General recommendations for hyperspectral measurements in bioassays,
greenhouse and field trials
As the sensor is sensitive to even minor changes, which can lead to some sources
of error. Beside the essential and frequent calibration of the sensor with
some kind of white standard and to take into account the warm up period of
the sensor, we have summarised some
essential points from our experiences:
We recommend contact probe measurements on leaf surfaces for greenhouse experiments.
The constant light supply provides a less noisy spectrum and problems with shadowing or light disturbances
are reduced/eliminated. The devices usually
generate a mean spectrum of a small area of the scanned object/leaf. You get the "cleanest"
signal, and the statistics emphasize on the treatment effect and do not focus on technological abnormities.
As the method introduced here requires a certain amount of samples with respect to the number of parameters fitted to
the spectra, we recommend experimental design with a sufficient number of repetitions of each factor level. Most important,
do not underestimate the numbers of repetitions for the control crops or plants, it should be equal weighted compared to the
Repeated measurement on one object, for example leaves of one crop, is also recommended. It is clear, these
measurement are pseudo replicates, but they account for the variance of the individual plant, and leaf differences, which
are not are not treatment related (and that are many).
A sampling equivalent of 13 parameters is needed for the analysis of measurements up to 1100 nm, but taking into account the domains up to
2500 nm 36 parameters and therefore a thrice sample size is required compared to the NIR domain only. To detect treatment differences spectra up to
1100 nm are sufficient in most cases. This is based on our current experiences.
We recommend an easily divisible number of repetition on one crop for high-throughput screenings (ex.: 2,5,10, aso.),
as the likelihood of a false assignment of a certain crop and spectrum increases drastically with increasing complexity.
An important add on for contact probe measurements: Only use closed light systems which do not heat up the plant tissue.
In case of contact with the light source the tissue gets burned and necrotic. The impact is massive, destructive and might
mask the treatment effect.
To avoid false conclusions measurements should be taken on leaves of similar age, as both growth and ageing affect the signal.
Use always the same white standard in one experiment, and only one sensor device. If possible measurements should be done by the same person.
Avoid time effects, in detail, choose an experimental design and size which is measureable within one day. Distribute the samples equally to
minimise errors by diurnal variations. In case the size of the experiment is not measureable in one day, again, all factors and samples have
to be distributed with respect to the time face of the measurements. If not, time effects are depicted by the signatures and not the
intended treatment effects.