Developing proxies through digital phenotyping
Last time I looked at how we can use the information – or datasets – collected through digital phenotyping and introduced the idea of proxy measurements. Here I want to delve into that aspect of the subject in more detail.
Both the Rothamsted Field Scanalyzer and the gantry in CHAP’s Phenotyping and Soil Health Facility at Cranfield University have a laser capability for extracting three-dimensional information. So rather than just having 2D imagery we can have images in 3D. Crucially this means you can measure the size of things without having to have destructive sampling, removing leaves or stems off the plant in order to weigh or measure them. You can measure plants in situ, non-destructively.
Something that Rothamsted Research does a lot is just measuring plant height. Height is no big deal to measure manually, but if you have to go everyday, to around say 1000 different plots or plants, and measure the height of each, this can be quite tedious, but with the laser imagery you can do all that quite routinely and in an automated fashion. Such laser imagery is part of the Scanalyzer systems and is now also being deployed from the drones.
As height is a proxy for growth, using the 3D imagery systems described above, we can measure rates of growth, a critical parameter of health and productivity. This may be with respect to the genetics, agronomic treatments, or pathogen infection. The automated systems allow for multiple measurements and high temporal resolution of growth rates throughout the crop/plant cycle or in response to inputs or stresses. Using the systems described, this may be achieved for large numbers of germplasm, and offers many new opportunities for examining crop performance.
Non-destructive measurements are a feature of most but not all digital phenotyping systems and avoids the need to take leaves or parts of plants into the lab for most routine measurements. However, if we need chemical analysis then usually this would be destructive.
Ideally spectral features might avoid this requirement. In the first instance calibrations are required, which would need destructive sampling with lab-based chemical analysis, but in the long term would facilitate some non-destructive chemical analysis,
The goal would be to find spectral proxies that could tell us something about the chemical composition.
So to take an obvious one: if there is insufficient nitrogen then plants are less green. The more nitrogen you give them the greener they’ll get, so the greenness is potentially a proxy for the nitrogen status.
Unfortunately it’s not quite that simple, there are lots of things that make plants turn yellow, for example if you don’t water them enough or they have pathogen infections. However even this might be resolved with the appropriate spectral signatures.
A major goal is to align spectral signatures with a range of plant parameters, from structural aspects through to chemical composition, and this remains a priority activity.
To read the earlier articles in this series, go to: