Morphometric Properties of Olive (Olea europaea) Pits: A Dataset for Cultivar Identification and Analysis

Abstract

Image analysis of pits and grains provide alternative routes for overcoming the invasive approach of genomic tools in the investigation of archaeological or modern plant material, which is only seldom a viable option due to the complex and laborious methodologies required. Nevertheless, any investigation of pit morphology and cultivar interpretation requires a high quality, comprehensive dataset for comparison. Such a benchmark dataset for the morphology of olive (Olea europaea) pits is presented in this paper, designed to facilitate similar research and establish a base for future investigations. The dataset was established by image analysis of pits of 18 olive cultivars that were photographed in both lateral and dorsal positions. A dedicated MATLAB® code was developed to extract the silhouettes of each pit and to calculate 16 morphometric traits of each view of the pit. Altogether, a total of 1008 photos of 504 pits of the 18 cultivars, together with their detailed morphometric description and statistical analysis are available here. These were used to test the accuracy of the dataset and the new approach in representing the different cultivars.

Background & Summary

The olive tree (Olea europaea var. europaea) has been an important member of the Mediterranean natural forest and rural landscape for millennia, as well as a central pillar of the Mediterranean agricultural heritage¹. The Olea database (http://www.oleadb.it/) lists several hundred olive cultivars that are grown over nine million hectares (https://www.fao.org/).

Genetically inherited morphometric characteristics of the fruits and their pits are widely used for cultivar identification^2–5. Recently, by implementing a machine learning approach, Blazakis et al.⁶ have shown that shape variation of olive pits, fruits and leaves, can discriminate between 14 different cultivars with high accuracy⁶. Morphometric characteristics have been widely used for examining archaeobotanical findings in their geographical and historical contexts to determine environmental adaptation and cultivation practices^3,4,7,8. Recent advancements in imaging and statistical methods are improving the accuracy and applicability of morphometric analysis. Morphometric measurement is especially useful for quantitative analysis of large samples or when Paleogenomic data is limited due to DNA degradation⁹. As such, morphometric analysis of ancient olive pits excavated in Neolithic sites at the Carmel coast (6–8 kya before present) pointed at the utilization of wild olives (var. sylvestris) for the production of table olives and olive oil^3,10.

Here, we present a new open dataset of morphometric traits of olive pits. The data represents 18 different cultivars originating from different geographical regions around the Mediterranean Basin, sampled during a single year from a single plot. The dataset includes 32 different morphometric descriptors of 390 pits that passed our statistical testing, serving as a benchmark for morphometric studies involving either modern, or archaeobotanical material. We provide 1008 images of 504 pits (including pits that were removed as part of the dataset refinement through statistical testing) before and after processing, photographed in both dorsal and lateral views. Furthermore, a relevant analysis of variance of the resulting dataset is presented in order to refer to data that showed larger variance than expected, and therefore was not considered to represent the specific cultivars. By openly publishing this morphometric modern reference collection we aim to assist in implementing the proposed methodology, encourage and facilitate other research studies on this monumental and important crop. Beside archaeological application, morphometric analysis of olive pits offers a practical cost-efficient tool for proof identification of the of Mediterranean olive germplasm, particularly traditional living ancient olive trees of unknown heritage. Identification of cultivar identity in germplasm collections mainly relies on laborious genetic tools which require complex infrastructure and specialized analysis expertise. The morphometric approach suggested here provides an alternative cost-effective approach. This approach is valuable for preliminary screening of large-scale germplasm collections, identification of mislabeled accessions or field assessment of cultivar diversity in traditional groves, where molecular analysis of a large number of individuals may be too expensive and time consuming. In spite of its limitations, such as morphological overlap among cultivars, morphometric identification offers an accessible screening tool that may help researchers in ambiguous cases, therefore optimizing resource allocation in germplasm curation and diversity assessment programs. Part of the dataset presented here has already been successfully utilized for the identification of Olea europaea subsp. europaea var. sylvestris population in Atlit, Israel¹¹ and in examining the geographical origin of a non-fossilized waterlogged ancient olive cargo in the Ma’agan Mikhael B merchant shipwreck (600–700 AC)¹². Morphometric assessment of archaeological olive pits has proven effective in categorizing ancient pits as the wild ancestor of cultivated varieties^2,4. However, the majority of archaeobotanical assemblages consist of charred pits, which likely underwent distortion and shrinkage during carbonization. Such taphonomic alterations raise concerns regarding the direct application of standard morphometric protocols to ancient material. To address these challenges, analysis of archaeobotanical remains should put more emphasis on additional dimensional ratios and shape descriptors, an approach successfully demonstrated in the study of Phoenix seeds¹³. Consequently, while our dataset offers promising applications for archaeobotanical research, its validation on charred archaeological material warrants that further investigation should be directed to determine which morphometric features withstand various forms of alteration¹⁴.

Methods

Olive pits inventory

Fruits of 18 Olea europaea cultivars originating from various areas across the Mediterranean region were collected at the Volcani institute olive germplasm collection (Agricultural Research Organization, Israel) (Table 1, Fig. 1). To account for possible variability between trees of the same cultivar, 10 fruits of either two or three trees of each cultivar were collected, as available in the germplasm. Collection was conducted at the same day in November 2021, resulting in a total of 504 pits. The mesocarp (fruit flesh) of each fruit was removed and pits were clean in running water and disinfected by soaking in a 3% Sodium Hypochlorite solution before photographing.

Table 1.

Details of the olive pits used for the morphometric analysis, depicting cultivar name, its code name, country of origin and number of processed (analyzed) and tested (passed quality control) pits.

#	Cultivar	Code	Origin	Pits Processed	Pits used
1	Arbequina	ARB	Spain	30	27
2	Baladi	BAL	Palestine/Jordan	30	27
3	Chemlal	CHE	Tunisia	18	0
4	Coratina	COR	Italy	30	18
5	Frantoio	FRA	Italy	30	27
6	Gemlik	GEM	Turkey	30	27
7	Kalamata	KAL	Greece	28	18
8	Leccino	LEC	Italy	25	22
9	Landrace	MLL1	Southern Levant	30	27
10	Landrace	MLL7	Southern Levant	30	27
11	Muhasan	MUH	Palestine/Jordan	28	18
12	Nabali	NAB	Palestine/Jordan	28	17
13	Picual	PIC	Spain	30	27
14	Picholine Languedoc	PILA	France	20	18
15	Picholine Marocaine	PIMA	Algeria	28	27
16	Saiali Magloube	SAI	Tunisia	30	27
17	Shami	SHA	Egypt	20	18
18	Sorani	SOR	Syria	30	18
Total				504	390

Fig. 1.

Medoid pit of the 18 olive cultivars in dorsal view. Pits in each row are designated to cultivars, from left to right: 1–6 representing Gemlik, Baladi, Arbequina, Picual, Muhasan, Leccino; 7–12 representing Shami, Saiali Magloube, Picholine Languedoc, MLL1, Nabali, Coratina; and 13–18 representing Sorani, Picholine Marocaine, Chemlal, Frantoio, MLL7,  and Kalamata. Cultivar’s names and their country of origin are detailed in Table 1.

Imaging

Pits (endocarp) were photographed (Figs. 2 and 3) according to a standard workflow presented by Terral et al.¹⁵. Each pit was placed and stabilized on a blue cloth background and photographed in two positions capturing the dorsal and lateral views. A millimetric graphing paper with cm and mm scales was placed on the side of each photo for scaling (Fig. 3). The photos were taken using a Canon EOS 200D camera (24 MP, 6000 by 4000 pixels), equipped with an EFS 18–55 mm lens. The camera was fixed on a single-arm stand to ensure vertical photography and a stable permanent distance from the object, while being lit using two led photography lights from both sides, to prevent shadowing. In order to ensure the suitability of the images for analysis, which may be hampered by their quality, the GIMP image editing software (version 2.4.1.3) was used to clean any dirt, particles and fragments by covering them with blue pixels (Fig. 4).

Fig. 2.

Schematic workflow of the methodology.

Fig. 3.

Detailed workflow showing the image processing- (a) original image, (b) background specks and dirt removed using GIMP, (c) cropped image for analysis, (d) filling of the image and (e) the final silhouette that is being analyzed.

Fig. 4.

Example of a images after processing in GIMP software. The images represent the same pit in both dorsal and lateral views (VD and VL respectively).

Image analysis

The images were processed using a dedicated MATLAB® code, developed for this research. A scale was determined manually for each image, which allowed an accurate conversion from pixels to millimeters. Because the images of each cultivar were acquired in a single photography session, the scale was determined as the median scale determined for the entire set of images of each cultivar in order to avoid user errors. Prior to segmentation, noise was filtered using a simple median filter with a fixed size of 3 × 3 pixels. Because the cropped image contains only the blue cloth background and the pit, without the millimetric scale, each image was segmented using a built-in function for K-means clustering segmentation. After segmentation the image was binarized, filled and filtered using two median filters of increasing size (10 × 10 and 20 × 20). After eroding the boundaries, the perimeter of the pit in each image (i.e., its silhouette) was extracted for additional analyses. The resulting filled and binarized image was analyzed by collecting key morphometric descriptors using the MATLAB regionprops function and complementary key features previously used for morphometric description of zircon grains in geological research¹⁶.

Data Record

After screening of 1008 images of the total number of pits, the dataset was refined, and a data set of 16 morphometric parameters (Table 2) was established for 390 pits. All the images are available on a dedicated Zenodo repository¹⁷. The file name of each image comprises an abbreviation of the cultivar name (Table 1), tree number (a, b or c), pit number (1–30) and the pit position (VD VL for dorsal and lateral, respectively). Accordingly, the file “ARB_a_001_VD.jpg” for example, represents the “Arbequina” cultivar, tree “a”, pit #1 in its dorsal view.

Table 2.

Morphometric parameter used for analyses, their description, and formula used for its calculation.

Parameter	Description	Formula
Area	Pixel count of the pit	—
Circularity	How round the object is	$\left(\frac{4\pi \times \mathit{Area}}{{\mathit{Perimeter}}^2}\right)\times {\left(1-\frac{0.5}{r}\right)}^2$ $\left(\frac{4\pi \times \mathit{Area}}{{\mathit{Perimeter}}^2}\right)\times {\left(1,-,\frac{0.5}{r}\right)}^2$
Convex Area	Pixel count of smallest enclosing convex polygon	—
Eccentricity	Eccentricity of ellipse with same second-moments	$\frac{\sqrt{({\mathit{major\; axis\; length}}^2-{\mathit{minor\; axis\; length}}^2)}}{\mathit{major\; axis\; length}}$
Equivalent Diameter	Diameter of circle with same area	$\sqrt{\left(\frac{4\times \mathit{area}}{\pi }\right)}$
Major Axis Length	Major axis of best-fit ellipse (pixels)	—
Maximum Feret Diameter	Maximum distance between convex hull boundary points	—
Minor Axis Length	Minor axis of best-fit ellipse (pixels	—
Minor Feret Diameter	Minimum distance between convex hull boundary points	—
Perimeter	Boundary length (pixels)	—
Solidity	Proportion of convex hull filled by region	$\frac{\mathit{Area}}{\mathit{ConvexArea}}$
Form Factor	Similarity to a circle	$\frac{4\times \pi \times \mathit{area}}{{\mathit{perimeter}}^2}$
Roundness	Ellipse aspect ratio; measures elongation	$\frac{4\times \mathit{area}}{\pi \times {\mathit{major\; axis\; length}}^2}$
Compactness	Elongation relative to equal-area circle	$\frac{\sqrt{\frac{4}{\pi }\times \mathit{area}}}{\mathit{major\; axis\; length}}$
Aspect Ratio	Major axis length / Minor axis length	$\frac{\mathit{Major\; axis\; length}}{\mathit{Minor\; axis\; length}}$
Orthogonal Diameter	Longer of two perpendicular diameters through center	—

The morphometric data (converted into mm) is stored in a csv file format that includes 16 morphometric parameters for each image (as specified above). The parameters are indicated with the suffix “.VD” or “.VL” to denote a dorsal or lateral views, respectively.

The quantitative parameters are presented in millimeters (mm) for: equivalent diameter, major axis length, max. feret diameter, minor axis length, minor feret diameter, orthogonal diameter, and pit perimeter. Area-related parameters (area and convex area) are presented in square millimeters (mm²). Each of the quantitative parameters were transformed from pixels to the corresponding unit in millimeters (e.g., length, area) using relevant formulas (Table 3). Descriptive parameters with relative values between zero and one include circularity, eccentricity, solidity, form factor, roundness, and compactness. The aspect ratio is provided as a relative parameter, describing the ratio between the length and the width. Subsequently, each pit’s pair of images (lateral and dorsal) were combined, resulting in 504 rows, presenting the complete set of data for each pit, for a total of 32 parameters. The complete data is accessible on a dedicated Zenodo repository¹⁷.

Table 3.

Formulas used to transform data from Pixels to Millimeters.

	Pixel→Millimeters (mm)	Pixel→Millimeters squared (mm2)
Parameters	Equivalent Diameter, Major Axis Length, Maximum Feret Diameter, Minor Axis Length, Minor Feret Diameter, Perimeter, Orthogonal Diameter	Area, Convex Area
Formula	$\left({\mathit{Length}}_{\left(\mathit{pixel}\right)}\times \frac{1}{{\mathit{scale}}_{\left(\mathit{pixel}/\mathit{cm}\right)}}\right)\times 10$	$\left({\mathit{Area}}_{\left(\mathit{pixel}\right)}\times {\left(\frac{1}{{\mathit{scale}}_{\left(\mathit{pixel}/\mathit{cm}\right)}}\right)}^2\right)\times 100$

Technical Validation

The resulting dataset was investigated using principal component analysis (PCA) which was performed on the resulting morphometric dataset using R (4.4.2). Prior to PCA, the data was standardized to account for different measurement scales. Outlier screening was conducted using a distance-based approach in principal component space. For each cultivar, the medoid (geometric median) was calculated based on the first three principal components, which together explained over 90% of the variance. Individual pits were ranked by their Euclidean distance from their cultivar’s medoid, where the 10% of the furthermost pits were identified as outliers and have been removed from the dataset. This relatively conservative threshold was selected to eliminate morphologically exceptional individuals while retaining most of the natural variation within each cultivar. This approach was set to establish a reference dataset representing typical morphometric characteristics, while excluding pits that may represent genetic admixture or extreme developmental variants.

Following outlier removal at the individual pit level, tree level homogeneity was assessed using multivariate analysis of variance (MANOVA) on the first three principal components of the PCA (Table 4), with a significance threshold of α = 0.05. This analysis was conducted separately for each cultivar in order to identify cases where pits from one tree systematically differed from others, potentially indicating mislabeling, undetected genetic variations or other environmental effects.

Table 4.

P value of MANOVA analysis comparing the difference between different trees of each cultivar, before and after removal of inconsistent trees.

Cultivar (code)	P value (before removal)	P value (after removal)
ARB	0.0042	0.0041
BAL	0.4763	0.4384
CHE	0.0001	—
COR	0.0002	0.3204
FRA	0.0146	0.0167
GEM	0.6203	0.5043
KAL	0.2678	0.3428
LEC	0.3463	0.4045
MLL1	0.0325	0.0293
MLL7	0.0046	0.0045
MUH	0.0010	0.0607
NAB	0.0006	0.0519
PIC	0.6876	0.7022
PILA	0.1167	0.1213
PIMA	0.2999	0.2304
SAI	0.7702	0.7642
SHA	0.0865	0.0863
SOR	<0.0001	0.4004

P values below 0.05 are considered significant. Underlined values represent cultivar with significant variance between trees.

This exclusion of outlier pits and trees was adopted to establish a referential dataset that represents the core morphometric characteristics of each cultivar. Nevertheless, it should be noted that this strategy has important implications. This is because the removal of extreme morphological variants may lead to underestimation of the natural morphological variability which can potentially exclude legitimate intra-cultivar diversity. The observation of overlapping morphospaces among some cultivars (e.g., the overlapping ellipses in Fig. 6) highlights a fundamental challenge, indicating that the morphological variation within a cultivar may exceed the variation between cultivars. This suggests that morphometric identification based on individual pit measurements have inherent limitations, particularly when applied to archaeological specimens representing unknown genetic backgrounds and preservation state. Therefore, in the published dataset we also include the complete dataset comprising 504 pits, prior to the abovementioned identification and exclusion.

Fig. 6.

PCA analysis performed on 32 parameters per pit. Pits in the plot indicate the medoid pit of each cultivar, variance ellipse for 0.5 confidence level. Medoid pits are presented for each cultivar, colored and numbered according to the legend (Table 1).

The cultivars’ Chemlal, Coratina, Kalamata, Muhasan, Nabali and Sorani showed substantial differences between individual trees, suggesting increased morphotypic variance (Table 4). The other cultivars demonstrate a more compact spread, with no clear separation of the trees within each cultivar. Cultivars that showed significant variance were further examined and outlier trees were removed from the analysis. The Chemlal cultivar was removed entirely because of inconsistency between the two examined trees, which hindered a clear representation of the cultivar. Dissimilarity between pits of tree c to these of a and b occurred in the case of Coratina, Kalamata, Muhasan, Nabali and Sorani cultivars, which could be associated with technical mistakes that occurred during planting. The removal of the outlier trees resulted in homogeneity within most cultivars (P ≥ 0.05) (Table 4), with the exception of Arbequina, Frantoio, MLL1 and MLL7 cultivars which did not meet the criteria of P ≤ 0.05. Thus, we recommend that these are not suitable for further use. These constraints arise from several possible reasons such genetic variation within the cultivar as described above.

Generally, it can be seen that the morphometric parameters form three clear clusters, the first was based on features related to pit size (e.g. length, width, diameters, area, and perimeter). The other two clusters reflect features associated with pit shape: the first includes compactness, form factor, roundness and circularity, while the second comprises eccentricity and aspect ratio (Fig. 5). The cultivars were separated into four different quartiles of the graph where cultivars in quartile 1 are elongated and wide, cultivars in quartile 2 are short and wide, cultivars in quartile 3 are short and thin and cultivars in quartile 4 are long and thin (Figs. 5 and 6). In general, groups of shape descriptors were closely correlated, and features related to size are correlated separately (Fig. 5). The pits are therefore spread along the main properties of size and shape. Overall, the PCA analysis (Fig. 6) showed relatively high level of explained variance: 58.9% by the first component and 31.9% by the second component, summing to 90.8%, indicative of the quality of data set as taxonomical descriptors suitable for the separation of olive cultivars.

Fig. 5.

Loading plot showing the vectors of the principal components analysis (PCA) for the scores of the morphometric parameters. Vector color represents the level of contribution. Percent of variance explained by each axis is indicated in the axis titles.

Limitation and consideration for archaeological applications

The presented dataset provides a comprehensive set of pit photographs and morphometric reference descriptors of modern olive cultivars, which are highly valuable for large-scale investigations of olive heritage. The high variance explained by the first two principal components (90.8%) points at its potential to differentiate modern O. europaea cultivars. However, several important limitations must be acknowledged when considering its application to archaeological material. Various post depositional processes may alter the olive pits morphology, which may obscure cultivar specific morphometric properties. Fossilization and mineralization processes, for example, may cause dimensional changes through compression and deformation of the pit structure. Burning, which is common in archaeological contexts, causes shrinkage and warping that modifies the size parameters of pits (e.g., Terral 2004; Portillo 2019), whereas desiccation over archaeological timescales may lead to non-uniform shrinkage, particularly affecting width and size measurement. Mechanical damage from soil pressure and handling during the excavation may further compromises the morphological integrity. All of the abovementioned taphonomic alterations might introduce systemic biases and increased morphometric variance that may artificially enhance the natural morphological differences between archaeological material and cultivars documented in this modern reference dataset. The overlap between cultivars in the morphospaces, combines with the effects of taphonomic processes on archaeological specimens, suggests that morphometric identification alone may have limited reliability without complementary evidence such as paleogenomics. Therefore, while this dataset provides a comparative research resource, its direct application to archaeological material requires careful consideration of the preservation context. Further research integrating both morphometric and Paleogenomic data from multiple well-preserved archaeological specimens will be essential to establish the reliability and limitations of the morphological approach for cultivar identification in archaeological context. We hope that the establishment of this open-source freely accessible dataset would encourage researchers to share additional images and data in the future, forming a community-driven asset for the research of olive, and other species as well.

Author contributions

O.B, A.D and G.BO acquired funding, and conceived the study. G.BA supplied the biological material. E.B. collected the material, prepared the samples, photographed the pits and prepared the dataset. Y.B. organized the code and organized it for compilation as a stand-alone app. E.B. and Y.B. co-developed metric calculations, wrote the code, extracted the trait data, conducted the technical validation, and wrote the manuscript with contribution from all co-authors.

Code availability

The code that was used in this work is compiled as a stand-alone software based on MATLAB “PitAnalyzer”. The software is available to download at the following repository, where any use of it should be attributed appropriately to this publication (https://zenodo.org/records/18789307).

Data availability

The data published in this paper is openly available in a dedicated Zenodo repository at 10.5281/zenodo.18789307.

Competing interests

The authors declare no competing interests.