# Table 1 Proxy traits proposed here

Proposed trait Description Nearest traditional equivalent
Centroid The weighted centre of mass of the root system Centre of mass of all root pixels
Mass A normalised sum of all likelihoods generated by L in a given image or well Sum of all root pixels
Width/depth (M) Bounding box width and depth of the root system. Calculated as maximum point of mass on the extremities of the root system. We define this as: $$\arg \max_{x} (L(x,y) \cdot x )$$ and similarly for y and the other sides of the bounding box Maximum width and depth reached by all root pixels
Width/depth (p95) Alternative bounding box width and depth of the root system. Calculated to enclose 95% of the calculated root likelihood Maximum width and depth reached by all root pixels, discounting a small number of root outliers
Depth (p99) Alternative depth measurement, calculated as 99% of the root likelihood Maximum depth reached by all root pixels, discounting less outliers than p95
Quadrant mass The mass trait split horizontally into four regions for each well, giving a measure of root material within each quadrant (see Fig. 2a) Root pixel count in four regions (at varying depths)
Orientation Ten brackets of orientation representing the direction of the root system at each pixel. These range from 0°, horizontal, to 90°, vertical. (see Additional file 1: Figure S1) Histogram of all root angles
Quadrant orientation The orientation trait split horizontally into four regions for each well (as previously). Orientations are now grouped into four brackets per quadrant, rather than 10, giving 16 values for each well in total Histogram of root angles at different depths
Leaf hue The average hue for each leaf pixel in the top image, for each well Average pixel hue i.e. leaf colour
Leaf area Total pixel count for all non-white pixels in the top image, per well. Non-white is defined as having a saturation value above a low threshold of 20% Pixel count of leaf area
1. Where appropriate, a nearest traditional measure is listed. L refers to the root likelihood function (Eq. 2). Note the final two traits are measured from a top-down camera view