Phenotyping robot takes to the field

Global warming and rising demand for food, coupled with global population growth, are among the key challenges of the 21st century. In this context, the ability to breed new, adapted varieties of food crops is a crucial asset. The budding research field of digital phenotyping is devoted to non-destructive trait detection to enable selective breeding of food crops.

 

At a leisurely pace, the field robot makes its way back and forth across the densely planted wheat field. To a casual observer, the machine’s outlandish appearance might seem more suited to the exploration of distant planets. The contraption, designed to cause as little damage as possible to the delicate crops as it traverses the field, is equipped with a collection of sensors that literally watch the plants in the field grow.

With prototypes like this one, researchers at Fraunhofer IIS and industry partners such as the companies PhenoKey, Strube, and Saatzucht Streng & Engelen, have taken an important step forward in the field of phenotyping research: »This approach enables us to closely observe plant growth in its natural environment, providing us with objective and undistorted data,« says Dr. Stefan Gerth, head of the »Innovative System Design« group at the Fraunhofer Development Center X-Ray Technology EZRT.

People are not robots#

The foremost criterion by which varieties are selected for crop breeding purposes is the yield of a given species. To date, modern varieties have been selected by breeding experts primarily on the basis of visually detectable above-ground traits. However, this method suffers from limited accuracy and reproducibility due to the subjective nature of the experts’ observations and assessment.

In wheat breeding, for instance, breeders typically select for stalk height and number of ears. While these traits do correlate with a variety’s yield, they are only used because there is no technological solution so far that allows biomass to be accurately determined using non-destructive methods. Accordingly, biomass is measured only indirectly by measuring the correlated traits.

In sugar beet breeding, meanwhile, the focus of phenotyping research is on entirely different plant traits. Here, breeders aim for uniform crop emergence to ensure a harvest of sugar beets of roughly the same size. Leaf area, shape, and orientation continue to provide an indication of growth patterns and sugar content.

Multimodality improves data#

A section of the test field planted with wheat is divided into plots between 1.25 and 1.5 meters wide. Comparability of the individual measurements currently depends mainly on optical volumetric measurements using technologies such as LiDAR and laser line scanning. »However, depending on the density of the section – i.e. the number of stalks and ears in the plot – the relationship between biomass, that is yield, and volume is very imprecise for both optical methods. Optical methods can – in much the same way as a breeding expert – determine
stalk height,« explains Gerth. »In conjunction with the distance traveled, this information can be used to calculate the volume of a plot. But as soon as the leaf canopy closes, there is no information on the crop’s actual density. The leaf cover also means that the quality of the ears can no longer be visually assessed.«

This is where X-ray technology comes into play: We can use X-rays to uncover areas concealed by the leaf canopy. And that is not all: Using X-ray technology, we can measure the exact biomass of the stalks and ears. Combined with the optical image data, this provides highly relevant data for the selection of varieties.

3D modeling for precise leaf analysis#

Digital phenotyping of plants in test fields has become an indispensable tool for breeders in sugar beet breeding, too. As the beet itself is hidden underground, the focus is on the plant leaves. The task of these »solar collectors« is to absorb sunlight as efficiently as possible, harnessing it for sugar production, among other purposes. By contrast, excessive exposure to sunlight is stressful to the plant. Accordingly, leaf shape and orientation are crucial factors when it comes to assessing varieties.

»In order to determine the area and shape of a leaf as precisely as possible, three-dimensional scanning is indispensable,« explains Oliver Scholz, group manager in the Contactless Test and Measuring Systems department at the EZRT. »Depending on the environmental conditions, we use various methods tailored to the situation at hand to provide optimal data for subsequent analysis.«

The resulting three-dimensional data on each individual leaf provide the basis for a digital model of the leaf that captures the characteristics of relevance to breeders. In this process, hundreds of thousands of data points per leaf are compressed into a handful of parameters that nevertheless accurately characterize the leaf. This is comparable to mp3 encoding of audio data, another process in which a very large volume of data is distilled to its essence. The leaf descriptions obtained in this way provide breeders with detailed insights into how their varieties grow, giving them a sound basis for subsequent breeding decisions.

Human-machine collaboration#

The key benefit of phenotyping based on non-destructive measurements performed by inspection robots is the objectivity of the digital measurements and reproducibility of the results obtained. The robot can be thought of as a tool that supports human decision-making by giving breeders a precise data evidence basis on which to plan the breeding process. We are nevertheless working toward a higher degree of automation: In the next few years, we hope to develop field robots that are able to autonomously navigate crop fields, scan the plants they find there, and evaluate the resulting data.