Garryanalyzer: A morphometric workflow and open‐source ImageJ plug‐in for quantitative morphological analysis of Pacific Northwest Quercus leaves Authors: Ziqi Xie, Tan Bahar, Keith Karoly Published: 2026 ABSTRACT AbstractPremiseAccurate species identification is crucial for ecological restoration and can be especially challenging for understudied non‐model species. Quercus garryana is the only native oak species in the Pacific Northwest and is an important component of the endangered oak savanna ecosystem. Quercus robur is an imported ornamental species from Europe and has been found to be mistakenly planted as Q. garryana in habitat restoration projects.MethodsWe measured leaf morphological traits sampled from herbarium collections in their native ranges using the digital morphometric tools MorphoLeaf and Tomato Analyzer. We then used Lasso logistic analysis to generate a predictive model and tested it on leaves from Portland, Oregon. To streamline this species detection process, we developed Garryanalyzer, an ImageJ plug‐in that automatically measures leaf traits and outputs species predictions.ResultsGarryanalyzer demonstrated 95% accuracy in predicting the species identity of herbarium specimens of oaks. Garryanalyzer correctly identified all Q. robur individuals sampled in Portland but showed lower accuracy for Q. garryana.DiscussionMany existing morphometric software are not open source, which makes them unable to be customized to specific study systems. Garryanalyzer is built upon the widely used open‐source ImageJ platform. This study also demonstrates a viable workflow for developing similar tools for other ecologically important non‐model plant species.Keywords: applied ecology, conservation biology, digital morphometrics, non‐native species, oak morphology, species identification The detection and control of non‐native species have been a longstanding topic of interest in ecological restoration (Berger, 1993; Graham et al., 2024). Many non‐native species are introduced to new areas as a result of human activities, with landscaping being the most frequent reason in the case of woody plant taxa (Reichard and Hamilton, 1997; Reichard and White, 2001). Non‐native species have the potential to disrupt the local ecosystem by outcompeting or predating upon native species, by introducing pathogens, or by altering ecosystem functioning (Simberloff, 2015; Pyšek et al., 2020). A further concern arises when non‐native species are introduced into habitats with closely related native species where hybridization and introgression might occur (Vanden Broeck et al., 2005; Metcalf et al., 2008; Thompson et al., 2010; Millar et al., 2012; Johnson et al., 2016; Fitzek et al., 2018). Distinguishing between morphologically similar native and non‐native species is a critical step towards minimizing any conservation impacts from potential ecological disruption and genetic exchange between closely related taxa (Rhymer and Simberloff, 1996; Vilà et al., 2000; Allendorf et al., 2001; Bohling, 2016; Todesco et al., 2016). In the Pacific Northwest, the coexistence of a native oak species and its imported lookalike is a good representation of such an issue. The Oregon white oak (Quercus garryana Douglas ex Hook.) has one of the largest latitudinal ranges of oaks in western North America, extending from southern California to southern British Columbia (Burns and Honkala, 1990). North of the mid‐Willamette Valley, Q. garryana is the only native oak species in its range and serves as a key component of the prairie–oak ecosystem that is among the most endangered native habitats in the Pacific Northwest (Dunwiddie and Bakker, 2011). The pedunculate or English oak (Quercus robur L.) is native to Europe and western Asia and was introduced to northern Oregon in the late 19th century where it was cultivated primarily as an ornamental (Christy et al., 2009). The two oaks are both placed in section Quercus (white oak) of the oak genus (Denk et al., 2017; Manos and Hipp, 2021). Quercus robur is the most common non‐native white oak species in Portland, Oregon, and can look highly similar to the native Q. garryana. These two species have not been reported to hybridize. The morphological similarity between these two oak species has sparked concern from park and natural area managers in Portland. Non‐native Q. robur individuals have been observed in natural areas (Christy et al., 2009) and found to have been mislabeled as the native Q. garryana in some parks (Ed Benedict Park and Lair Hill Park; Montgomery, 2022). Furthermore, as a keystone species of Pacific Northwest native habitats, Q. garryana is often planted in restoration projects, but there have been records of Q. robur being mislabeled as Q. garryana and commercially supplied for such projects (Christy et al., 2009). This concern is further verified by the discovery of several young Q. robur individuals in the Baltimore Woods Nature Corridor, a natural area managed with the goal of supporting native species (Xie, 2023). In the field, it can be quite difficult to distinguish between Q. robur and Q. garryana due to their morphological similarities, especially for younger trees. While for older trees one can rely on traits related to the bark, crown, and acorns, younger individuals typically have not yet developed these characteristics. An alternative species identification method with high accuracy is genetic markers, but access to wet lab resources may be limited under field conditions and morphological indicators are often preferred. Therefore, leaf morphology is the easiest trait to examine and compare for species identification, especially for trees that have not reached maturity. The confounding morphology of Portland white oaks calls for a quantitative identification of outlier characteristics that reliably distinguishes the two species. One such solution is geometric morphometric methods (GMM). Over the past decades, GMM—which includes mapping the outline of shapes and placing landmarks that allow for mathematical analysis of the relative positions among them—have become increasingly popular (Klein and Svoboda, 2017; Oso and Jayeola, 2021). Relative to leaf morphology, digital GMM tools with automated functions of shape capture and landmark placement on leaf images have been widely employed in taxonomic and developmental studies, such as MorphoLeaf (Andrey and Maurin, 2005; Biot et al., 2016; Oughou et al., 2023), MASS (Chuanromanee et al., 2019), LEAFPROCESSOR (Backhaus et al., 2010), LeafJ (Maloof et al., 2013), and MuLES (Bowman et al., 2023). Some other useful plant GMM software targets structures other than leaves: for example, Tomato Analyzer quantitatively measures the shape of tomato fruits (Gonzalo et al., 2009; Rodríguez et al., 2010), but some of its functions are applicable to leaf shapes. These programs not only expedite leaf morphometric analysis by automating landmark placement and calculation but also reduce human‐induced error by standardizing measurement procedures. The first goal of this study was a quantitative analysis of leaf morphometrics of Q. garryana and Q. robur. We looked for digital GMM tools that can perform both basic measurements such as length, width, and area, as well as complex shape analyses such as conformity to a best‐fit circle or ellipse. Additionally, because white oaks have lobed leaves, software that measures the position and size of lobes was preferred. Based on these criteria, MorphoLeaf (Andrey and Maurin, 2005; Biot et al., 2016; Oughou et al., 2023) and Tomato Analyzer (Gonzalo et al., 2009; Rodríguez et al., 2010) were chosen. As neither of these programs measured petiole length, we performed manual measurement of petiole length in Fiji, a pre‐packaged distribution of the scientific image analysis software ImageJ (Abràmoff et al., 2004; Schindelin et al., 2012; Schneider et al., 2012). ImageJ is a widely used open‐source platform for extraction and measurement of shapes, and there have been automated or semi‐automated plug‐ins for ImageJ that focus on leaf morphology, such as MuLES (Bowman et al., 2023) and LeafJ (Maloof et al., 2013). However, these two plug‐ins were not selected for morphometric analysis of our oak leaf samples, as most parameters measured by MuLES and LeafJ overlap with those in Tomato Analyzer and MorphoLeaf, with MorphoLeaf being prioritized because of its inclusion of individual lobe measurements. LeafJ has a unique function of automatic petiole length extraction, but its requirement of manual adjustment of the images’ color channel prior to analysis made this function less intuitive to use. In this study, we specifically aimed to investigate whether a predictive model for Portland white oak species identity can be derived from a quantitative analysis of leaf morphometrics of Q. garryana and Q. robur individuals sampled from each of their native ranges. The second goal was to develop a tool that streamlines the process of species identification between Q. garryana and Q. robur. While many automated digital GMM tools already exist, they are not taxon‐specific and therefore include many parameters that may not be relevant to the species of interest. After screening quantitatively distinguishing traits with general digital GMM tools, we aimed to create a simple program, Garryanalyzer, that readily informs the user of the species identity of a white oak after the input of a leaf image and a few clicks. Garryanalyzer was developed as an ImageJ plug‐in because ImageJ is highly customizable due to its open‐source nature and has a user‐friendly GUI and a large user base. Because other oak species in the Portland area are more taxonomically distant (Manos and Hipp, 2021) and morphologically dissimilar from Q. garryana and Q. robur and because these two species are not known to hybridize, they serve as a good model for demonstrating a workflow of narrowing down morphometric traits for native and non‐native species distinction. Garryanalyzer can not only help natural resource managers and the general public who need a tool to quickly determine if a white oak in Portland is the native species, but also serve as an example for researchers developing customized software for species identification in non‐model plants. METHODS Herbarium database sampling Twenty herbarium sheets collected in the native habitat for each species were selected from the Global Biodiversity Information Facility (GBIF; https://www.gbif.org/) website under the occurrence section using the following filters: “Occurrence status: Present”, “Basis of record: Preserved specimen”, “Location: including coordinates”, and “Media type: Image.” For Q. garryana, the taxon searched was “Quercus garryana Douglas ex Hook.” To limit the results to the Pacific Northwest, the following coordinates range was used: 44°S to 90°N, −180°W to −116°E. These coordinates limit the range to the Willamette Valley, where the most common morphotype (var. garryana) is the same as that found in Portland. As there were only 89 results within this range, each herbarium sheet was first examined for its viability for digital morphometric analysis based on the quality of the image and the existence of at least one isolated, intact, and flattened leaf. Out of these 89 results, there were 20 sheets that met the above criteria with minimal photo‐editing required, along with 18 sheets that could be viable with some photo‐editing as they had intact and flattened leaves that overlaid on top of other leaves. Upon examining the location distribution of the 20 qualifying sheets, we found that only one sample was from Washington and a disproportionate number of samples were located relatively close to each other in British Columbia. Therefore, three samples from British Columbia were randomly taken out and replaced with three random Washington samples from the potentially viable group. The finalized selection contained four samples from British Columbia, four samples from Washington, and 12 samples from Oregon (Appendix S1; see Supporting Information with this article). For Q. robur, the taxon searched was “Quercus robur L.” The coordinates range was set to encompass its European habitat: 60°N to 40°S, −10.5°W to 60°E. Because there were a large number of results, herbarium sheets were randomly selected by adjusting the screen display to fit 10 samples per row and identifying the first viable sheet from 20 randomly selected rows. The locations of the samples were then examined to make sure they were relatively evenly distributed across the native range of Q. robur (Appendix S1). Sampling of Portland leaves According to the City of Portland's online database of the location and identity of ~25,000 trees found in city parks (https://www.portland.gov/trees/get-involved/treeinventory), more than 2000 of these trees are oaks from 23 species; 258 Q. garryana individuals are present in 55 city parks and 223 Q. robur are present in 25 city parks. Locations where both species coexist were prioritized for sampling. In fall of 2020 and summer of 2022, leaf and bud tissue from trees of each species were collected from nine of the 13 city parks that have both species (Appendix S2). Following conversations with City of Portland park and natural area managers, we also sampled trees in four natural areas in spring of 2023 (Appendix S2; Sellwood Riverfront Park, Oaks Bottom Wildlife Refuge, Baltimore Woods Nature Corridor, and Powell Butte Nature Park; 6–7 trees at each site). These areas were all managed with the goal of supporting native Q. garryana trees as a component of the habitat. At all four sites, either volunteer oaks or oaks planted recently for habitat enhancement included trees of uncertain species identity. A few trees from each park or natural area were selected to undergo morphometric analysis, totaling 42 individuals. One adult leaf was randomly selected from each individual for the analysis. Genotyping of Portland trees As part of a previous project aimed at developing genetic markers to distinguish between the two species, a subset of Portland samples was genotyped with one chloroplast and one nuclear DNA marker (Montgomery, 2022; Xie, 2023). Genotyping was used to verify species identities of some labeled park trees, as well as to test trees of unknown species identities in natural areas. The chloroplast DNA marker used was rbcL‐atpB (National Center for Biotechnology Information [NCBI] accession: AY042495; Manos and Stanford, 2001) and the nuclear DNA marker was a sequence with the ID of STS 8745 (NCBI accession: HF563416; Hubert et al., 2014). A detailed description of DNA marker selection and genotyping protocol can be found in Montgomery (2022) and Xie (2023). Imaging and photo‐editing Leaves of Q. garryana and Q. robur collected in Portland were flattened and dried with a plant press. The leaves were then individually placed into an Ortery Photosimile 200 Desktop Product Photography Studio (Ortery Technologies, Irvine, California, USA) and photographed using a Nikon 1 J5 mirrorless digital camera (Nikon, Tokyo, Japan). A 10‐cm ruler was placed next to the leaves as a size marker. As several leaf images from GBIF were below the minimum recommended image dimension size for MorphoLeaf (1000 pixels along the main leaf axis), we resampled these images to 400% the original size using the Enhanced Super‐Resolution Generative Adversarial Networks (ESRGAN; Wang et al., 2019), implemented through Cupscale (version 1.39.0; https://github.com/n00mkrad/cupscale). A bicubic scaling filter at 100% factor was applied to increase the low‐resolution images to the minimum viable image size; visual artifacts produced from this process were deemed minimal. To meet the requirements for MorphoLeaf, one viable leaf was randomly selected from each GBIF herbarium sheet image and cropped out using Photoshop CS6 (Adobe Systems, San Jose, California, USA). Shadows around each leaf were erased, and the leaves were given a light background and rotated to a horizontal position with the apex of the leaf pointing to the right. Leaves imaged in the lab required minimal editing besides cropping off the ruler after calibration. To analyze the leaves in Tomato Analyzer, the images were edited onto a black background, with petioles erased and the leaf blades rotated to a vertical position with the apex pointing down. The color channel of some of the leaves was adjusted to create more contrast with their background. Morphometric analysis MorphoLeaf MorphoLeaf was used to perform basic size measurements and to quantify lobe‐related traits. Calibration of pixel width was performed in accordance with the ruler in each GBIF herbarium sheet or Portland leaf image photographed in the lab. Contour extraction and identification of sinuses and teeth (lobes) were automatically performed in MorphoLeaf (Figure 1). Sinus extraction was performed with the parameters “Half neighborhood” = 25 and “Maximum negative curvature” = 145; extraction of teeth tips was performed with the “Local curvature maximum” method with the setting 10% length of the tooth. No hierarchy was applied to the teeth. For some samples, some manual adjustments were made to the contour outline and the position of the sinuses and teeth to better fit the leaf outline. The following parameters were measured using MorphoLeaf: blade length, blade width BB (between peaks of lobes), blade width IS (between dips of sinuses), blade area, blade perimeter, petiole width, total teeth number, individual tooth width, individual tooth height (latitude), individual tooth height (median), individual tooth area, and individual tooth position relative to the base and tip (Andrey and Maurin, 2005; Biot et al., 2016; Oughou et al., 2023). Mock‐up of MorphoLeaf interface of contour, sinuses, and teeth detection. Image created in Photoshop CS6. ImageJ The petiole length of the GBIF and Portland samples was measured manually in Fiji (Abràmoff et al., 2004; Schindelin et al., 2012). Calibration in accordance with the rulers was made in the Image > Properties section by changing the pixel width. A straight or segmented line was used to mark out the petiole of each leaf image, and the length of this line was recorded as the petiole length. Tomato Analyzer The ellipsoid score, circular score, and absolute ovoidness measurements in Tomato Analyzer (Gonzalo et al., 2009; Rodríguez et al., 2010) were used to analyze the homogeneity and asymmetry of the blades of GBIF and Portland leaves (Figure 2). The petiole was erased on each leaf so that only the blade portion remained. As the ellipsoid score, circular score, and absolute ovoidness reflect within‐leaf proportion instead of absolute sizes, calibration was not required. The ellipsoid and circular scores reflect the proportion of error between the actual leaf area and a best‐fit ellipse or circle, respectively. A smaller value for either measurement means there is less deviation of the leaf shape from the best‐fit shape. The absolute ovoidness measurements were derived from two measurements in Tomato Analyzer—obovoid and ovoid—so that when positioned with apex up, blades that fit an ovoid shape are given a positive value of absolute ovoidness and blades that fit an obovoid shape are given a negative value. Tomato Analyzer interface of (A) ellipsoid, (B) circular, and (C) ovoidness and obovoidness measurements. On the left is a Quercus garryana leaf and on the right is a Q. robur leaf. Composite morphometric measurements As MorphoLeaf measured the size and position of individual lobes, to quantify lobe (tooth) characteristics for individual leaves, we calculated the means for tooth width, area, height, perimeter, and position from the base for each leaf. To obtain more ratio‐based measurements, we also calculated the sum of teeth area:blade area, blade length:blade width (BB), and mean tooth height:mean tooth width (latitude). Finally, petiole length:blade length was calculated based on measurements from Fiji (ImageJ) and MorphoLeaf, respectively. There were a total of 22 morphometric parameters measured combining direct and derived results from all software. Statistical analyses Univariate tests of mean differences for the 22 morphometric traits were performed in JMP 15 (SAS Institute, Cary, North Carolina, USA) for the GBIF dataset to preliminarily identify traits that were significantly different between the two species found in their native range. To further identify the traits that contribute the most to the morphological differences between the two species, all 22 traits of the GBIF dataset were then analyzed with Lasso logistic regression (Tibshirani, 1996; Friedman et al., 2010), using the rlassologit() function from the hdm R package (High‐dimensional Metrics; Chernozhukov et al., 2016). Lasso analysis works by implementing a shrinkage estimator that pulls the smaller coefficients to zero, leaving only the most significant parameters as variables to generate a logistic regression model for species prediction. The species identity of each GBIF sample was converted to 1 for Q. garryana (“GA”) and 0 for Q. robur (“RO”). Lasso analysis derived a logistic regression equation that calculated a log‐odds x = log(P(GA)/P(RO)). This log‐odds would then be converted to a probability using the logistic function p = 1/(1 + e −x ), which ranges between 0 to 1, where a value closer to 1 indicates that the leaf sample is more likely to be from a Q. garryana individual. This logistic regression equation was then tested on the Lasso‐identified significant traits of Portland oaks to assess the equation's accuracy in species prediction. Development of Garryanalyzer An ImageJ plug‐in, Garryanalyzer, was developed to streamline the process of measuring the significant traits—petiole length:blade length ratio and ellipsoid score (later renamed to “elliptical” in Garryanalyzer)—identified by Lasso and subsequent species prediction based on fitting a logistic regression. For the petiole length:blade length ratio calculation, a graphical interface was programmed to prompt the user to draw a straight line that covers the petiole. This line serves as an indicator for the plug‐in to both measure and mask the petiole. For the remaining blade portion of the leaf, the plug‐in sequentially converts the image to grayscale, generates a threshold‐based separation for high contrast between the blade outline and background, isolates the blade outline with a wand selection, and records blade length as the length of a two‐dimensional box created around the outline. The outline is subsequently used for the next steps, namely elliptical score calculation. To accurately replicate the calculation method of Tomato Analyzer's ellipsoid score, a template image of different elliptical shapes (Appendix S3) was used to cross‐reference with Tomato Analyzer's ellipsoid score outputs, which is derived from a measure of deviation from a best‐fit ellipse superimposed over the actual leaf shape. Tomato Analyzer's documentation for the ellipsoid function was also referenced (Gonzalo et al., 2009; Rodríguez et al., 2010). A script for generating a best‐fit ellipse in ImageJ was adapted from the ImageJ Macro library (https://imagej.net/ij/macros/DrawEllipse.txt) and edits were made to subdivide the blade outline's resultant best‐fit ellipse, as well as the outline itself, each into a set of 300 vertices as per the default Tomato Analyzer settings. These vertices were tested against each other to come up with the shortest distance along the ellipse from the closest point of a leaf, averaging to produce an elliptical score. Because the elliptical score value generated by Garryanalyzer is not exactly the same as the ellipsoid score in Tomato Analyzer, a new logistic regression was generated using rlassologit() by replacing the ellipsoid measurements from Tomato Analyzer with the elliptical scores from Garryanalyzer. With this new logistic regression equation, we retested species prediction with both GBIF and Portland area leaves and compared the results to the prediction based on the original logistic regression. RESULTS Species identification with the logistic regression model Univariate tests of mean differences for the 22 morphometric traits measured on GBIF leaves suggested that only five of the variables (elliptical score [equivalent to ellipsoid score in Tomato Analyzer], petiole length, blade length:blade width ratio, tooth area:blade area ratio, and petiole length:blade length ratio) differed statistically significantly between the two species. After applying the shrinkage estimator, the Lasso logistic regression model identified elliptical score and petiole length:blade length ratio (“PBratio”) as the only predictor variables with non‐zero coefficients, with the following equation: where a predicted probability p closer to 1 indicates that the leaf sample is more likely to be from a Q. garryana (“GA”) individual. Using a threshold of 0.5 for classification, 38 out of the 40 GBIF samples were correctly identified using this logistic regression equation (Appendix S4). Of the two samples that were misidentified, one was Q. garryana and the other was Q. robur. When the equation was applied to Portland leaf samples (42 total), all Q. robur samples were correctly identified by this equation, including, notably, the six Q. robur trees in the Baltimore Woods Nature Corridor that were previously found through genotyping to be mislabeled as Q. garryana in the city's inventory. However, 14 individuals labeled as Q. garryana in the City's database had a predicted probability of less than 0.5 (Appendix S5). Five of these individuals were previously confirmed by genotype to be Q. garryana (Appendix S5; Xie, 2023). When the distribution of predicted species probabilities was plotted out using the R package ggplot2 (Wickham, 2016), the probabilities of both datasets showed a bimodal distribution, but the predicted probabilities for Q. garryana individuals were more spread out and did not congregate into a single peak like those for Q. robur (Figure 3). The distribution of predicted species probability of (A) GBIF and (B) Portland trees. GA = Quercus garryana, RO = Q. robur. Garryanalyzer Parameters and user input Garryanalyzer is an open‐source ImageJ plug‐in that automatically differentiates between leaves of Q. garryana and Q. robur based on two significantly distinct morphometric traits identified in the statistical analyses: elliptical score and petiole length:blade length ratio. It works by first calculating values for the two traits, followed by fitting them into a logistic regression and deriving a predicted probability for each leaf image input. The plug‐in asks for an image where an intact leaf is placed horizontally on a light background. The user will be prompted to manually select the petiole (Figure 4). This enables the program to measure the length of the petiole and determine the contour of the blade. Garryanalyzer user interface showing (A) petiole selection and (B) region of interest (ROI) generation and species prediction. Regions of interest With each leaf input image, the following region of interest (ROI) items are automatically generated by the plug‐in. These ROIs aid the calculation of the two determining traits (petiole:blade length ratio and elliptical score). The ROIs can also be referenced for troubleshooting purposes. Petiole length: This value is generated after the user manually selects the petiole. It is used by ImageJ to generate a vertical separation of petiole from leaf to determine the elliptical score as well as the petiole:blade length ratio. Petiole length: This value is generated after the user manually selects the petiole. It is used by ImageJ to generate a vertical separation of petiole from leaf to determine the elliptical score as well as the petiole:blade length ratio. Petiole length flat: This is a flattened, horizontal version of the petiole length. It is used as a visual aid to show user‐input petiole length versus generated blade length in the image. Petiole length flat: This is a flattened, horizontal version of the petiole length. It is used as a visual aid to show user‐input petiole length versus generated blade length in the image. Blade outline: An outline of the blade is created from the remaining blade structure (i.e., the part of the leaf without the petiole). After the petiole is erased, the shape of the blade is isolated from the background with its outline extracted. This outline will later be used for quantifying the elliptical score together with the best‐fit ellipse. Blade outline: An outline of the blade is created from the remaining blade structure (i.e., the part of the leaf without the petiole). After the petiole is erased, the shape of the blade is isolated from the background with its outline extracted. This outline will later be used for quantifying the elliptical score together with the best‐fit ellipse. Blade bounds: This is a two‐dimensional box created around the blade outline. It is used for generating blade length. Blade bounds: This is a two‐dimensional box created around the blade outline. It is used for generating blade length. Blade length: This is a line along the lower end of the blade bound. It is used for calculating the petiole:blade length ratio. Blade length: This is a line along the lower end of the blade bound. It is used for calculating the petiole:blade length ratio. Best‐fit ellipse: A best‐fit ellipse is created in reference to the blade outline. The ellipse is then used for quantifying the elliptical score together with the blade outline. Best‐fit ellipse: A best‐fit ellipse is created in reference to the blade outline. The ellipse is then used for quantifying the elliptical score together with the blade outline. Petiole:blade length ratio After the user manually identifies the position of the two ends of the petiole, the length of the petiole is automatically measured and converted to “petiole length flat.” The petiole is then masked and separated from the rest of the leaf, and the outline of the blade is automatically detected. Afterwards, the “blade outline” is converted to “blade bounds” and “blade length” (Figure 4). The resulting distances of the lines “blade length” and “petiole length flat” are used to compute the petiole:blade length ratio. Elliptical score A best‐fit ellipse is generated for the “blade outline” of each leaf after the petiole is removed (Figure 4). The program subdivides both the resultant ellipse and the “blade outline” each into a set of 300 vertices as per the default Tomato Analyzer settings. These vertices are tested against each other to come up with the shortest distance along the ellipse from the closest point on the leaf blade, averaging to produce an elliptical score (Video S1). Species output equation To account for the difference in outputs between elliptical score in Garryanalyzer and ellipsoid in Tomato Analyzer, a new logistic regression was calculated with Lasso and used for species prediction in Garryanalyzer. The new predictive equation is as follows: where a predicted probability of p closer to 1 indicates that the leaf sample is more likely to be from a Q. garryana individual, using a threshold of 0.5 for classification. With this new equation, 37 out of 40 samples from GBIF were correctly identified (Appendix S4). Two out of the three misidentified samples were the same as when the original equation was used, and an additional Q. robur sample was misidentified. When the equation was applied to the Portland leaf samples (42 total), 13 individuals identified as Q. garryana in the City's database had a predicted probability of less than 0.5 (Appendix S5). All 13 misidentified individuals were among those misidentified using the original equation, and five of these individuals were previously genotyped to be Q. garryana (Appendix S5; Xie, 2023). DISCUSSION This study used leaf morphometrics to distinguish between the most common native and non‐native white oak (section Quercus) species in Portland that are visually similar to each other. Accurately distinguishing between these two species is important as the native Q. garryana is a keystone species often used in restoration projects and the non‐native Q. robur has been mistakenly supplied to such projects (Christy et al., 2009). For younger trees that have not yet developed acorn, bark, or crown features, as is the case with most volunteer saplings in natural areas and trees planted for restoration, leaf morphology is the only reliable trait to visually distinguish between these two species. The first contribution of this study is the identification of quantitative leaf morphological traits that can be used to distinguish between Q. garryana and Q. robur. We measured various leaf traits of herbarium samples collected from GBIF using the digital morphometric tools MorphoLeaf and Tomato Analyzer. Univariate tests of these traits showed that, apart from petiole length, the most quantitatively significant morphological differences between leaves of the two species are general shape homogeneity (e.g., elliptical score) and ratio measurements (e.g., petiole length:blade length ratio) between defined regions of the leaves, instead of absolute geometric measurements (e.g., blade length or blade width). Lasso logistic regression analysis further narrowed down the distinguishing characteristics to two key leaf traits—one ratio and one shape trait, specifically, petiole length:blade length ratio and elliptical score—which were used to construct a model for species identity prediction. Simply put, our analysis suggested that leaves of the native Q. garryana have a longer petiole compared to their blade length, while leaves of the non‐native Q. robur conform better to an ellipse (i.e., have shallower sinuses). The logistic regression model proved to be successful with leaves from GBIF sampled from the native range of the two species, with a 95% accuracy. When applied to individuals in the Portland area, the logistic regression model correctly identified all the non‐native Q. robur trees. Notably, all the individuals that were shown by genetic markers to be mislabeled in the Baltimore Woods Nature Corridor were correctly identified by our model. Most of these trees were young and located in or near an area managed with the goal of native species restoration. However, the overall success rate with Portland leaves (70%) is lower compared to that with GBIF leaves, due to several Q. garryana individuals demonstrating leaf morphology closer to that expected of the other species. This showed that some Q. garryana leaves in the Portland area had a different morphology than the ones in its native range of the entire Pacific Northwest, possibly a result of phenotypic plasticity due to differing environmental conditions under an urban setting. Nonetheless, for the purpose of detecting putative non‐native Q. robur, our logistic regression model with two morphometric traits proved to be effective, although follow‐up tests with other leaves from the same tree or genotyping with genetic markers would still be recommended to eliminate the possibility of false positives. The second contribution of this study is a novel software that streamlines the species identification process specific to the two white oak species of interest. Although several digital GMM tools for leaf morphology have been developed, when the goal is the quick and accurate identification of specific taxa of interest, such as Q. garryana and Q. robur in Portland, collecting measurements with these GMM tools and conducting statistical analysis in other programs from scratch can be time‐consuming. Furthermore, the two programs used in our morphometric analysis, MorphoLeaf and Tomato Analyzer, are not open source and are thus unable to be modified into a tool specific to Q. garryana and Q. robur. Garryanalyzer was developed as a plug‐in for ImageJ, a widely used open‐source image analysis platform. Garryanalyzer is capable of automatically calculating the two key leaf morphological traits identified through Lasso logistic regression analysis of a white oak leaf, and then predicting whether the input leaf belongs to Q. garryana or Q. robur based on a logistic regression equation. Garryanalyzer successfully replicated the measurement of the significant traits—petiole length:blade length ratio and elliptical score—and generated species predictions that were highly congruent with those from the original logistic regression model. Additionally, the measurement of best‐fit ellipse conformity is a new feature introduced by Garryanalyzer that is not present in other shape analysis base functions or plug‐ins of ImageJ. With the leaf morphometric analysis of two white oaks of the Pacific Northwest and the development of Garryanalyzer, we demonstrate a workflow for customizing a specialized digital tool to a few taxa of interest by narrowing down quantitative morphological traits. Such a workflow is useful in the conservation of non‐model plants, especially those involving native and non‐native species that are taxonomically close and morphologically similar. While Garryanalyzer was designed to distinguish between only Q. garryana and Q. robur, it demonstrates a framework that can be followed and modified for morphometric studies in other non‐model plants. Even though Q. garryana and Q. robur are not known to hybridize, the workflow of Garryanalyzer can be applied to identifying non‐native plants that could potentially hybridize with native species. Because Garryanalyzer is able to automatically extract blade shape, it can also be used for the general purpose of measuring some leaf traits that are applicable to many taxa. In addition to computing the petiole length:blade length ratio and elliptical score, a few other shape homogeneity measurements can also be performed on the extracted blade outline using ImageJ's Analyze Particles base function. These parameters—circularity, solidity, roundness, and aspect ratio—are referred to as shape descriptors and are conceptually analogous to some lobe and sinus‐related measurements in MorphoLeaf (Appendix S6, p. 15). Because Garryanalyzer already performs automated extraction of the blade outline, the particle measurements are also automatically measured and included in the output. Together with the elliptical score generated by Garryanalyzer, the particle measurements are capable of quantifying whether a leaf blade is elongated or circular (roundness and aspect ratio) and the homogeneity of the blade outline, represented by the overall number and size of lobes or serrations (circularity, solidity, and elliptical). The ImageJ plug‐in MuLES automates the particle measurements, with the incorporation of Morphology and BioVoxxel Toolbox, two comprehensive libraries of shape extraction and analysis functions in ImageJ (Landini, 2008; Brocher, 2022; Bowman et al., 2023). While MuLES works efficiently with automated analysis of large numbers of leaf images, occasionally its results need to be manually adjusted at the very end of the automation due to inaccuracy in blade shape capturing. When there are fewer leaves to analyze, Garryanalyzer may be a preferable option because it performs shape extraction and analysis on individual images with single leaves, providing the users with more stopping points to make necessary adjustments. By combining Garryanalyzer, MuLES, and the particle analysis function of base ImageJ, morphometric analysis of many other taxa can be performed in an open‐source environment, as a substitute for the closed‐source MorphoLeaf and Tomato Analyzer. For our initial collection of leaf morphometric data in Q. garryana and Q. robur, we prioritized MorphoLeaf over MuLES due to the postulation that individual lobe and sinus measurements might be distinguishing features between the two Portland oaks, but this turned out not to be the case. With elliptical score being one of the two distinguishing features identified by Lasso, it seems that the overall conformity of the blade shape (i.e., the collective of all lobes and sinuses, instead of separate measurements of individual ones) is more important in distinguishing taxa using white oak leaves. For other taxa that have “toothed” (lobed or serrated) leaves, Garryanalyzer's elliptical score and ImageJ's particle measurements are capable of describing the depth and density of the “teeth” of the blade relative to its area, convex shape, and best‐fit ellipse. Although these measurements reflect only the overall homogeneity of the blade shape instead of individual “teeth” sizes, as long as the pattern of lobes or serrations is relatively consistent on the same leaf and within the same taxon, as is the case with many species with “toothed” leaves, analysis of overall homogeneity of the blade shape should be sufficient for taxa‐level distinctions. Nonetheless, the actual effectiveness of Garryanalyzer and ImageJ's particle analysis in distinguishing between other taxa, especially those with serrated leaves, remains to be tested in future experiments. For taxa with compound leaves, on the other hand, testing will need to be done with all morphometric software to determine the most optimal distinguishing features. As an open‐source software based on a widely used image analysis platform, Garryanalyzer provides a modular framework that can be easily customized for digital morphometrics in other non‐model plants. The source code of Garryanalyzer also includes a detailed annotation for this purpose (see the Data Availability Statement). There remain a few improvements that can be made to future versions of Garryanalyzer. The first improvement pertains to the analysis method used and the predictive model generated by this study. This may be especially important for improving the accuracy of species prediction of white oaks in the Portland area. Both the GBIF and Portland datasets contain a limited number of sampled trees, which can be expanded through future collection and analysis. Moreover, as we elected to maximize the number of trees to be analyzed within our available collection from Portland, we measured only one leaf per tree. Our experiments operated under the expectation that the variation in morphology for mature leaves sampled from different trees will be at least as great as the variation in morphology among mature leaves sampled from a single tree. Because our analysis utilized a single leaf per tree, we cannot conclude whether the correct versus incorrect assignment of a leaf to species is representative of all leaves on those trees, or whether, instead, a small proportion of leaves from all trees might be incorrectly assigned by our model. Although, with our current data, the observation of significant differences in several morphometric traits between the two species suggested a plausibly high degree of interspecific variation, future testing with more individuals and more leaves from the same individual could further corroborate our predictive model. The second improvement pertains to the applicability and accessibility of the Garryanalyzer software. The effectiveness of Garryanalyzer on species complexes with more than two taxa could be assessed by performing tests on multiple closely related Quercus lineages, as a starting point. A future version of Garryanalyzer could also incorporate parameters that are not currently covered by the extended ImageJ platform, such as individual “tooth” measurements from MorphoLeaf and other homogeneity and asymmetry measurements from Tomato Analyzer. Additionally, integration with other open‐source image analysis platforms such as PlantCV (Gehan et al., 2017) would increase the accessibility of Garryanalyzer to researchers of non‐model plants. AUTHOR CONTRIBUTIONS Z.X. designed experiments, helped with field collection, performed genotyping analysis, and conducted morphometric analysis. T.B. developed the Garryanalyzer software, edited images used in morphometric analysis, and helped with field collection. K.K. designed experiments, conducted field collection, helped with genotyping analysis, and supervised the project. The manuscript was written by Z.X. with sections contributed by T.B. and K.K. All authors approved the final version of the manuscript. Supporting information Appendix S1: Locations of GBIF samples from (A) Quercus robur and (B) Q. garryana. Appendix S2: Location of sampled parks and natural areas in Portland, Oregon, USA. Appendix S3: Shapes used to test the mechanism of ellipsoid score in Tomato Analyzer. The values in the image are elliptical scores generated by Garryanalyzer, which were programmed to work in a way similar to Tomato Analyzer. Appendix S4: Basic information and logistic regression fit result of the GBIF samples. Appendix S5: Basic information and logistic regression fit result of the Portland samples. Appendix S6: Installation and use of Garryanalyzer. Appendix S7: Sequence information of the two DNA markers used in genotyping. Video S1: Animation demonstrating the mechanism of elliptical score generation in Garryanalyzer. ACKNOWLEDGMENTS The authors thank Albyn Jones for his help with statistical analysis. We thank Mary Bushman, Laura Guderyahn, Angie Kimpo, and Dominic Maze (City of Portland) for introducing us to the research question, helping us identify sampling sites, and assisting with the collection of oak samples. We thank Editor‐in‐Chief Briana Gross, Associate Editor Aarón Vélez‐Ramírez, and two anonymous reviewers for their constructive suggestions, which improved this manuscript. We thank members of the Cruzan Lab for providing helpful comments on an earlier version of this manuscript. Lenny Blair assisted with oak sampling in the field. Victoria Sork provided helpful guidance for DNA extraction methods. This research was supported in part by the Betty C. Liu Summer Research Fellowship Fund and the Ruben Chair Faculty Research Fund of Reed College. Xie, Z. , Bahar T., and Karoly K.. 2026. Garryanalyzer: A morphometric workflow and open‐source ImageJ plug‐in for quantitative morphological analysis of Pacific Northwest Quercus leaves. Applications in Plant Sciences 14(4): e70069. 10.1002/aps3.70069 This article is part of the special issue “Non‐model plants: Challenges and solutions.” DATA AVAILABILITY STATEMENT The source code and installation instructions for Garryanalyzer can be accessed on GitHub at https://github.com/zxie8561/Garryanalyzer. The DNA marker sequences used for genotyping are provided either as NCBI accession numbers or as aligned sequences in Appendix S7. All images used in the morphometric analyses, both original and modified, are available on Zenodo (https://doi.org/10.5281/zenodo.17462266). The full dataset of leaf morphometric measurements of both GBIF and Portland samples, R code for Lasso analysis, and other miscellaneous files are available on a separate Zenodo repository (https://doi.org/10.5281/zenodo.17546152). REFERENCES Associated Data Supplementary Materials Appendix S1: Locations of GBIF samples from (A) Quercus robur and (B) Q. garryana. Appendix S2: Location of sampled parks and natural areas in Portland, Oregon, USA. Appendix S3: Shapes used to test the mechanism of ellipsoid score in Tomato Analyzer. The values in the image are elliptical scores generated by Garryanalyzer, which were programmed to work in a way similar to Tomato Analyzer. Appendix S4: Basic information and logistic regression fit result of the GBIF samples. Appendix S5: Basic information and logistic regression fit result of the Portland samples. Appendix S6: Installation and use of Garryanalyzer. Appendix S7: Sequence information of the two DNA markers used in genotyping. Video S1: Animation demonstrating the mechanism of elliptical score generation in Garryanalyzer. Data Availability Statement The source code and installation instructions for Garryanalyzer can be accessed on GitHub at https://github.com/zxie8561/Garryanalyzer. The DNA marker sequences used for genotyping are provided either as NCBI accession numbers or as aligned sequences in Appendix S7. All images used in the morphometric analyses, both original and modified, are available on Zenodo (https://doi.org/10.5281/zenodo.17462266). The full dataset of leaf morphometric measurements of both GBIF and Portland samples, R code for Lasso analysis, and other miscellaneous files are available on a separate Zenodo repository (https://doi.org/10.5281/zenodo.17546152).