Depth-Resolved Root Analysis and Trait Extraction from Minirhizotrons and Flatbed Scanner • RootScanR

Overview

RootScanR is a package which has been designed to make the analysis of minirhizotron root scans just a little less stupid. The package relies on prior Image segmentation. We recommend using RootDetector or RootPainter. This packages enables the user to map various root features to a continue s depth distributions. The user has the choice to include the tube insertion angle and tube diameter - or ignore it (?). Rotational bias of root prevalence can be tested. Let me know if you have questions, improvements, or found bugs & errors.

# install the package
if (!require("devtools")) {
  install.packages("devtools")
}
devtools::install_github("jcunow/RootScanR")

Calibration

More accurate results require easy to gather in-situ calibration of each tube. Traditionally, the assumption is that all tube have been installed perfectly, i.e., exact same insertion angle, insertion depth, same scanner rotation position, and tape overshoot for each tube. RootScanR offers the possibility to approximate the soil start and rotation center based an assumption of well-taped tubes.

Soil Surface Position Estimation

start.soil = SoilSurfE(img)

Scanner Rotation Position, Rotation Differences Between Two Scans, and Removal of Scan Edges only present in one Scan and not the other.

RotationE(img)
RotShiftDet(img1,img2)
RotCensor(img,center.offset = 0.3,fixed.rotation = FALSE)

# under construction: estimating rotation by root distribution (assuming top-tube bias)
LR_rhythmicity(rotation.cuts, root.px, period= number.rotation.slices)$phase * (dim(roation.cuts)[1]/number.rotations.slices) %>% round()

The core function: Create a depthmap

requires: soil surface estimate, rotation center estimate, tube insertion angle, tube diameter, scan resolution

Phase shifted, trimmed sine depth mapping including tube diameter and tube insertion angle

depthmap = create_depthmap(img,mask, tube.thicc = 7, tilt = 45, dpi = 300, start.soil = 0,center.offset = 0)

Zoning - Create analytical units

Creates discrete sub-units of the whole image. This determines the resolution of the depth distributions. Currently, the package only supports analysis within discrete depth bands. Future releases will tackle response functions on continuous depth.

Depth-wise zoning

binned.map = binning(depthmap, nn = 1,"rounding")
root.zone = zone.fun(img, binned.map, indexD = 5, nn = 1)

Feature Extractions

Parameter from segmented image cut

RootScapeMetrics(root.zone, metrics = c("lsm_c_ca","lsm_c_pland","lsm_c_enn_mn"))
Root.px = count_pixel(root.zone)

Parameter from skeletonized image cut

root.zone.skl = skeletonize_image(root.zone, method = "MAT", layer = 2)
RL = root_length(root.zone.skeleton)

Combine both

root.thickness(Root.px, RL)

Parameter from RGB image cut

Rhizosphere = Halo(root.zone, width = 5, halo.only = FALSE)
soil = rgb.img[Rhizosphere == 1] < - NA
analyze_soil_texture(soil)
soil.color = tube_coloration(soil)

Turnover - either two images from different time points, or the ‘RootDetector’ root tracking output

turnover_tc(root.zone1,root.zone2, method = "kimura",dpi = 300, unit = "cm")
turnover_dpc(root.zone.dpc, product.layer = 2, decay.layer = 1, im.return = F)

Rotational Bias

Rotational zoning, Feature Extraction,

rotation.rootpx = data.frame(kk = 1:12,px = NA)

for(i in rotation.rootpx){
rotation.zone = zone.rotation.fun(img,k = c(i-1,i),kk = 12, mm = c(1500,5000))
rotation.rootpx$px[i,] = count_pixel(rotation.zone)
}

Test Amplitude Difference Can be used to text e.g., differences in top-bottom-tube bias between varying insertion angle or tube diameter To test differences between sinus fits, we make use of the diffCircadian package https://github.com/diffCircadian/diffCircadian