High-throughput research: Maximizing ‘n’ use efficiencyFebruary 20, 2018
My first job in university was as a research intern for a major agribusiness. I quickly developed a passion for science under the guidance of the field biologist leading our team of four, and it was a significant influence on my decision to pursue my MSc. The overarching aim of my thesis research was to investigate how UAV imagery could improve nitrogen use efficiency in corn, which I investigated through intensive plant- and soil-measurements. As much as I love the power of a controlled experiment (where else in crop science do we ever get to make such definitive conclusions, however small or incremental they might be?), plot work involves hours of monotonous, mentally straining visual evaluations. Humans are subjective, and limited in the quality and quantity of the observations they can make. An experiment could have more explanatory power if a biologist adds just one more treatment, but this must be balanced against the time it takes to properly assess its effects. When we move from the world of crop protection and agronomy into plant breeding, these issues are magnified astronomically due to shrinking plot size. The following excerpt highlights the problem for a breeding trial of 5000 lines with 20,000 plots:
“Using single-row, 1m wide by 4m long plots and ignoring the need for walkways or borders, the net row-length would be 80km (roughly 50mi), occupying 8ha (20ac). A person walking 3km/h (2mph) would need about 27h to visually score traits, assuming no stopping. Halting at each plot for 30s …would require an additional 165 h”
-White, et al. 2012. Field-based phenomics for plant genetics research. F. Crop. Res. 133: 101–112.
Some researchers would suggest that these limitations make small plot research passé in the age of ‘Big Data’. Why go through the effort to execute an experiment that is limited in scope when we could ‘mine’ massive datasets of yield, inputs, or soil characteristics for information? For one thing, there needs to be a representative, well-organized and maintained database to even begin such an endeavor. Any exercise without an a priori hypothesis is inherently non-scientific and as susceptible to bias as the humans running the process, for another. We need the empirical results of field research to first show us what to look for rather than casting a digital net and hoping exciting (and real) patterns surface.
One thing that is undeniable, however, is that our results become more accurate as a truly random and representative sample dataset (‘n’) grows. Big Data is already showing promise due to its large sample sizes, but we can also use technology in small-plot research to increase the number of site-year replications or depth of experimentation for a trial. Rapid data collection by UAVs can overcome the labour and time restrictions that reduce the size and scope of agricultural experiments. The challenge so far has been, to paraphrase an old professor, to transform these visual observations into quantifiable information that feeds into statistical analyses.
Rapid data collection by UAVs can overcome the labour and time restrictions that reduce the size and scope of agricultural experiments.
As the demand for imagery in small-plot research grows, industry is responding with tools to generate and extract plot-level data. While agronomists might look at a population estimating algorithm for broadacre use, the same algorithm is equally beneficial to a corn breeder for recording the plant counts of thousands of plots. The Green Fraction analysis used to quantify winter wheat survival could alternatively be used in a time-series to calculate rate of soybean canopy closure. With geospatial tools, these analyses can be applied to small plot trials to produce tabulated data.
This isn’t a total paradigm shift in research–proximal and remote sensing have been investigated for decades–but it does seem to be the first time that academic and industry research programs are pushing to incorporate these data as standard measurements. One of my goals as Research Agronomist is to document workflows so that new users do not face the steep learning curve I did. As research-focused tools become available, and with careful planning and communication between data analysts, technicians in the field, and biologists designing experiments, UAV imagery can enable the expansion of field trials in both treatment count and geographic replication. This is nothing to say of the benefits from freeing up labour to delve deeper into existing experiments through more intensive plant-, microbial-, or soil-level measurements. As commercial farmers become more efficient with their inputs, researchers must also get more out of their limited resources by constantly improving on ‘n’-use efficiency.