, etc.) can then be compared.
This is possible with the classic exact, field and plot trial. In order to minimise the influence of the essential factor of soil, the area/plot for a test element is kept small in addition to other measures; the size is usually a few square metres and work is carried out on homogeneous fields wherever possible. This small plot size means that special test technology must be used to conduct the trial.
In contrast, on-farm experiments (OFE) also known as on-farm research (OFR) or Production-Integrated Large Plot Trials (PiG), are usually carried out on whole or partial plots on real farms. This has the advantage that no specialist technology is required, and the trial system and harvesting can be integrated into the normal production process. The test results were thus obtained under largely farm-specific conditions. This means that practical results are obtained that are also meaningful for the farm.
But what is the problem? The principle of exact trials, that all influencing factors that are not to be analysed should be as similar as possible, cannot be adhered to here. The larger the area, the more the factors such as soil, nutrient and water supply, microclimate etc. vary randomly. The classic experimental design rules such as repetition and randomisation, which are intended to prevent the influence of these factors, do not bring the hoped-for success here.

Figure 1: OFR trial design, on the left a non-randomised trial, on the right a design with randomised test elements for comparison

What can you do about it? – Record these very factors and take their influence on the result into account! Incidentally, we call these disturbance variables or explanatory factors. High-resolution geo-referenced data (relief, nutrient, soil quality, soil moisture, biomass, etc. maps) are helpful for this. The effects of these disturbance variables can then be taken into account when analysing the trials. In addition to available digital maps, for example, a digital elevation model can be calculated from the GPS coordinates recorded during yield mapping, from which various other relief parameters can in turn be derived. In the statistical evaluation, the influences of the disturbance variables on the target variable are determined, and in the analysis it is calculated whether there are significant influences on the target variable. Here, too, the statistical evaluation provides a statement as to how high the probability is that the difference between two test elements can be significantly attributed to the factor under investigation (e.g. the level of N fertilisation). In contrast to the normal exact test, however, the statistical analysis is much more complex. Prof Brenning (University of Jena, Chair of Geoinformatics) helped us here. Working together, we designed, created and tested our geostatistical evaluation programme PiGSTAT (Statistical Analysis of Production-Integrated Large Plot Trials with R). This has enabled us to successfully analyse all OFR trials to date on a scientific basis.
In a series of trials, at the suggestion of the trial organiser, the results of our OFR trials were compared with those of exact trials also conducted in the field, and the results of both types of trial were in good agreement.

Figure 2: Map of the standardised residuals of a PiGSTAT evaluation

An example of an analysis step in PiGSTAT is the geostatistical examination of deviations in the results. The map of standardised residuals shows possible spatial clusters of outliers or areas of negative or positive model deviation. In blue areas, which occur here mainly along the edges of the field, the observations are at least two standard deviations below the model prediction. In this case, it would be advisable to remove these marginal areas from the data set if they are artefacts (e.g. deviations in yield mapping when entering the swath).

Rating

Since yield is not the only important factor in trials, whether exact trials or OFR, many other data are recorded during the trial. These can be parameters such as emergence rate, growth height, weed infestation, disease infestation, plant cover, straw cover, straw incorporation into the cultivation horizon, etc. We develop and, where possible, utilise digital processes for this purpose. This not only reduces the amount of work involved in scoring, but also leads to more objective results. The scoring data is assigned to the individual test elements either via plot numbers or via the GPS coordinates of the individual photos.

Figure 3: Automatic plant recognition from digital photos to determine the rate of emergence

For which experimental topics is OFR used in practice?

Varieties

  • Farmers: Variety comparisons
  • Breeders: Introduction of new varieties, qualified demonstration trials to close the gap between federal and state trials and practice, basis for variety recommendations

Crop protection

  • Farmers: comparison of means, timing, etc.
  • Companies: Demo trials, practical testing of new agents or methods

Fertilisation

  • Lime, macro, N, micro > type, time, location, etc.

Soil cultivation/sowing:

  • Processes and devices

Technologies for precision agriculture:

  • Fertilisation > N, lime, macro, micro, timing, location (under root, strips, etc.)
  • Sowing > cereals, rape, maize, develop operational control curves
  • Tillage
  • Crop protection

Conclusion

Testing is important and the only way to be able to read nature’s answers correctly. Exact trials have and will retain their importance in the experimental system. On Field Research (OFR) is not by definition inaccurate. The variability of influencing factors outside the experimental question can be taken into account by recording disturbance variables and processing them further in the statistical evaluation of the trial. The evaluation of such tests is more complex, but the statements are just as statistically reliable as in the exact test, which in turn requires a great deal of effort to implement.
Whether exact test or PiG, assessments are an important source of additional data during the execution of the test; digital methods help to reduce the workload and objectify the results.

Sources (german):