Root image acquisition and determination of root growth rate
Starting two days after seedling transfer to rhizoboxes
(30th July 2021), roots were pictured from the
transparent window of rhizobox with a digital camera (Canon EOS 5D Mark
III) through a 28-mm lens. Pictures were taken on alternate days
starting from 2nd August 2021 until the roots reached
the bottom of the rhizobox (18th August 2021) giving a
total of 9 measurement times in 18 days. All the root pictures were
cropped within ImageJ (Schindelin et al. 2012) to exclude
rhizobox boundaries for further analysis. To detect all the roots of our
images, we trained a convolutional neural network using RootPainter
(Smith et al. 2022). For this purpose, we used RootPainter to
generated a dataset of images by randomly selecting three sub-regions pe
cropped image (width: 861 pixels, height: 897 pixels) and annotate the
roots from this image dataset to improve the model until it was able to
identify most of the roots in our images. After achieving satisfactory
performance of the model, we used it to segment all of our original
images and extract the roots present in them. Segmented pictures were
used to determine the total visible root length (VRL) by using
RhizoVision Explorer using the batch-processing mode (Seethepalliet al. 2021) (Fig. 1a). Visible root length was plotted against
time and the slope of this regression was used as a proxy for root
growth rate (RGR).
\begin{equation}
\text{RGR\ }\left(slope;cm\ \text{day}^{-1}\right)=\frac{\text{Δ\ }Visible\ root\ length\ (cm)}{\text{Δ\ }Time\ (day)}\nonumber \\
\end{equation}Afterward, plants were moved to the greenhouse facility of Leuphana
University of Lüneburg on 19th August 2021 and allowed
to acclimatize to the greenhouse conditions (comparable to climate
chamber conditions: day/night temperature and relative humidity were
22/15.3 °C and 60/73%, respectively) for two days before installing the
root exudation traps.