) the photo ID of the leaf indicating the vineyard position of 206 the vine it was collected from, 2) an enumerating value 1 through 4 specifying which of four 207 leaves for the variety the data corresponds to, and 3) which vector the data file represents. 208 These files, the raw data, are available at the following link: 209 https://github.com/DanChitwood/grapevine_ampelometry/tree/master/0_visual_check/ampel 210 ometry_data. Tracing all data for a single leaf took approximately 15 minutes. Because the data 211 was traced by hand, it was important to visually verify its accuracy. Analyses in Python were 212 undertaken using NumPy (Oliphant, 2006), pandas (McKinney, 2010), and Matplotlib (
Open resource ↗DanChitwood/grapevine_ampelometry · pdf-layout-page:5 lines:1-54Unverified paper record
The shapes of wine and table grape leaves: an ampelometric study inspired by the methods of Pierre Galet
bioRxiv · 10 May 2020 · 10.1101/2020.05.08.085175
Abstract
The shapes of grapevine leaves have been critical to correctly identify economically important varieties throughout history. The correspondence of homologous features in nearly all grapevine species and varieties has enabled advanced morphometric approaches to mathematically classify leaf shape. These approaches either model leaves through the measurement of numerous vein lengths and angles or measure a finite number of corresponding landmarks and use Procrustean approaches to superimpose points and perform statistical analyses. Hand illustrations, too, play an important role in grapevine identification, as details omitted using the above methods can be visualized. Here, I use a saturating number of pseudo-landmarks to capture intricate, local features in grapevine leaves: the curvature of veins and the shapes of serrations. Using these points, averaged leaf shapes for 60 varieties of wine and table grapes are calculated that preserve features. A pairwise Procrustes distance matrix of the overall morphological similarity of each variety to the other classifies leaves into two main groups--deeply lobed and more entire--that correspond to the measurements of sinus depth by Pierre Galet. Using the system of Galet, pseudo-landmarks are converted into relative distance and angle measurements. Both Galet-inspired and Procrustean methods allow increased accuracy in predicting variety compared to a finite number of landmarks. Using Procrustean pseudo-landmarks captures grapevine leaf shape at the same level of detail as drawings and provides a quantitative method to arrive at mean leaf shapes representing varieties that can be used within a predictive statistical framework.
Plant phenotyping relevance
ブドウ葉の形状を擬似ランドマークとプロクルステス解析で定量化し、品種識別・平均葉形状推定に用いる手法の開発が中心である。
abstractHere, I use a saturating number of pseudo-landmarks to capture intricate, local features in grapevine leaves: the curvature of veins and the shapes of serrations.
abstractUsing Procrustean pseudo-landmarks captures grapevine leaf shape at the same level of detail as drawings and provides a quantitative method to arrive at mean leaf shapes representing varieties that can be used within a predictive statistical framework.
Code and data availability
The paper explicitly links public GitHub repositories and a Dryad DOI containing its leaf photographs, hand-traced landmark/pseudo-landmark raw data, visual-check outputs, interpolation code and outputs, and Procrustes analysis code and outputs — all paper-specific, public, and actionable.
the data 211 was traced by hand, it was important to visually verify its accuracy. Analyses in Python were 212 undertaken using NumPy (Oliphant, 2006), pandas (McKinney, 2010), and Matplotlib (Hunter, 213 2007) to plot the data on the actual photo. The code for plotting vectors onto the original photo 214 can be found here: 215 https://github.com/DanChitwood/grapevine_ampelometry/blob/master/0_visual_check/ampel 216 ometry_visual_check.ipynb. The visual checks for each of the 240 leaves analyzed in this study 217 can be found here: 218 https://github.com/DanChitwood/grapevine_ampelometry/tree/master/0_visual_check/outpu 219 t_visual_check 220 5
Open resource ↗DanChitwood/grapevine_ampelometry · pdf-layout-page:5 lines:1-54ith assigned numbers of points to every vector, interpolation was 240 used to calculate equidistant pseudo-landmarks. A function was created using the scipy 241 (Virtanen et al., 2020) interp1d function to interpolate the correct number of equidistance 242 points for each vector. The code used to interpolate points is here: 243 https://github.com/DanChitwood/grapevine_ampelometry/blob/master/1_interpolation/ampe 244 lometry_interpolation.ipynb. The interpolated points can be found here: 245 https://github.com/DanChitwood/grapevine_ampelometry/blob/master/1_interpolation/outp 246 ut_interpolated_points.txt 247 248 With corresponding points between all leaves, a Procrustes analysis could be pe
Open resource ↗DanChitwood/grapevine_ampelometry · pdf-layout-page:6 lines:1-54esults saved as a pairwise distance matrix. The hclust() function in R using the 259 “mcquitty” method was used to hierarchically cluster varieties based on the pairwise distance 260 matrix and overall morphological similarity. The code for performing a Procrustes analysis for 261 each variety and outputs can be found here: 262 https://github.com/DanChitwood/grapevine_ampelometry/tree/master/2_procrustes_by_vari 263 ety 264 6
Open resource ↗DanChitwood/grapevine_ampelometry · pdf-layout-page:6 lines:1-54ll Procrustes mean 265 shape, super-imposed Procrustes coordinates for all leaves, and eigenvalues and eigenleaves 266 from a PCA. The superimposed Procrustes coordinates of all leaves and the mean shape were 267 plotted together. The code for the Procrustes analysis for all 240 leaves and the outputs can be 268 found here: 269 https://github.com/DanChitwood/grapevine_ampelometry/tree/master/3_overall_procrustes 270 271 Data analysis 272 273 To calculate allometry for each line segment, distances between all points were converted to 274 cm using the pixel to cm scale measured for each leaf. The lm() function in R was used to model 275 the natural log of the distance from each point to the ne
Open resource ↗DanChitwood/grapevine_ampelometry · pdf-raw-page:7 lines:1-92This is an automatically classified, unverified record. Curator approval is required before any resource enters the Catalog.