{"paper_id":"16cf2a37-be1e-4cf8-8f71-b7353ce9a417","body_text":"1 \nQuantifying growth and lodging in Tef (Eragrostis tef) with Uncrewed Aerial Systems 1 \n(UAS)  2 \n 3 \nKeely E. Brown1,2, Haley Schuhl1, Dhiraj Srivastava1, Getu Beyene1, Mao Li1, Noah 4 \nFahlgren1,2, Katherine M. Murphy1,2 5 \n1 Donald Danforth Plant Science Center, 975 N. Warson Rd., St. Louis, MO 63132 6 \n2 Corresponding Authors 7 \nRunning Title: Quantifying tef phenotypes with uncrewed aerial systems. 8 \nSignificance Statement 9 \nExtreme weather or heavy grain can cause plant stems to bend, a process called 10 \nlodging. Lodging significantly reduces crop yields globally, particularly in grain crops 11 \nsuch as tef (Eragrostis tef). Semidwarf crops have previously been reported to be 12 \nlodging-resistant, increasing crop yields. Here, we used uncrewed aerial systems (UAS) 13 \nto measure plant growth, height, and lodging in gene edited semidwarf tef lines, and 14 \ncompared the results to ground-truth data. Using a UAS equipped with a red-green-blue 15 \n(RGB) camera or LiDAR sensor, we measured plant height and lodging, and found that 16 \nearly-season height measurements could predict future lodging potential. The tools 17 \nused were contributed to the open-source software PlantCV-Geospatial for community 18 \nuse. This work contributes to a broader understanding of genetic resistance to lodging, 19 \nproviding valuable insights for tef crop improvement and reduces the need for labor-20 \nintensive manual measurements.  21 \n 22 \nKeywords: tef, lodging, UAS, drone, structure-from-motion, LiDAR, phenotyping 23 \n 24 \nAbstract 25 \nLodging is a major contributor to decreased yield in tef, a staple cereal crop in Ethiopia. 26 \nSemidwarf varieties have been developed with a goal to increase yield through reduced 27 \nlodging, but studying lodging susceptibility currently requires a labor-intensive, 28 \nimprecise, manual scoring method. Here we present workflows for analyzing tef stand 29 \nheight from UAS sensors across time to both predict lodging later in the season with 30 \nearly height and to measure the severity of lodging after a storm event. We compare 3D 31 \npoint clouds generated by photogrammetry from RGB images with those generated 32 \nfrom LiDAR to estimate height, demonstrating that they produce similar results, despite 33 \ndifferences in cost. Stand height and lodging can both be accurately measured with low-34 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n2 \ncost UAS, reducing the need for manual measurements and increasing precision and 35 \ntemporal resolution in plant breeding programs.  36 \nIntroduction  37 \nTef (Eragrostis tef (Zucc.) Trotter) is a C4 grass generally considered tolerant to 38 \ndrought, flooding, and pests both in the field and during grain storage (Bekele-Alemu & 39 \nLigaba-Osena, 2023), and is a major staple cereal crop in Ethiopia (Tadele & Hibistu, 40 \n2021). In addition to favorable agronomic traits related to stress tolerance, consumer-41 \nfacing traits, such as gluten-free grain with high nutritional value, are driving increasing 42 \nglobal interest in tef (Abebe et al., 2007; Assefa et al., 2011; Gebremariam et al., 2014; 43 \nSpaenij-Dekking Liesbeth et al., 2005). However, widespread adoption of tef is hindered 44 \nby low yields compared to other grain crops such as maize and wheat (Mihretie et al., 45 \n2022). Lodging, competition from weeds, grain shattering, and low productivity are 46 \nmajor limitations to tef production (Assefa et al., 2011). Like other cereals, lodging in tef 47 \nis exacerbated by high wind speeds, heavy rainfall, and agronomic practices that favor 48 \ntop growth and often cause tef to lodge, a phenomenon in which the stem bends or 49 \nsnaps permanently at the base, causing the plant to fall over (Ben-Zeev et al., 2020; 50 \nBerry et al., 2004; Merchuk-Ovnat et al., 2020). Lodging in tef has been reported to 51 \nreduce grain yield by up to 25% and affects the quality of both grain and straw (Assefa 52 \net al., 2011; Ben-Zeev et al., 2020; Gebru et al., 2023; Zeid et al., 2012).  53 \n 54 \nSemidwarf plants have been associated with lodging resistance, so plant height has 55 \nbeen a target for traditional breeding in many crops. In wheat and rice, lodging resistant 56 \nvarieties were bred using spontaneous mutations, resulting in shorter, thicker stems 57 \n(Dalrymple, 1985; Hedden, 2003). Maize breeding and biotechnology efforts recently 58 \ndeveloped new short-stature varieties to increase yields by reducing susceptibility to 59 \nlodging (Barten et al., 2022). In cultivated barley, lodging is known to correlate with stem 60 \nheight and thickness (Haaning et al., 2020). Tef stand height and lodging resistance is 61 \ninfluenced by variety, row spacing, seeding rate, and more (Blösch et al., 2020; Tasew 62 \net al., 2024; Wato, 2019). To accelerate the process of traditional breeding, Beyene et 63 \nal. (2022) developed CRISPR/Cas9-based genome edited tef lines with reduced plant 64 \nheight and improved lodging resistance, tested in a controlled growth environment.  65 \nIn order to evaluate lodging resistance and its relationship to stature for breeding 66 \npipelines, researchers must accurately measure plant height. However, the current 67 \nstandard methods for measuring plant height are time-consuming and manual (Sun et 68 \nal., 2018; Wang et al., 2018). Additionally, to associate variation in height with lodging 69 \nresistance, a lodging index must be calculated from a subjective scale and human 70 \nestimations for the percentage of a plot at each lodging score (Caldicott & Nuttal, 1979). 71 \nThis method provides information about the overall extent of lodging across a plot, but 72 \nremains subjective and time-consuming. Another challenge in studying lodging is that 73 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n3 \nresearchers must wait for an event like a heavy storm and return to the field to manually 74 \nscore soon after, which may not always be possible.  75 \nHigh-throughput phenotyping provides new opportunities to accurately, precisely, and 76 \nefficiently quantify plant height and lodging with imaging and computational methods to 77 \nmeasure plants and plots (Pauli et al., 2016; Shi et al., 2016; Vergara-Díaz et al., 2016; 78 \nWilke et al., 2019). Measurements made from images captured from Uncrewed Aerial 79 \nSystems (UAS) and satellites are becoming increasingly popular due to their accuracy 80 \nand reduced manual labor (Haghighattalab et al., 2016; Hoffmann et al., 2018; Jindo et 81 \nal., 2021). In particular, drones equipped with RGB (red-green-blue), multispectral, 82 \nhyperspectral, and/or LiDAR (Light Detection and Ranging) sensors have provided 83 \nresearchers with measurements of plant height, lodging, canopy cover, plant 84 \narchitecture, and more (Ayankojo et al., 2023; Barbedo, 2019; Matias et al., 2022). The 85 \nreduced labor needs not only provide more reproducible measurements, but the 86 \nopportunity to increase the frequency or number of measurements feasibly taken in a 87 \ngrowing season. Thus, rather than manual lodging estimations during crop growth as 88 \nwell as after a weather event, lodging can be measured as the change in height before 89 \nand after an event, providing a more holistic and dynamic approach to plant 90 \nmeasurements over time.  91 \nGenerating 3-Dimensional (3D) reconstructions of plants and plots has been shown to 92 \nprovide useful information on plant height and other measurements, but there is a trade-93 \noff between sensor cost and accuracy. LiDAR uses laser pulses to measure distances 94 \nfrom the sensor to objects below, making it particularly useful for estimating plant 95 \nheight, stand height, and canopy density (Swinfield et al., 2019; ten Harkel et al., 2019). 96 \nHowever, LiDAR sensors and the UAS required to carry their often heavy payloads are 97 \ngenerally more expensive than RGB cameras and their associated UAS, and data 98 \nanalysis is often complex. RGB imaging can be used to generate 3D point clouds using 99 \nstructure from motion analysis methods (Leberl et al., 2010), which may have reduced 100 \naccuracy compared to LiDAR (White et al., 2013). While these methods are exciting, it 101 \nis critical that new software tools and methods to analyze these datatypes are 102 \naccessible and reproducible so that other researchers can utilize the measurement 103 \npipelines. Lack of usable software remains a well-documented challenge in the plant 104 \nphenotyping community (Lobet et al., 2013).  105 \nIn this study, we compare manual measurements of plant height and lodging for a group 106 \nof semidwarf tef lines and their parental control with measurements obtained using a 107 \nUAS equipped with RGB and LiDAR sensors. We present reproducible, sustainably 108 \nmaintained analysis pipelines for estimating height using digital elevation models and 109 \nthe open-source software package PlantCV-Geospatial. UAS-based phenotyping 110 \ngenerated accurate measurements of stand height and lodging from both RGB and 111 \nLiDAR data. This confirmed that semidwarf tef lines were less likely to lodge during 112 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n4 \nvegetative growth, in extreme weather 113 \nevents, or during the grain filling stage. 114 \nUAS with either RGB or LiDAR sensors 115 \nthus provide new opportunities for 116 \nresearchers and farmers to evaluate plant 117 \nvarieties for improved lodging resistance 118 \nwithout subjective, manual measurements. 119 \nMaterials and Methods 120 \nPlant growth - Tef lines were grown at the 121 \nDonald Danforth Plant Science Center 122 \nField Research Site in St. Charles, 123 \nMissouri, USA (38.848 N, 90.458 W) from 124 \nJune to September 2023. Two wild-type 125 \nlines (cultivar Magna) and seven genome 126 \nedited, semidwarf tef lines were planted 127 \neach on a 12 m2 plot (4 m x 3 m) 128 \nreplicated three times in randomized 129 \ncomplete block design (a total of 27 plots). 130 \nPlots 101, 102, 103, 901, 902, and 903 131 \nwere wild-type (Figure 1B). Plots 201, 132 \n202, and 203 were sd-1-1 lines, previously 133 \npublished (Beyene et al., 2022), and the 134 \nremaining five were lines generated 135 \ntargeting tef orthologs of known dwarfing 136 \ngenes. Row spacing was 30 cm and 137 \nseeds were planted in each row manually 138 \nusing a salt shaker to give an estimated 1-2 cm spacing between plants. The distance 139 \nbetween plots was 2 m and between replicated blocks 3 m. Plants were irrigated with 140 \nsprinkler irrigation as needed and plots were kept weed-free by manual hoeing. The 141 \nfield used for tef planting was preceded by soybean the previous year, and chemical 142 \nfertilizers were not applied during the growing season. A severe windstorm occurred on 143 \nJuly 29th, 2023, ~8 weeks after planting. Wind speeds reached almost 22 mph (35.4 144 \nkph), as compared to averaging less than 4 mph (6.4 kph) the day before the storm 145 \n(measured by a PheNode environmental sensor, Agrela Ecosystems, Inc.; Fig. S1).  146 \nManual data collection - Stand heights per plot were measured during vegetative growth 147 \nat 6 weeks after planting (July 13, 2023). Stand height was measured from three 148 \ndifferent locations in each plot by measuring the height of tef plants from ground level to 149 \nthe top of the canopy. To do so, an A4 sized piece of paper was placed on top of 150 \nstanding plants in a uniform section of the canopy, and measured from the ground to 151 \nFigure 1 - UAS orthomosaics allowed for clear \nplot delimitation. (A) RGB orthomosaic from \nJuly 17th, 2023. Note the powerlines (white \nlines) above the field. (B) DEM created from \nthe point cloud during stitching of the \northomosaic. The powerlines were converted \nto no data values by PlantCV-Geospatial using \na height threshold. (C) LiDAR from the same \ndate produced point clouds for height \ncalculation. Higher points are toward the red \nend of the scale. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n5 \nthe paper with a ruler. The process was repeated at three randomly selected positions 152 \nin the plot, but only for standing, non-lodged plants. No plant was bent in the process, 153 \nas they supported the lightweight paper. Lodging index was measured at 15 weeks after 154 \nplanting, which was 2 weeks after the last UAS flight, as per Caldicott and Nuttal (1979). 155 \nIn brief, plots were scored from 0 (no lodging) to 5 (flat plants) on a subjective severity 156 \nscale, and the percent of the plot with each score was estimated to calculate a final 157 \nlodging score (Caldicott & Nuttal, 1979). 158 \nUAS data collection - UAS images were acquired at 12 timepoints throughout the 159 \ngrowing season, roughly every 1-2 weeks from the middle of June to the end of August. 160 \nImages were captured with a DJI M600 UAS mounted with RGB (Sony) or LiDAR 161 \n(Phoenix) sensors, and flown at 80 m with a constant horizontal speed of 28.8 kph 162 \nabove the ground for all flights. Front and side imaging overlap for RGB images was 163 \n70%. Image resolution for RGB orthomosaics is reported in Table S1. For LiDAR, the 164 \npulse repetition rate was 700kHZ, the field of view was 90 degrees, with two returns per 165 \npulse and an accuracy of ~2 cm.  166 \nRGB image analysis - Raw images were used to construct orthomosaics using the 167 \nphotogrammetry software Agisoft Metashape (Agisoft LLC, St. Petersburg, Russia) 168 \nthrough the Data to Science (D2S) platform (Jung et al., 2024; Jung, M., B. G. Hancock, 169 \nZ. C. Qian, N. Zhuo, Z. Gong, J. S. Doucette, J. Jung., n.d.), which also created a digital 170 \nelevation model (DEM) representing height of pixels in meters above sea level in the 171 \northomosaic from a dense point cloud. The orthomosaic blending mode was Mosaic 172 \nwith “Fill Holes” enabled, the point cloud quality was set to high, the align photos 173 \naccuracy was set to high, surface was DEM, and the software version of Agisoft 174 \nMetashape Professional was 2.1.3 build 18946. Both orthomosaics and DEMs were 175 \ngeoreferenced using ground control points with the QGIS Georeferencer tool (Dawson 176 \net al., 2025). Georeferencing was done to a single timepoint as a reference rather than 177 \nto an absolute coordinate system. To do so, a reference orthomosaic was opened in the 178 \nmain map canvas, and the second orthomosaic was opened in the Georeferencer 179 \nwindow. The August 10th flight was used as the reference orthomosaic because this 180 \nwas the timepoint used to create shapefile plot boundaries. Ground control point 181 \ncoordinates were entered by clicking on corresponding points in the two orthomosaics. 182 \nTransformation type was Projective, resampling method was Nearest Neighbor, and 8 183 \nground control points per orthomosaic were used to calculate the transformation for 184 \ngeoreferencing. A shapefile was also made in QGIS to generate polygons around each 185 \ntef plot on August 10, and used on the remaining images after georeferencing.  186 \nTo measure height, georeferenced orthomosaics and DEMs were cropped to the field 187 \nand opened in Python using the PlantCV-Geospatial package, which is a library of 188 \nPython tools for analyzing geospatial data. PlantCV is a free, open-source image 189 \nanalysis package for analyzing images of plants (Gehan et al., 2017; Schuhl et al., 190 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n6 \n2025) that provides a framework for measuring and storing observations extracted per 191 \nobject within each image. All code associated with these analyses is available on 192 \nGitHub (https://github.com/danforthcenter/teff-manuscript), as well as the PlantCV-193 \nGeospatial package (https://github.com/danforthcenter/plantcv-geospatial). As observed 194 \nin the orthomosaic (Fig. 1A), tef plots were planted under power lines in the field, which 195 \ncould not be flown under due to UAS safety restrictions. Pixels belonging to powerlines 196 \nneeded to be removed to measure plot heights. During import, PlantCV-Geospatial was 197 \nused with a height percentile threshold for filtering, so values above the threshold were 198 \nconverted to no data values. We used a threshold of 0.995 (unitless, 0-1 range), which 199 \nwas sufficient to remove powerlines in this field (Fig. 1B).  200 \nTo obtain the stand heights of each plant pixel in the tef plots, pixel height needed to be 201 \nsubtracted from the elevation of the soil with respect to the mean sea level. Because the 202 \nland across the field was not uniformly level, a single soil elevation could not be used for 203 \nall plots. However, the soil surrounding each individual tef plot was relatively flat (Fig. 204 \n1B). Therefore, shapefiles per plot included a perimeter of soil so height was measured 205 \nby the difference between plot and soil elevation (Fig. 1). For each plot, the soil 206 \nelevation was estimated to be the 1st percentile in the distribution of plot height (Fig. 2), 207 \ncalculated using ranked pixel values from each plot’s DEM. This soil value (red) was 208 \nthen subtracted from the 95th percentile (blue) in the height distribution of each plot, 209 \nwhich was estimated to represent an average canopy height (Fig. 2).  210 \nLiDAR data analysis - The 211 \nLiDAR data in LAS format was 212 \nimported into MATLAB (R2024a) 213 \nas a 3D point cloud. An image 214 \nacquired on July 17, 2023 was 215 \nselected as a reference for 216 \ndefining the region of interest 217 \n(ROI) of the tef field. A 218 \ncustomized  algorithm was 219 \ndeveloped, allowing rotation and 220 \nprecise definition of the ROI 221 \nboundaries by selecting the 222 \nupper-left and lower-right corner 223 \npoints of the field. Based on prior 224 \nknowledge of tef plant height 225 \nranges, a height threshold of 3 m 226 \nabove the ground level was 227 \napplied to exclude objects and 228 \nnoise exceeding this height, 229 \nFigure 2- (A) Stand height was calculated as the \ndifference between top (95%, blue line) and bottom (1%, \nred line) percentiles from the point cloud distribution. (B) \nAn example of the 3D point cloud showing surrounding \nsoil pixels. (C) A slice inside of a plot in the point cloud \nshowing the distribution of plant heights across the \ncanopy. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n7 \nsuch as power lines. Once the mask parameters were determined from the reference 230 \ndata, they were applied to mask all other LiDAR datasets. 231 \nThe ROI was partitioned into a grid consisting of 3 rows and 9 columns (Fig. 1A). This 232 \nconfiguration was chosen to ensure that each grid cell contained soil points. Within each 233 \ngrid cell, points located at least 25 cm above the minimum elevation were identified as 234 \nthe initial plant subset. The initial plant subsets from all grids were then combined to 235 \nform the entire initial plant point cloud. A density-based noise removal method was 236 \napplied to this point cloud using MATLAB’s pcdenoise() function and the refined data 237 \nwas segmented into clusters of interest (i.e. plots). The edge mask for each plot was 238 \ngenerated and was subsequently applied to all ROIs in other LiDAR datasets.  239 \nWe estimated the plant height using two different approaches. The first approach is 240 \nsimilar to the RGB images, soil elevation for each plot was estimated as the 1st 241 \npercentile of the plot height distribution. This soil elevation was then subtracted from the 242 \n95th percentile of the height distribution, which was used to represent the average 243 \ncanopy height for each plot. 244 \nThe second approach is to estimate stand height by mimicking the manual 245 \nmeasurement process, which involves placing a flat paper over the plant canopy and 246 \nmeasuring its height above the ground. To simulate this, 10 x10 cm squares were 247 \noverlaid to cover each plot, excluding the 30 cm boundary region to remove the 248 \nboundary effects. Within each square, the maximum height of the plant points inside the 249 \nsquare area were measured. The final stand height for each plot was then determined 250 \nas the average height of the top 80th percentile of these maximum square heights. 251 \nStatistics - Linear models, ANOVAs, and correlation coefficients were fit and estimated 252 \nusing base R (version 4.4.2) (Posit team, 2025). Linear mixed-effects models were fit 253 \nusing the R package nlme (Pinheiro et al., 2025) in an R environment running version 254 \n4.1.3.   255 \nResults 256 \nShort tef lines are resistant to lodging after a high wind event 257 \n Shorter plant varieties have consistently been shown to be resistant to plant 258 \nlodging; thus, we hypothesized that semidwarf tef would both be shorter than wild-type 259 \ntef grown in the field, as well as have reduced lodging. Indeed, height measured 260 \nmanually at a single date showed that semidwarf tef was significantly shorter than wild-261 \ntype (F=12.573, p=0.00231). Semidwarf tef also had significantly lower manual lodging 262 \nscores at the end of the season than wild-type (F=112.29, p=3.63e-09). Statistics are 263 \nfrom an ANOVA including both genotype (wild-type or semidwarf) and line as factors.  264 \nHeight estimated from RGB images approximates manual measurements  265 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n8 \nWe next sought to determine whether height estimated with RGB imaging would 266 \ncorrelate to manually-collected measurements, and how additional timepoints of 267 \nmeasurement might add temporal insights. To do so, we used the open-source software 268 \nPlantCV-Geospatial to estimate height from DEMs at each date (see Methods for 269 \ndetails). Using plant area at the 95th percentile threshold and soil area at the 1st 270 \npercentile had the highest correlation to manual measurements, with a Pearson 271 \ncorrelation coefficient (R2) of 0.83 (Fig. 3A), and was used as the height calculation 272 \nparameters for all plots and timepoints. This supports previous research suggesting 273 \nRGB imaging from UAS is an effective method for measuring stand height, as expected 274 \n(Hassan et al., 2019; Matias et al., 2022). 275 \n 276 \nFigure 3- Measurements of height estimated from either RGB images (A) or LiDAR data (B) 277 \ncorrelated with manual measurements. RGB-estimated height was also correlated with LiDAR-278 \nestimated height (C). 279 \nHeight estimated from LiDAR more closely approximates manual measurements   280 \nNext, we hypothesized that LiDAR would also correlate to manual stand height 281 \nmeasurements. Unlike for RGB images, we did not collect LiDAR within a day of the 282 \nmanual measurements (July 13), so we compared the closest timepoint from a flight 283 \nfour days later (July 17). Despite the intervening days, LiDAR-estimated height 284 \npredicted manual height measurements accurately (R2=0.89, Fig. 3B). RGB-estimated 285 \nheight from images collected on July 12th had a lower correlation to manual 286 \nmeasurements compared to LiDAR, despite being closer in time (Fig. 3A).  287 \nImportantly, while RGB compared to manual measurements had a slope of 0.985 288 \n(95% CI 0.803-1.168), LiDAR slope was significantly different from 1 at 0.764 (95% CI 289 \n0.651-0.877) (see intersection of red and black lines in Fig. 3B). The reduced slope of 290 \nthe correlation between LiDAR and manual measurements suggests that either taller 291 \nplots were underestimated by LiDAR, or that shorter plants grew faster than taller plants 292 \nin the 4-day span between the manual measurements and LiDAR flight. Which of these 293 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n9 \nhypotheses explains the relationship cannot be determined from this dataset. When 294 \ncomparing RGB and LiDAR measurements captured on the same date (July 17), we 295 \nfound a strong correlation between measurements from both sensor types (R2=0.90, 296 \nFig. 3C), suggesting they provide similar results despite different methods. The intercept 297 \nof the correlation between RGB and LiDAR does differ from 0, however (t=4.202, 298 \np=0.00295) due to RGB analysis producing larger height estimates. Correlation 299 \nbetween rank order is significant (Spearman’s rho = 0.963, p = 2.41e-07) indicating that 300 \neither method produces measurements with utility for assessing relative height 301 \ndifferences between plots. 302 \nBecause the method for estimating height manually involved averaging the height of a 303 \npaper laid on top of several places in a plot, we also analyzed the LiDAR data using a 304 \nmethod based on this idea (see Methods for more detail). The plot heights as estimated 305 \nby the LiDAR “paper” method were well correlated with the plot heights estimated by the 306 \nLiDAR plant and soil threshold subtraction method (R2 = 0.986, t = 85.50, p < 2e-16). 307 \nThe slope of the correlation is slightly, but significantly, larger than 1 (95% CI 1.024 - 308 \n1.072), indicating that the soil subtraction method produces larger height estimates at 309 \nlarger height values than the “paper” method.  310 \nUAS confirms semidwarf tef lines are shorter than wild-type and more resistant to 311 \nlodging 312 \nManually collected data confirmed that 313 \nsemidwarf tef lines were more resistant to 314 \nlodging. We next tested if UAS methods 315 \ncould provide efficient metrics for lodging, 316 \nrather than the labor-intensive and subjective 317 \nmanual lodging scores.  First, we 318 \nhypothesized that height at a single timepoint 319 \nafter a weather event may be a suitable 320 \nmetric for lodging, and that plots with greater 321 \nlodging scores would have reduced height. 322 \nAfter the storm, wild-type plots had lower 323 \naverage heights compared to semidwarf 324 \nplots, and remained lower for the remainder 325 \nof the season (Fig. 4).  There was a 326 \nsignificant linear correlation (p-value = 327 \n6.627e-07) between the manual lodging 328 \nscores and the UAS measurements of plot 329 \nheight on August 31, which was the last RGB 330 \nFigure 4 - Wild-type plants (orange) were \ntaller than semidwarf lines (blue) early in \nthe season, but lodged more both after a \nstorm event (red vertical line) and \nprogressively towards the end of the \nseason. Colored line represents average of \nall wild-type or semidwarf lines, \nrespectively. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n10 \ntime point and the closest measurement temporally to when lodging was scored.  331 \nHowever, even pre-storm height predicted final manual lodging scores. The plot 332 \nheights from July 27, the midpoint of the growing season and last timepoint before the 333 \nstorm, had a significant relationship (t-value = 6.439700, p-value < 0.0001) to end-of-334 \nseason lodging scores using a linear mixed-effects model (Fig. 5A). This relationship 335 \nindicates that measurements extracted from UAS images at mid-season time points can 336 \npredict the best performing lines. This 337 \nsuggests that while height after a weather 338 \nevent may be an effective high-339 \nthroughput measurement of lodging, 340 \nwithout additional context of size before 341 \nan event, it’s not possible to know if the 342 \nresult is due to starting size or the height 343 \nchange.  344 \nFor a more robust UAS 345 \nmeasurement of lodging, we measured 346 \nthe change in height using DEMs from 347 \nRGB images before and after the high-348 \nwind event (Figure S1). This change in 349 \nheight was strongly correlated to height 350 \nbefore the storm, suggesting taller plants 351 \nwere more likely to lodge (R2=0.55, p-352 \nvalue = 1.03e-05; Figure 5B). This 353 \nmeasurement of change, rather than 354 \nabsolute height, requires multiple 355 \ntimepoints, but eliminates conflating 356 \nstand height with lodging, where a short 357 \nvariety may be considered lodged if only 358 \ncompared at one timepoint. We 359 \nhypothesize that height thresholds might 360 \npredict lodging (> ~0.75m) or continued 361 \ngrowth (< ~0.65m), but sample sizes are 362 \ntoo small to statistically test for more complex dynamics (Figure 5A). More data could 363 \nenable explicit fitting of a changepoint model to capture information that could be useful 364 \nfor breeding to a specific height threshold. 365 \nDiscussion 366 \nFigure 5 - Taller plants experienced a greater \ndecrease in height after a storm (A) and had a \nhigher manual lodging score at the end of the \nseason (B). There was variation in height among \nsemidwarf lines (circles) but the shorter lines \nwere more resistant to lodging. Color represents \na replicated semidwarf line (n = 3). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n11 \nHere, we present a method for using UAS sensor data, both RGB and LiDAR, to 367 \nconstruct 3D point clouds to estimate height of genetically variable plots of Eragrostis tef 368 \nacross a growing season. Both methods for image analysis of point clouds estimate 369 \nheight accurately as compared to manual measurements (Figure 3), but decrease 370 \nnecessary human time, allowing for higher temporal resolution. By imaging at multiple 371 \ntime points through the season, we were able to capture the dynamics of lodging after a 372 \nhigh-wind speed storm. Breeding for higher yield in tef requires adopting strategies for 373 \nreduction in loss from lodging, such as the development of semidwarf varieties. We 374 \nshow that the modified lines used in this study are both shorter and lodge less, and that 375 \nearly season height variance among the lines predicts the severity of lodging.  376 \nOur method uses average stand height to compare lines and replicates. While average 377 \nstand height of a plot is important, the distribution of stand heights within a plot is an 378 \nimportant indicator of plant growth and lodging that can be missed when only 379 \nconsidering a plot-level average. While a person may be able to make a small number 380 \nof manual stand height measurements for a plot, UAS imaging measures stand height 381 \nfor every pixel, and thus is often a more accurate representation of the whole plot. In 382 \nthis study, plant height was described as a singular measure, but the level of detail 383 \navailable to extract from this 3D image data will be valuable in understanding growth 384 \ndynamics, plant health indices, and more. UAS can provide an objective and data-rich 385 \napproach to describing plot height and offers a higher throughput approach to 386 \nquantification in lodging studies in the field. 387 \nWe found that LiDAR and RGB analyses both produced similar results for stand height. 388 \nWhile LiDAR provided a higher correlation to manual measurements (Figure 3), the data 389 \nsizes were larger, analysis was more complex, and the equipment was more costly. 390 \nRGB images also provide additional data on plot color that could be analyzed with 391 \nadditional calibration during photogrammetry for other research questions, such as plant 392 \nhealth, flowering time, and senescence. We recommend researchers consider the 393 \nobjectives of the experiment and the expected height differences of their control and 394 \ntest lines when determining which sensor is appropriate to achieve the experimental 395 \ngoals while keeping costs and analysis time low. For example, more subtle differences 396 \nbetween control and test varieties may require the higher accuracy afforded by LiDAR 397 \nto detect differences.  398 \nThe height analysis from DEMs described above does not require elevation values to be 399 \ngeorectified to an absolute coordinate system, since height is determined by the 400 \ndifference in elevation between plants and soil within a plot. This approach may be less 401 \naccurate when bare soil is not visible, such as when there is total canopy coverage, but 402 \ndepending on the shape of plot height distributions, optimizing for a threshold may still 403 \nwork. Residue (such as from no-till management), cover crops in biculture, and weeds 404 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n12 \nwould also impact the appropriate approach to using UAS for plot heights. We also 405 \nfound that the specific plot boundaries (i.e. how much soil area is included) can affect 406 \nthe appropriate choice for soil and plant thresholds for the pixel distributions, which are 407 \nused to calculate average plot height. As such, we recommend visually inspecting the 408 \nshape of pixel height distributions (such as in Figure 2A) prior to running this analysis to 409 \nhelp choose appropriate threshold values, as was performed here (see Methods). Even 410 \nso, this method  can produce variability due to stochasticity in plot boundaries, which 411 \nshould be considered when interpreting results. 412 \nImportantly, the occurrence of the storm during this experiment was unintentional, and 413 \nobscured possible differences in lodging later in the season that can occur during grain 414 \nfilling. It is possible that the effect of height is consistent across both lodging from wind 415 \nor rain and lodging from grain weight, but that cannot be determined in this study. 416 \nInstead, we focus on the ability of 3D point clouds to aid in both the tracking of height 417 \nchanges through a season and on the prediction of possible yield loss from lodging. 418 \nBecause of the correlation with both manual height measurements at a single time point 419 \nand the increased time resolution afforded, we conclude that UAS image analysis could 420 \nprovide a benefit to tef breeding strategies that target resistance to lodging.    421 \nFigure legends 422 \nFigure 1 - UAS orthomosaics allowed for clear plot delimitation. (A) RGB orthomosaic 423 \nfrom July 17th, 2023. Note the powerlines (white lines) above the field. (B) DEM created 424 \nfrom the point cloud during stitching of the orthomosaic. The powerlines were converted 425 \nto no data values by PlantCV-Geospatial using a height threshold. (C) LiDAR from the 426 \nsame date produced point clouds for height calculation. Higher points are toward the 427 \nred end of the scale.  428 \n 429 \nFigure 2 - (A) Stand height was calculated as the difference between top (95%, blue 430 \nline) and bottom (1%, red line) percentiles from the point cloud distribution. (B) An 431 \nexample of the 3D point cloud showing surrounding soil pixels. (C) A slice inside of a 432 \nplot in the point cloud showing the distribution of plant heights across the canopy. 433 \n 434 \nFigure 3 - Measurements of height estimated from either RGB images (A) or LiDAR 435 \ndata (B) correlated with manual measurements. RGB-estimated height was also 436 \ncorrelated with LiDAR-estimated height (C).  437 \n 438 \nFigure 4 - Wild-type plants (orange) were taller than semidwarf lines (blue) early in the 439 \nseason, but lodged more both after a storm event (red vertical line) and progressively 440 \ntowards the end of the season. Colored line represents average of all wild-type or 441 \nsemidwarf lines, respectively.  442 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n13 \n 443 \nFigure 5 - Taller plants experienced a greater decrease in height after a storm (A) and 444 \nhad a higher manual lodging score at the end of the season (B). There was variation in 445 \nheight among semidwarf lines (circles) but the shorter lines were more resistant to 446 \nlodging. Color represents a replicated semidwarf line (n = 3).  447 \n 448 \nAcknowledgements 449 \nUAV data were collected and provided by Remote Sensing Lab at Saint Louis University 450 \nas part of a Taylor Geospatial Institute Block Grant to the Donald Danforth Plant 451 \nScience Center. We acknowledge the use of Data to Science (D2S, https://d2s.org) 452 \nplatform, an open-source project developed by Geospatial Data Science Lab 453 \n(https://gdsl.org) at Purdue University. We thank the Phenotyping Core Facility 454 \n(RRID:SCR_019049), particularly Joseph Duenwald for ground control point 455 \nmaintenance, and the Field Research Site at the Donald Danforth Plant Science Center 456 \nfor plant care. The developers of PlantCV-Geospatial thank Sam Taylor and Jalissa 457 \nPirro for help and guidance. 458 \n 459 \nAuthor Contributions 460 \nKMM, GB, and NF designed experiments. GB developed tef lines, planted, and 461 \nperformed manual measurements of height and lodging score. KEB, HS, ML, and DS 462 \nperformed data analysis. KEB, HS, and KMM wrote the article with contributions from all 463 \nauthors. 464 \nDeclarations of interests 465 \nGetu Beyene has patent “Lodging resistance in eragrostis tef” pending to Donald 466 \nDanforth Plant Science Center. 467 \n 468 \nFunding 469 \nThis work was supported by a Taylor Geospatial Institute Block Grant to K.M.M. and 470 \nN.F., the National Science Foundation (grant numbers 2120153 and 2346101 to N.F.), 471 \nthe USDA NIFA AFRI (grant number 2022-67021-36467 to N.F.), and by the Bellwether 472 \nFoundation.  473 \n 474 \nData Availability 475 \nCode and data associated with this manuscript are available on GitHub 476 \n(https://github.com/danforthcenter/teff-manuscript).  477 \n 478 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n14 \n 479 \nSupplemental Figure S1: Wind speed measured over a 48h period during which a 480 \nstorm occurred in the vicinity of the field plots. 481 \n 482 \nReferences 483 \nAbebe, Y., Bogale, A., Michael Hambidge, K., Stoecker, B. J., & Gibson, R. S. (2007). Phytate, zinc, 484 \niron and calcium content of selected raw and prepared foods consumed in rural Sidama, 485 \nSouthern Ethiopia, and implications for bioavailability. Journal of Food Composition and 486 \nAnalysis: An Official Publication of the United Nations University, International Network of 487 \nFood Data Systems, 20(3), 161–168. 488 \nAssefa, K., Yu, J.-K., Zeid, M., Belay, G., Tefera, H., & Sorrells, M. E. (2011). Breeding tef [Eragrostis 489 \ntef (Zucc.) trotter]: conventional and molecular approaches. Plant Breeding = Zeitschrift Fur 490 \nPflanzenzuchtung, 130(1), 1–9. 491 \nAyankojo, I. T., Thorp, K. R., & Thompson, A. L. (2023). Advances in the Application of Small 492 \nUnoccupied Aircraft Systems (sUAS) for High-Throughput Plant Phenotyping. Remote 493 \nSensing, 15(10), 2623. 494 \nBarbedo, J. G. A. (2019). A Review on the Use of Unmanned Aerial Vehicles and Imaging Sensors for 495 \nMonitoring and Assessing Plant Stresses. Drones, 3(2), 40. 496 \nBarten, T. J., Kosola, K. R., Dohleman, F. G., Eller, M., Brzostowski, L., Mueller, S., Mioduszewski, J., 497 \nGu, C., Kashyap, S., Ralston, L., Renaud, A., Hall, M., Mack, D., & Gillespie, K. (2022). Short-498 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n15 \nstature maize reduced wind damage during the 2020 midwestern derecho, improving yields 499 \nand greenhouse gas outcomes. Crop Science, 62(6), 2439–2450. 500 \nBekele-Alemu, A., & Ligaba-Osena, A. (2023). Comprehensive in silico analysis of the underutilized 501 \ncrop tef (Eragrostis tef (Zucc.) Trotter) genome reveals drought tolerance signatures. BMC 502 \nPlant Biology, 23(1), 506. 503 \nBen-Zeev, S., Rabinovitz, O., Orlov-Levin, V., Chen, A., Graff, N., Goldwasser, Y., & Saranga, Y. 504 \n(2020). Less Is More: Lower Sowing Rate of Irrigated Tef (Eragrostis tef) Alters Plant 505 \nMorphology and Reduces Lodging. Agronomy, 10(4), 570. 506 \nBerry, P. M., Sterling, M., Spink, J. H., Baker, C. J., Sylvester-Bradley, R., Mooney, S. J., Tams, A. R., 507 \n& Ennos, A. R. (2004). Understanding and reducing lodging in cereals. In Advances in 508 \nAgronomy (pp. 217–271). Elsevier. 509 \nBeyene, G., Chauhan, R. D., Villmer, J., Husic, N., Wang, N., Gebre, E., Girma, D., Chanyalew, S., 510 \nAssefa, K., Tabor, G., Gehan, M., McGrone, M., Yang, M., Lenderts, B., Schwartz, C., Gao, 511 \nH., Gordon-Kamm, W., Taylor, N. J., & MacKenzie, D. J. (2022). CRISPR/Cas9-mediated 512 \ntetra-allelic mutation of the “Green Revolution” SEMIDWARF-1 (SD-1) gene confers lodging 513 \nresistance in tef (Eragrostis tef). Plant Biotechnology Journal, 20(9), 1716–1729. 514 \nBlösch, R., Plaza-Wüthrich, S., Barbier de Reuille, P., Weichert, A., Routier-Kierzkowska, A.-L., 515 \nCannarozzi, G., Robinson, S., & Tadele, Z. (2020). Panicle Angle is an Important Factor in Tef 516 \nLodging Tolerance. Frontiers in Plant Science, 11, 61. 517 \nCaldicott, J. J. B., & Nuttal, A. M. (1979). A method for the assessment of lodging in cereals crops. 518 \nJournal of the National Institute of Agricultural Botany, 15, 88–91. 519 \nDalrymple, D. G. (1985). The Development and Adoption of High-Yielding Varieties of Wheat and 520 \nRice in Developing Countries. American Journal of Agricultural Economics, 67(5), 1067–521 \n1073. 522 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n16 \nDawson, N., Fischer, J., Kuhn, M., Pasotti, A., mhugent, Rouzaud, D., Bruy, A., Sutton, T., Dobias, 523 \nM., Pellerin, M., Rouault, E., Olaya, V., Blottiere, P., Macho, W., Blazek, R., Sherman, G., 524 \nSant-anna, H., Cabieces, J., Bartoletti, L., … Jurgiel, B. (2025). qgis/QGIS: 3.44.4. Zenodo. 525 \nhttps://doi.org/10.5281/ZENODO.17434324 526 \nGebremariam, M. M., Zarnkow, M., & Becker, T. (2014). Teff (Eragrostis tef) as a raw material for 527 \nmalting, brewing and manufacturing of gluten-free foods and beverages: a review. Journal 528 \nof Food Science and Technology, 51(11), 2881–2895. 529 \nGebru, M., Alemayehu, G., & Bitew, Y. (2023). Yield and lodging response of tef [Eragrostis tef 530 \n(Zucc) trotter] varieties to nitrogen and silicon application rates. Heliyon, 9(12), e22576. 531 \nGehan, M. A., Fahlgren, N., Abbasi, A., Berry, J. C., Callen, S. T., Chavez, L., Doust, A. N., Feldman, 532 \nM. J., Gilbert, K. B., Hodge, J. G., Hoyer, J. S., Lin, A., Liu, S., Lizárraga, C., Lorence, A., 533 \nMiller, M., Platon, E., Tessman, M., & Sax, T. (2017). PlantCV v2: Image analysis software for 534 \nhigh-throughput plant phenotyping. PeerJ, 5, e4088. 535 \nHaaning, A. M., Smith, K. P., Brown-Guedira, G. L., Chao, S., Tyagi, P., & Muehlbauer, G. J. (2020). 536 \nNatural Genetic Variation Underlying Tiller Development in Barley ( L). G3 (Bethesda, Md.), 537 \n10(4), 1197–1212. 538 \nHaghighattalab, A., González Pérez, L., Mondal, S., Singh, D., Schinstock, D., Rutkoski, J., Ortiz-539 \nMonasterio, I., Singh, R. P., Goodin, D., & Poland, J. (2016). Application of unmanned aerial 540 \nsystems for high throughput phenotyping of large wheat breeding nurseries. Plant Methods, 541 \n12, 35. 542 \nHassan, M. A., Yang, M., Fu, L., Rasheed, A., Zheng, B., Xia, X., Xiao, Y., & He, Z. (2019). Accuracy 543 \nassessment of plant height using an unmanned aerial vehicle for quantitative genomic 544 \nanalysis in bread wheat. Plant Methods, 15, 37. 545 \nHedden, P. (2003). The genes of the Green Revolution. Trends in Genetics: TIG, 19(1), 5–9. 546 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n17 \nHoffmann, M., Butenko, Y., & Traore, S. (2018). Evaluation of satellite imagery to increase crop yield 547 \nin irrigated agriculture. Agris On-Line Papers in Economics and Informatics, 10(3), 45–55. 548 \nJindo, K., Kozan, O., Iseki, K., Maestrini, B., van Evert, F. K., Wubengeda, Y., Arai, E., Shimabukuro, 549 \nY. E., Sawada, Y., & Kempenaar, C. (2021). Potential utilization of satellite remote sensing 550 \nfor field-based agricultural studies. Chemical and Biological Technologies in Agriculture, 551 \n8(1), 1–16. 552 \nJung, J., Fei, S., Tuinstra, M., Yang, Y., Wang, D., Song, C., Gillan, J., Bhandari, M., Ibrahim, A., Zhao, 553 \nL., Swetnam, T., Barker, B., Jung, M., & Hancock, B. (2024). Data to science: an open-554 \nsource online platform for managing, visualizing, and publishing UAS data. In C. Bauer & J. 555 \nA. Thomasson (Eds.), Autonomous Air and Ground Sensing Systems for Agricultural 556 \nOptimization and Phenotyping IX. SPIE. https://doi.org/10.1117/12.3021199 557 \nJung, M., B. G. Hancock, Z. C. Qian, N. Zhuo, Z. Gong, J. S. Doucette, J. Jung. (n.d.). Data-to-Science 558 \n(D2S): An open-source ecosystem for collaborative geospatial data science research. 559 \nJournal of Open Source Software. 560 \nLeberl, F., Irschara, A., Pock, T., Meixner, P., Gruber, M., Scholz, S., & Wiechert, A. (2010). Point 561 \nClouds. Remote Sensing, 76, 10. 562 \nLobet, G., Draye, X., & Périlleux, C. (2013). An online database for plant image analysis software 563 \ntools. Plant Methods, 9(1), 38. 564 \nMatias, F. I., Green, A., Lachowiec, J. A., LeBauer, D., & Feldman, M. (2022). Bison-Fly: An open-565 \nsource UAV pipeline for plant breeding data collection. The Plant Phenome Journal, 5(1), 566 \ne20048. 567 \nMerchuk-Ovnat, L., Bimro, J., Yaakov, N., Kutsher, Y., Amir-Segev, O., & Reuveni, M. (2020). In-568 \ndepth field characterization of teff [Eragrostis tef (Zucc.) Trotter] variation: From agronomic 569 \nto sensory traits. Agronomy (Basel, Switzerland), 10(8), 1107. 570 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n18 \nMihretie, F. A., Tsunekawa, A., Haregeweyn, N., Adgo, E., Tsubo, M., Masunaga, T., Meshesha, D. T., 571 \nEbabu, K., Nigussie, Z., Sato, S., Berihun, M. L., Hashimoto, Y., Kawbota, A., & Bayable, M. 572 \n(2022). Exploring teff yield variability related with farm management and soil property in 573 \ncontrasting agro-ecologies in Ethiopia. Agricultural Systems, 196, 103338. 574 \nPauli, D., Andrade-Sanchez, P., Carmo-Silva, A. E., Gazave, E., French, A. N., Heun, J., Hunsaker, D. 575 \nJ., Lipka, A. E., Setter, T. L., Strand, R. J., Thorp, K. R., Wang, S., White, J. W., & Gore, M. A. 576 \n(2016). Field-Based High-Throughput Plant Phenotyping Reveals the Temporal Patterns of 577 \nQuantitative Trait Loci Associated with Stress-Responsive Traits in Cotton. G3 , 6(4), 865–578 \n879. 579 \nPinheiro, J., Bates, D., & R Core Team. (2025). nlme: Linear and Nonlinear Mixed Effects Models. 580 \nhttps://doi.org/10.32614/CRAN.package.nlme 581 \nPosit team. (2025). RStudio: Integrated Development Environment for R. Posit Software, PBC. 582 \nhttp://www.posit.co/ 583 \nSchuhl, H., Brown, K. E., Sheng, H., Bhatt, P. K., Gutierrez, J., Schneider, D., Casto, A. L., Acosta-584 \nGamboa, L., Ballenger, J. G., Barbero, F., Braley, J., Brown, A. M., Chavez, L., Cunningham, 585 \nS., Dilhara, M., Dimech, A. M., Duenwald, J. G., Fischer, A., Gordon, J. M., … Fahlgren, N. 586 \n(2025). PlantCV v4: Image analysis software for high-throughput plant phenotyping. In 587 \nbioRxiv (p. 2025.11.19.689271). https://doi.org/10.1101/2025.11.19.689271 588 \nShi, Y., Thomasson, J. A., Murray, S. C., Pugh, N. A., Rooney, W. L., Shafian, S., Rajan, N., Rouze, G., 589 \nMorgan, C. L. S., Neely, H. L., Rana, A., Bagavathiannan, M. V., Henrickson, J., Bowden, E., 590 \nValasek, J., Olsenholler, J., Bishop, M. P., Sheridan, R., Putman, E. B., … Yang, C. (2016). 591 \nUnmanned Aerial Vehicles for High-Throughput Phenotyping and Agronomic Research. 592 \nPloS One, 11(7), e0159781. 593 \nSpaenij-Dekking Liesbeth, Kooy-Winkelaar Yvonne, & Koning Frits. (2005). The Ethiopian Cereal Tef 594 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n19 \nin Celiac Disease. The New England Journal of Medicine, 353(16), 1748–1749. 595 \nSun, S., Li, C., Paterson, A. H., Jiang, Y., Xu, R., Robertson, J. S., Snider, J. L., & Chee, P. W. (2018). 596 \nIn-field High Throughput Phenotyping and Cotton Plant Growth Analysis Using LiDAR. 597 \nFrontiers in Plant Science, 9, 16. 598 \nSwinfield, T., Lindsell, J. A., Williams, J. V., Harrison, R. D., Agustiono, Habibi, Gemita, E., 599 \nSchönlieb, C. B., & Coomes, D. A. (2019). Accurate Measurement of Tropical Forest Canopy 600 \nHeights and Aboveground Carbon Using Structure From Motion. Remote Sensing, 11(8), 601 \n928. 602 \nTadele, E., & Hibistu, T. (2021). Empirical review on the use dynamics and economics of teff in 603 \nEthiopia. Agriculture & Food Security, 10(1), 1–13. 604 \nTasew, W., Habte, A., & Laekemariam, F. (2024). Boosting tef (Eragrostis tef (Zucc.) trotter)) yield 605 \nthrough the use of different inter-row spacing and seeding rates. Advances in Agriculture, 606 \n2024, 1–11. 607 \nten Harkel, J., Bartholomeus, H., & Kooistra, L. (2019). Biomass and Crop Height Estimation of 608 \nDifferent Crops Using UAV-Based Lidar. Remote Sensing, 12(1), 17. 609 \nVergara-Díaz, O., Zaman-Allah, M. A., Masuka, B., Hornero, A., Zarco-Tejada, P., Prasanna, B. M., 610 \nCairns, J. E., & Araus, J. L. (2016). A Novel Remote Sensing Approach for Prediction of Maize 611 \nYield Under Different Conditions of Nitrogen Fertilization. Frontiers in Plant Science, 7, 666. 612 \nWang, X., Singh, D., Marla, S., Morris, G., & Poland, J. (2018). Field-based high-throughput 613 \nphenotyping of plant height in sorghum using different sensing technologies. Plant 614 \nMethods, 14, 53. 615 \nWato, T. (2019). Effects of Nitrogen Fertilizer Rate and Inter-row Spacing on Yield and Yield 616 \nComponents of Teff [Eragrostis teff (Zucc.) Trotter] in Limo District, Southern Ethiopia. 617 \nInternational Journal of Plant & Soil Science, 1–12. 618 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint \n\n20 \nWhite, J., Wulder, M., Vastaranta, M., Coops, N., Pitt, D., & Woods, M. (2013). The utility of image-619 \nbased point clouds for forest inventory: A comparison with airborne Laser Scanning. 620 \nForests, 4(3), 518–536. 621 \nWilke, N., Siegmann, B., Klingbeil, L., Burkart, A., Kraska, T., Muller, O., van Doorn, A., Heinemann, 622 \nS., & Rascher, U. (2019). Quantifying Lodging Percentage and Lodging Severity Using a UAV-623 \nBased Canopy Height Model Combined with an Objective Threshold Approach. Remote 624 \nSensing, 11(5), 515. 625 \nZeid, M., Assefa, K., Haddis, A., Chanyalew, S., & Sorrells, M. E. (2012). Genetic diversity in tef 626 \n(Eragrostis tef) germplasm using SSR markers. Field Crops Research, 127, 64–70. 627 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 7, 2026. ; https://doi.org/10.64898/2026.01.06.697717doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}