*2.2. Spectral Data Collection and Nutrient Processing*

Data collection and processing methods included in situ spectral data collection and ex situ laboratory nutrient analyses. The methodology proceeded in four general phases (Figure 2), which are briefly described here and detailed below. First, canopy spectral measurements and plant samples were collected at the four sites immediately prior to peak crop maturity (seed head stage). Next, plant samples were dried and separated in the laboratory where the grain was imaged and both the plant material and grain were subjected to nutrient analyses. Third, the relationship between the nutrients and spectral components were modeled using PLS regression with both the full spectrum and a waveband selection method. Lastly, the replicability of the PLS regression methods were tested through statistical comparisons of model fits and similarity of results.

**Figure 2.** Methods flowchart detailing data collection and processing methods.

In Phase 1, canopy-level hyperspectral measurements and plant material were collected at each site immediately prior to harvest during peak maturity, which was late June/early July 2016 in the US and October 2017 in Ethiopia. For each of the four sites, 40 random points were generated in ArcGIS, providing 80 possible sampling location in each country. These points were then located in the field using a hand-held GPS unit (Trimble Juno 3B, Corvallis, OR, USA). In some cases, locations could not be accessed due to sampling on the days of harvest and attempting to not interfere with the harvesting practices of the respective farmers, so the actual number of samples is below 40 for each field and below 80 for the region. At all sampling sites, canopy spectral data were captured using the same spectroradiometer (FieldSpec Pro FR: Analytical Spectral Devices [ASD], Boulder, CO, USA), measuring reflectance from 350–2500 nm with a spectral sampling width of 1.4 nm from 350–1000 nm and 2.0 nm from 1000–2500 nm. The spectroradiometer was calibrated using a Spectralon diffuse reference panel (Analytical Spectral Devices [ASD], Boulder, CO, USA) approximately every 15 min. The spectralon was held at a distance from the spectroradiometer fiber to ensure no shadows were measured. The spectroradiometer fiber was held 1.2 m above the ground; since the top of canopy was about 0.3 m above ground, the cone of acceptance was 25◦, resulting in a footprint with a diameter of 0.40 m for each sample. This diameter ensured a sufficient mass of plant/grain matter was collected for nutrient testing (10 g of grain; SSSA, 1990; [37]). Reflectance data were captured between 11:00 a.m. and 2:00 p.m. local time under a cloud-free sky. It is important to mention that the same spectroradiometer was operated by the same individual in all locations to ensure the exact same data collection process was reproduced. Each location was imaged five times in succession, and spectra were averaged. Plant material in the footprint of the imaging fiber was then clipped at the base, stored in plastic bags, and placed on ice in a cooler for transport back to the laboratory.

In the laboratory, samples were dried to remove excess moisture, and the grains were separated from the plant using a traditional method of hand threshing with the assistance of a basket weaved surface. The grains from each sample, which measure approximately 1 mm in length, were aggregated in a petri dish to generate a sufficient amount to fully cover the lens of the imaging equipment. The grains were spectrally imaged in a dark room using a contact probe (Contact Probe: Analytical Spectral Devices [ASD], Boulder, CO, USA) with a halogen light source, while not equivalent to the sun, still emitted spectral wavelengths (350–2500 nm) capable of being identified using the same spectroradiometer used in the field (Figure 3). As with the in situ samples, five spectral readings were collected for each sample, and the values averaged.

**Figure 3.** Use of spectroadiometer in the field (**left**) and a close-up of *Eragrostis tef* (tef; **right**).
