Harvest of change: Using drones and robots to select new plant varieties

Admittedly, a lot of very hard work goes into such a tour, and the glasses are Virtual Reality (VR) glasses which take you into a virtual world with technology that simulates physical conditions.

This virtual reality may well be a reconstruction of something real. As a concert or, in this case, a field.

Drones and robots taking over the fields
In a new research project in plant breeding at Norwegian University of Life Sciences (NMBU)  the scientists are adopting new technology to optimise the development of new plant varieties. New plant varieties are necessary to increase crop yield, and to improve adaptation to climate change.

Traditionally, plant breeding of this type has proven to be time and resource-intensive, and huge amounts of data have been recorded and registered manually. For more than 100 years traits like disease resistances, plant heights, dates for heading and maturation have been assessed using human observation and elegantly recorded using pen and paper.

In the new project, drones and a specially-designed version of the NMBU-developed agricultural robot Thorvald collect data using hyperspectral cameras and image analysis to automate measurements done in the field trials.

"By taking a series of photos in sequence we can create a three-dimensional model and recreate the field virtually. It is a bit like Google Maps," says NMBU-scientist and project manager Morten Lillemo.

Thorvald rolls along close to the plants and takes close-up images which are then connected to the three-dimensional image model created from the drone images taken from higher above the field.

"In doing so we know exactly where in the field we are when looking at the close-ups, and we can, for example, determine the number of ears per square metre," he explains.

The images are given additional information which enables the scientists to gather exactly the data they need. This may, for example, be information about performance in previous years, the plants' properties concerning resistance to disease or results from field tests at other locations, because testing of new crop varieties is being conducted across several locations in various environments and growing conditions.

Much of the research work is based on creating models which, in time, will take over and automate the data collection.

The result will be a virtual field which the scientists can visit and examine in depth without leaving their office chairs, using a pair of VR glasses and a computer.

Seeing the invisible
In order to create this virtual field the scientists use multispectral and hyperspectral cameras to take photos. The advanced imagery gives the scientists access to information that the naked eye cannot see. 

In a normal camera, there are three layers of colours; Red, Green and Blue (RGB) that are collected in every pixel or point, which together make the fields of an image, which is composed of the colours that the eye can recognise.

A hyperspectral camera can take images that show both the visible and the invisible light, and there can be hundreds of datapoints in each pixel.

"The human eye cannot see it, but reflected light from the chlorophyll and the green hues in the foliage indicate the plant's condition. The information that lies in the near-infrared area holds the key to the physiological status of the plant, for example, if the plants are stressed or sick we see a drop in the reflectance in these wavelengths," says Lillemo.

Would like to have known more
It is an accepted problem among plant scientists, that from time to time they find themselves wishing they had looked more closely at something in the field trials of previous generations. Up to now they have only had measurements and notes to refer to.

"From time to time the research results reveal something that we would have liked to have studied closer in previous generations. The high-resolution images that create the three-dimensional model facilitate a virtual tour of the field, in the past," says Lillemo.

The process of analysing data from the images from the field studies is called phenotyping. A phenotype is the properties that can be observed in an individual. The phenotype shows the interaction between hereditary properties and environmental factors affecting the individual.

Examining the genes
However, things do not occur only on the visual plane. Genomic selection has also made inroads into plant breeding. Genomic selection is based on all the hereditary material in the plant, termed the genome, being divided into small parts or markers. 

The hereditary material of wheat has been mapped and is used as a basic standard in breeding research, and following careful mapping the scientists know a lot about which markers affect what properties in the plant.

Using genomic prediction, the scientists can see all the markers in the genome simultaneously, whereas previously they were only able to see small parts of the hereditary material at a time.

Although plant breeders are skilled and well trained at picking out the right plants, the genomic information helps the plant breeder to make even better and more precise choices. The genome analyses reveal whether or not the plant has the desired combination of genes. 

"In order to improve the genomic selection models, we plan to combine phenotype data from the image analyses with the markers in the genomic selection models. Phenotype data will provide additional information and enables us to improve the selection models when selecting plants. We also gather data about relationships in the model," explains Lillemo - a hair's breadth from falling into scientific jargon.

How to make a new variety
Wheat breeder Jon Arne Dieseth at Graminor AS is constantly on the hunt for new and improved wheat varieties, which he develops by crossing existing varieties with bred ones. Graminor is responsible for the development of plant varieties for the Norwegian agriculture and horticultural sectors.

 "The starting point is that we have a basic variety that is good, but that has a few traits that we would like to improve. We cross it with another variety that is good for the traits we want to improve," says Dieseth.

Wheat, oats and barley are self-pollinating, whilst rye is cross-pollinating. The hereditary material of cereals is not a simple matter, by the way, as the genome of the wheat is a mix of three grass species and contains much more information than the human genome.

Fully conditional gene combinations
When scientists cross two varieties, they produce offspring with new genetic combinations, and the offspring can have all possible combinations of the genes from the parent plants. The work on improving the variety has thus only just begun.

"It's all about finding the offspring with the desired genetic combinations, which can be likened to finding the proverbial needle in a haystack. We need to find offspring that combine the good properties of the parents, and that are as close as possible to the ideal variety we want to cultivate," explains Dieseth.

The crossing itself takes place in a greenhouse during the winter, and it is the plant breeder who gets to play with Amor's arrows by opening the florets of the maternal variety and adding pollen from the paternal variety in the crossing phase.

"In self-pollinating plants such as barley, oats and wheat, all the individuals in the first generation from a similar cross-pollination will be genetically identical, but in the next generation the fun starts with the segregation of various genetic variants," enthuses Lillemo.

Grain from the cross-pollination is sown out in the field with more than 1000 plants in one rectangular test plot. The breeder selects the best plants for the next generation, with offspring from each selected plant being allocated its own small test plot.

"The process is repeated for five to six generations until we reach genetically stable breeding lines that can be trialled on larger test plots, and gradually with more replicates and across multiple locations," explains Dieseth.

Since these varieties are naturally self-pollinating, the next generations can be grown out in the fields and produce genetically identical offsprings.

Drones and robots help with mapping

During the growing season, the plant breeder collects information that will help determine which plants will be used in the next generation.

It is here that the drones, Thorvald the robot and genome analyses come in. The first field missions with the drones and Thorvald are used to map the test fields.

"We store the reference points in the field with the help of GPS positioning. Once the pattern or mosaic of the plant fields has been laid and we have checked that the images correlate with the pattern,  much of our work is done. Then we can fly the drones and let Thorvald loose on the fields maybe once per week during the growing season," Lillemo says.

There is a lot of research going on around the globe, which is similar to Lillemo's. Whilst much of the research is currently at the development stage, it provides the basis for major improvement and optimising the extensive breeding effort to create new varieties of traditional plants.

Drones, robots and hyperspectral image analysis are already in use within several areas of research at Ås, but this is only the start of an exciting technological leap forward in the research fields.


Published 26. October 2017 - 13:40 - Updated 8. September 2020 - 15:02