- Innovative analysis is enabling more accurate interpretation of large-scale trials to inform management decisions
- Data acquisition is being enabled through the use of precision ag equipment, satellite imageries, drones, and strategically placed sensors in paddocks together with the latest statistical analysis and big data analytics
- As the analysis develops further potential use cases will become evident to assist decision making
Experiments undertaken on farms, historically, have been conducted under the management of a research expert using small-scale ‘white peg’ trials with rigorous designs and statistical analysis to minimise the impact of within-site variation. Such experiments are designed to obtain unbiased and precise estimates of treatment effects – even if they are comparatively small – and these have been critical for assessing treatments relating to agrichemical inputs or new genetics, for example.
However, results from these small-scale plots do not always transfer well to grower’s practice at a paddock or farm scale where there may be substantial spatial variability in crop yields due to unknown or uncontrollable environmental effects (such as soil type).
Subsequently, growers may find insights from small scale research experimentation using small plot designs difficult to put into practice.
A solution has presented itself through the widespread adoption of yield monitoring equipment and variable rate machinery which provides an opportunity to undertake large scale or strip trials with growers’ equipment. Most farmers have been doing this for a long time as they assess new management options prior to adopting it further.
“The challenge is to make these strip trials a useful, robust and profitable part of grower’s decision making,” says Professor Mark Gibberd, Director of the Centre for Crop and Disease Management (CCDM), Curtin University, who is leading activities to improve the utility of these on farm experiments.
“Underpinning the drive to improve utility of these large-scale grower trials is the development of new generations of novel analytical methods which take into account paddock variability and can be visualised at paddock scale.”
Two concurrent projects have been run within the CCDM. Firstly, through co-investment between the GRDC and Curtin University in the Statistics for the Australian Grains Industry-West (SAGI-West) team, researchers are applying increasingly sophisticated forms of statistics to strip trial analysis.
“From regression analysis, to Bayesian statistics- which applies probabilities to statistical problems- to mixed models and more, we have the potential to move into machine learning for these types of trials” Prof Gibberd says.
“Rather than generating discrete values associated with treatment effects the outputs from the analyses of larger scale trials are commonly in the form of response functions. These response functions can then be applied across landscapes to predict the broader impact of a management practice or intervention.”
To identify the optimal treatment from strips in a large paddock, advanced statistical tools are used together with algorithms to analyse large dataset arising from several precision agriculture technologies, including yield monitoring systems. Data from satellite imageries, drones, and strategically placed sensors within paddocks can be analysed with the yield data to prescribe an optimal treatment map or to produce a spatial map of yield predictions.
“Secondly, we have worked with the Food Agility CRC to develop and implement methods to examine the role of fertilisers and soil amelioration at a paddock scale. This work will help to deliver new methods for farmers, agronomists, grower groups and others to independently analyse their own trial data sets.”
These grower-led trials draw on geographically weighted regression (GWR) analysis to estimate spatially varying response curves and visualising it in the Google Earth Engine.
“We are able to interpret whether the treatment or the spatial variability is driving yield response, as well as any interactions between the two. When we overlay gross margins and investigate return on investment at a paddock scale, this becomes a powerful tool for informing management options for growers and agronomists.”
“Our co-design approach, working with growers and their agronomists has fast-tracked the development of our analytical capability as well as the utility of the research outputs including how the data is visualised and the ability to investigate potential management scenarios. In essence, the results are more meaningful and practical for growers.”
Grower involvement essential
Professor Gibberd says the involvement of growers in these activities is essential in shaping the outcomes to ensure they are relevant and easily applied on farm.
“The team have really had to think hard about the pragmatics of growers’ decision making and consider their perceptions, for example their attitude to risk of certain management practices.”
“The research has focussed on practical and economic significance as growers are ultimately interested in profit and risk. We have been overlaying economic factors like return on investment with the response function to inform decisions.”
“This means growers can make really informed management decisions based on potential economic gain not just the highest yield.”
“It’s really about optimising the system.”
Grower James Heggaton from Kojonup has actively been involved with the research through Western Australian No-tillage Farmers Association.
The research team works with growers like James and their agronomists to develop an on-farm trial using a plan, prepare, test, analyse and then review and repeat approach. This means co-designing a simple trial that will answer a specific management question for the growers. The trial layout (strips) is prepared and implemented by the farmer using their machinery. The trial is reviewed mid-season using analysis of normalized difference vegetation index (NDVI) imagery, which offers an opportunity to reconnect with the trial as a group. Harvest yield data is analysed and then reviewed again as a group. The ability to visualise the analysis (yield and gross margin) “live” using a specialised data visualisation tool allows the group to investigate scenarios ‘what profit/loss would I make if this treatment was applied across the whole paddock or grain prices went up/down by $X/t?’
Whilst James says technological advances are pushing opportunities in agriculture, and trying to keep up with them through adjusted management is difficult.
“Simple is powerful and being able to visualise my trial results at such scale and understand the impact on economic returns is critical,” James says.
Future work and opportunities
“To begin with we have been focussing on inputs such as fertilisers, but into the future it may be possible to use this form of analysis to unravel plant trait variation on a paddock scale which could provide important information on the value of developing crop varieties with new traits,” says Prof Gibberd.
The team will continue to support grower groups and other researchers in Western Australia and across Australia to analyse on farm trials. Further improvements will be made to the regression methods for modelling paddock variation and they will start making recommendations for how to design paddock-scale trials.
“As we push the frontiers of analyses it opens up more potential use cases and makes new types of experiments possible which is really exciting” says Professor Gibberd.
“The team are becoming world leaders in this type of analysis which is a real bonus for Australian growers.”
More information: Dr Suman Rakshit, email@example.com and Dr Julia Easton, firstname.lastname@example.org, 0403311395