Developing objective tools to study rock hyrax (Procavia capensis) behaviour in the field

TL;DR

This paper creates a standardized ethogram for rock hyrax behavior to improve consistency in behavioral studies.

Contribution

The novel contribution is a validated ethogram with 42 behaviors grouped into six categories for rock hyrax.

Findings

A validated ethogram with 42 distinct behaviors was developed for rock hyrax.

Six behavioral categories were identified and statistically validated using multiple methods.

Seasonal behavioral patterns matched previous studies, confirming biological relevance.

Abstract

Animal behaviour research seeks methodological rigor and consistency within and across studies, yet maintaining shared standards requires continual evaluation. The rock hyrax (Procavia capensis) is a social mammal whose behaviour has been widely studied, but no standardized methodological ethogram currently exists for this species. Here, we developed an ethogram for the rock hyrax using a procedure designed to produce objective and reproducible behavioural classifications. We observed and filmed four groups of rock hyrax using focal animal sampling and constructed the ethogram based on two organizational rules: the level of behavioural description and the behavioural units. We identified 42 distinct behaviours and grouped them into six behavioural categories based on context. This grouping was statistically validated using correspondence analysis, multidimensional unfolding, and a cluster variables algorithm. In addition, we performed a biological validation by comparing seasonal behavioural patterns with those reported in previous studies of this system. The resulting ethogram, together with its validated behavioural categories, provides a standardized framework that can improve objectivity and consistency in future studies of rock hyrax behaviour. The approach described here is broadly applicable to other species and may facilitate comparative research across taxa.

Figures

Terms attributed to the three body parts positions.Colours denote the different body parts and pictures illustrate the positions. The positions assigned to the head: on the ground, low, parallel to the ground, or high. The positions assigned to the trunk: on the ground, inclined, or parallel to the ground. The positions assigned to the legs: extended (to the side/behind), folded (below the trunk), or standing.

Multivariate analysis for behaviour constellations (blue circles) and categories (red crosses).Correspondence analysis plots are provided for coordinates 1 and 2 (a), coordinates 1 and 3 (b) and coordinates 2 and 3 (c). The amount of variance explained by each coordinate is denoted in parenthesis. Multidimensional unfolding map for behavioural constellations based on Jaccard similarity distance (d) and Bray-Curtis dissimilarity distance (e). The measures of stress and dispersion accounted for (DAF) are provided in parenthesis. Clustering of behavioural constellations by the Cluster Variables SAS PROC VARCLUS algorithm (f). Coefficient level for each behavioural constellation and cluster are denoted by colour and details are provided in the legend. The amount of variance explained within cluster and in total are provided in parentheses.

Observations of behaviours by month.(a) The frequency of occurrence (i.e., number of cases observed) for each of the seven behavioural categories by month during the 2024 field season. (b) Correspondence analysis showing the significant association (χ224 = 220.9, P ≤ 0.0001) between the behavioural categories (red cross) and months (blue circle). In parentheses the amount of variance explained by each coordinate.

Full text

Introduction

Language uses specific words to construct sentences and transfer information. While each word is defined, the meaning of sentences depends on context, which may be subjective and influenced by the experience and emotional state of both the writer or speaker and the reader or listener [1]. Similarly, the study of animal behaviour requires the creation of standardized terms and detailed descriptions of behaviours, a convention known as an ethogram [2]. The task of labelling and defining behaviours is particularly challenging in social species with complex behavioural repertoires. For example, “open mouth” has been described either as a distinct behaviour [3] or as one element within the broader description of attack behaviour [4].

It can be useful to divide behaviours into static (e.g., standing, sitting) versus dynamic (e.g., walking, jumping) [5–7], or into states (i.e., behaviours whose duration can be measured) versus events (i.e., instantaneous behaviours whose duration is difficult to measure) [2]. Static behaviours are typically classified as states, whereas dynamic behaviours are classified as events, such that one classification scheme is often sufficient. Furthermore, the level of descriptive complexity should be consistent throughout an ethogram. For example, walking can be described as a dynamic posture [7], a slow type of locomotion [8], or as a series of limb positions—head forward, neck intermediate, legs intermediate, and toes alternating in forward and downward positions [9].

Ethograms may also group behaviours into categories based on function or context, a process that may introduce subjectivity [8,10–13]. Thus, objectively categorizing and interpreting behaviour remains a central challenge in behavioural research and can benefit from tools such as statistical similarity [3] and analyses of behavioural sequences [14]. For example, based on statistical similarity, grimace can be considered a submissive behaviour if grouped with retreat and ears back [3]. Similarly, if behaviours that occur sequentially belong to the same category, behaviours preceding a fight may be considered aggressive [15].

Rock hyraxes (Procavia capensis; hereafter hyrax) are social, diurnal mammals [16] whose behaviour has been widely studied since the 1960s, yet no standardized ethogram exists for this species. Each study defined its own set of focal behaviours, resulting in a total of 24 behaviours across studies (Table SI1), often using different definitions for the same behaviours. The purpose of this study was to create a standardized ethogram, comparable to those developed for well-established model species (e.g., zebrafish Danio rerio) [4], for use by rock hyrax researchers.

In this study, we describe a structured process for developing an objective ethogram, which may also be applicable to other species. We first defined behaviours as static, dynamic, or complex (i.e., constellations composed of multiple behaviours). For each behaviour, we described the position of the trunk, legs, and head. We then classified behaviours according to context and tested the objectivity of this categorization using three approaches: correspondence analysis, multidimensional unfolding, and the cluster variables algorithm. Finally, we biologically validated the categorization by comparing the temporal occurrence of behavioural categories across sampling months with seasonal patterns reported in the literature (Table SI1).

Materials and methods

Species, study site and field procedures

The rock hyrax is widespread across Africa and the Middle East in rocky terrain [16–17]. It lives in crevices that provide protection from predators and environmental conditions [17–18]. The typical hyrax group includes a resident male, adult females, juveniles and pups [18–19].

This study is part of the long-term hyrax study, ongoing since 1999 in the Ein Gedi Nature Reserve (31°28′N, 35°24′E), west of the Dead Sea in the Judean Desert, Israel. Hyraxes have been studied for six months each year (March to August). In the 2024 field season, four social groups were observed (Table SI2) five days a week, in the mornings, beginning before sunrise and continuing until all hyraxes were either in shelter or inactive for 30 minutes. All groups included pups that were born in March. All hyraxes in the study were captured using live box traps (Tomahawk Live Trap Co., Tomahawk, WI, USA), tagged with a unique collar and/or earrings to allow behavioural observations [20]. We observed hyrax using focal animal sampling [21], which records all behaviours of a focal individual over a pre-established period. In this study, we observed and filmed hyrax behaviour using a video camera (Nikon CoolPix P1000) for 10 minutes or until the focal individual moved out of sight (following [22]).

Ethogram rules

To create a consistent ethogram, we described the behaviour on two levels:

We defined the behaviour as static (e.g., stand, sit, lie) or dynamic (e.g., run, walk, jump) [6–7]. Complex behaviours that combined several static or dynamic behaviours were defined as constellations (e.g., chase, play, nurse) [13].For each static or dynamic behaviour, we specified the position of three body parts: the trunk, the legs and the head in relation to the ground (Fig 1) [5,6,9].

Terms attributed to the three body parts positions.Colours denote the different body parts and pictures illustrate the positions. The positions assigned to the head: on the ground, low, parallel to the ground, or high. The positions assigned to the trunk: on the ground, inclined, or parallel to the ground. The positions assigned to the legs: extended (to the side/behind), folded (below the trunk), or standing.

Behavioural categories

Videos (193 ± 258 seconds, 2.3 ± 2.6 behaviours) were manually annotated using the BORIS software [23]. Using the data from BORIS, we constructed a binary matrix where the columns are the presence or absence of behavioural constellations during a session, and the rows are the observation sessions. All behaviours were given an equal weight in each session. We assigned each session to one of six behavioural categories (affiliative, agonistic, foraging, mating, social, vocalizations) by the context and our experience with the study species. To evaluate the robustness of these categories we used three methods. First, we used correspondence analysis (CA) and the presence/absence matrix of behavioural constellations to test for association between behavioural categories and constellations. Specifically, we aimed to identify behavioural constellations that are associated with more than a single behavioural category. CA is conceptually like principal component analysis (PCA) for categorical (i.e., presence/absence) rather than continuous data. Then, we used two multivariate ordination procedures to classify constellations into clusters using the presence/absence matrix of behavioural constellations in the 156 video sessions. In the first approach we calculated the Jaccard similarity distance and the Bray-Curtis dissimilarity distance between the 23 behavioural constellations in SPSS (version 29, IBM Inc.). These two distance measures are the most frequently used for binary data. Using multidimensional unfolding (PREFSCAL, SPSS), we projected the Jaccard and Bray-Curtis distance matrices onto two-dimensional space. The stress measure (SM) and the dispersion accounted for (DAF) were used for evaluating the fit of projection of the two distance matrices. In the second approach, we used the cluster variables algorithm, which is a dimension-reduction method that constructs clusters by linear combinations of variables through a series of iterative steps and orthoblique rotation using SAS PROC VARCLUS [24], implemented in JMP Pro (version 18, SAS Institute Inc.). Unlike principal component analysis (PCA), this algorithm iteratively splits clusters of variables and reassigns variables to clusters until no more splits are possible. The cluster components produced in this analysis are not orthogonal but explain as much of the variation as possible among the variables within each cluster [24].

To examine the temporal congruency between previous studies of hyrax behaviour (Table SI1) and the behavioural categories in our ethogram, we counted the occurrences of the seven behavioural categories per month (March, April, May, June, July) for the 2024 field season. Then, we tested the association between each behavioural category and month using contingency table exact chi-square statistics and correspondence analysis (CA) in JMP (Version 18 Pro; SAS Institute Inc., USA).

Ethical note

The hyrax research is conducted under annual permits from the Israeli Nature and Parks Authority (NPA) for capturing, handling and tagging at the Ein Gedi Nature Reserve (2024/43519). We have been using collars to identify individual hyrax since 2000. Before starting to use them on the study population we observed hyrax social behaviour with and without the collars on captive hyraxes and found that the collars do not change the behaviour. All procedures performed in the hyrax research in general are in accordance with the ethical standards of the ARRIVE guidelines. This specific study does not include experimental procedures. It involves behavioural observations from a distance (~ 50 meters) using binoculars, telescope, and a video camera.

Results

Ethogram

Using 1528 behaviour videos, we described 7 static and 11 dynamic behaviours (Table 1). Based on these, we defined 24 constellations, which we grouped into 6 categories (Table 1): affiliative behaviours (n = 4), foraging behaviours (n = 3), agonistic behaviours (n = 4), social behaviours (n = 3), mating behaviours (n = 6), and vocalization (n = 3). An additional behaviour (sand bathing) was categorized as miscellaneous.

Table 1: Rock hyrax ethogram: static positions, dynamic positions and behavioural constellations. For each behaviour there is a definition and a link to a picture or a short video with an example of it.

Association between behavioural constellations and behavioural categories

Correspondence analysis (CA) based on the frequency of occurrence of each constellation over all events and each of the seven behavioural categories showed that the first three coordinates (i.e., vectors) accounted together for 67.6% of the total variance (Fig 2a - 2c). CA showed a significant and clear association between constellations and the assigned behavioural categories (n = 284, χ^2^132 = 899.1; P < 0.0001) except for two cases (social and agonistic and affiliative and miscellaneous categories). In all three CA plots, the same behavioural constellations are shared for the social and agonistic categories, and the same behavioural constellations are shared for the affiliative and miscellaneous categories (Fig 2a - 2c).

Multivariate analysis for behaviour constellations (blue circles) and categories (red crosses).Correspondence analysis plots are provided for coordinates 1 and 2 (a), coordinates 1 and 3 (b) and coordinates 2 and 3 (c). The amount of variance explained by each coordinate is denoted in parenthesis. Multidimensional unfolding map for behavioural constellations based on Jaccard similarity distance (d) and Bray-Curtis dissimilarity distance (e). The measures of stress and dispersion accounted for (DAF) are provided in parenthesis. Clustering of behavioural constellations by the Cluster Variables SAS PROC VARCLUS algorithm (f). Coefficient level for each behavioural constellation and cluster are denoted by colour and details are provided in the legend. The amount of variance explained within cluster and in total are provided in parentheses.

Classification of behavioural constellations

We attempted to cluster the behavioural constellations based on resemblance of occurrence using two approaches: multidimensional unfolding and the cluster variables algorithm. The multidimensional unfolding plot for both distance measures showed a very low stress values (0.04) and nearly all the variance in the distance matrices was accounted for in the plots (DAF = 0.96; Fig 2d - 2e). Both plots showed a separate clustering for the mating and agonistic constellations. However, the constellations assigned to affiliative, foraging, miscellaneous, social, and vocalizations were not clustered by behavioural categories but formed two clusters of a mix of constellations that are associated with social, affiliative, and foraging (Fig 2d - 2e).

The Cluster Variables SAS PROC VARCLUS algorithm constructed a more structural set of clusters (Fig 2f), which accounted for 51% of the variance in the data (S1 Fig in S1 File). Cluster 1 is a set of constellations associated with mating behaviour. Cluster 2 clumped all the constellations associated with agonistic behaviour. Cluster 3 is a collection of behaviours related to maternal care. Foraging behaviours are united under Cluster 4. Cluster 5 and 7 combined behaviours that are related to stressful situations in adults and pups, respectively. Finally, Cluster 6 combined behaviours that are associated with male competition.

Temporal analysis of behavioural constellations

During the 2024 field season, parturition spanned late March to April, followed by extensive maternal care (Fig 3a). Maternal care gradually decreased throughout the field season, which ended in late July. Matings started in late June, with mating-related behaviours increasing in frequency to the end of July (Fig 3a). The contingency table showed a significant association between behavioural category and month (n = 1785, χ^2^24 = 220.9, P ≤ 0.0001; Fig 3b). The CA showed an association between affiliative behaviours and the month of April, sand bath and the month of May, social, agonistic and foraging behaviours and the month of June, and mating and vocalization behaviours and the month of July (Fig 3b).

Observations of behaviours by month.(a) The frequency of occurrence (i.e., number of cases observed) for each of the seven behavioural categories by month during the 2024 field season. (b) Correspondence analysis showing the significant association (χ224 = 220.9, P ≤ 0.0001) between the behavioural categories (red cross) and months (blue circle). In parentheses the amount of variance explained by each coordinate.

Discussion

An ethogram is a fundamental tool for objective animal behaviour studies [25–26]. Here, we developed an ethogram for the rock hyrax, a species whose behaviour we have studied since 1999, and defined and classified 42 behaviours according to strict organizational rules. These behaviours were grouped into six behavioural categories, which were validated using three statistical approaches. All three methods provided a good representation of the data, indicating that the resulting categorization captured the main behavioural structure. The high proportion of explained variance in the correspondence analysis (CA) suggests that the extracted dimensions accounted for most of the behavioural variation. The clear association between constellations and behavioural categories indicates that the constellations were appropriately classified and that the predefined categories are representative of hyrax behaviour. The multidimensional unfolding analysis revealed clear clustering of mating and agonistic constellations. In contrast, constellations that were initially assigned to social, affiliative, and foraging categories were distributed across clusters dominated by vocalization and agonistic behaviours. This pattern likely reflects the temporal overlap of social and foraging behaviours within the daily activity routine of hyraxes [27]. Similar results were obtained using the cluster variables algorithm, which also identified distinct clusters for mating and agonistic behaviours, while foraging, social, affiliative, and vocalization behaviours were distributed across other clusters. Across the three clustering approaches, mating emerged as the most robust behavioural category, followed by agonistic (supported by CA and the cluster variables algorithm), vocalization (supported by CA and the cluster variables algorithm), and foraging (highlighted by CA). In contrast, social and affiliative categories showed substantial overlap, and the clustering methods produced inconsistent results for these behaviours. By integrating the results of all three analyses, we were nevertheless able to assign each constellation to a single behavioural category with greater confidence. Consequently, the pre-established behavioural categories were revised: play and nursing behaviours were reclassified as affiliative behaviours, and back-to-back behaviour was reassigned from the agonistic to the mating category, as supported by the clustering analyses.

The temporal analysis provided a biological validation of the categorization, demonstrating that the seasonal patterns observed during the 2024 field season closely matched those reported in previous studies (Table SI1). For example, in 2024, parturition occurred in early April, and accordingly, babysitting, nursing, and affiliative behaviours declined from May onward, reaching low levels by the mating season in July. Nursing is expected to decline once pups are weaned at around 2–3 months of age, typically before the mating season [17]. We also observed that social and agonistic behaviours increased from March to June, consistent with patterns reported in the literature [17]. Mating behaviours were highly seasonal, showing a marked increase throughout the mating season and peaking in July, when mating activity was most prevalent [17,28]. Vocalizations also increased in July, coinciding with the mating season. Singing in hyraxes is thought to signal male quality to rivals and potential mates [29–30]. Finally, foraging behaviours peaked between May and July in 2024, although comparable seasonal trends were not documented in earlier studies [17,27].

Overall, clustering results indicated that most behavioural categories were consistent with the literature and broadly supported across the three analytical approaches. However, some constellations were difficult to assign to a single category because they occurred in multiple contexts. For example, approach behaviour can occur in both mating and social contexts. In addition, some apparent inconsistencies may stem from differences in terminology across studies. Behaviours such as grooming [17,31] heaping [31], submissive behaviour [17], courtship [28], and vigilance [17,31,32] are not explicitly labelled in our ethogram, although the behaviours they describe are included. For instance, grooming represents a subjective interpretation (e.g., cleaning) and was therefore omitted as a label. Instead, the underlying behaviour: scratching with teeth, paws, or an object, is explicitly described. Courtship, previously defined as a male and a female approaching and following each other while sniffing and/or licking [28], comprises several constellations in our ethogram (approach, follow, sniff). This suggests that behavioural levels above constellations may exist, and that alternative hierarchical organizations of behavioural units are possible. Vigilance is another example of a subjective interpretation that we omitted. However, our ethogram includes the static behaviour stand with the head parallel to the ground, which captures the same mechanical behaviour without relying on contextual interpretation. Human observers are prone to assigning meaning to behaviours, which can result in over-interpretation and reduced objectivity.

In this study, we propose a procedure for defining and objectively testing the clustering of complex behaviours into categories, beginning with descriptions based on body-part positions and clearly defined behavioural units. Alternative approaches are possible, including the use of higher levels of descriptive complexity that incorporate additional body parts (e.g., mouth, eyes, nose, tail, and neck movements) [5,9]. Other methods, such as alternative statistical clustering techniques [3,33,34] free-choice profiling to assess behavioural intensity [35] or refined behavioural-unit frameworks with clearer boundaries [36], may further improve ethogram accuracy. Here, we present a comprehensive ethogram for wild rock hyraxes using a method designed to be simple and reproducible, thereby minimizing variability and enhancing objectivity. The ethogram was validated through temporal analyses, demonstrating its reliability. Together with the accompanying video material, this ethogram can be used to train new researchers and to assess inter-observer reliability prior to future studies. We suggest that this methodological framework may be applicable to other systems and broadly useful in animal behaviour research.

Supporting information

S1 FileCluster Variables (SAS PROC VARCLUS) model results. Correlation color map for the 23 constellations.Cluster assignment of all the constellations, R2 of each constellation with its own cluster, and with the next best cluster are listed on the left side of the map.(ZIP)