Human disturbance and habitat structure drive eurasian otter habitat selection in heavily anthropized river basins

We evaluated otter habitat selection using occurrence, habitat use intensity and breeding data from the Tordera and Besòs river basins, which are situated inside and around the Barcelona metropolitan area (41°25’-41°52’N, 2°05’-2°51’E, Fig. 1). There are among the most anthropized river systems across Europe. Both rise in the Montseny mountains (maximum altitude: 1712 m) and run through alluvial plains with a highly altered and fragmented ecological matrix until reaching the Mediterranean Sea. The predominant uses in the alluvial plains are residential, industrial, and agricultural. The human population density ranges from 1.5 inhabitants/km² at the headwaters to 15,000 inhabitants/km² at the Besòs river mouth. Despite their small size (1038 km² and 898 km²), both basins have a great variety of river habitats due to a combination of a great diversity of biogeographical conditions and types of human alteration on a small scale. The mosaic of heavily disturbed and well-preserved reaches adds relevance to the ecological assessment because it allows working with very long environmental gradients. The human alterations include most of the impacts related to the current river habitat degradation, among others: loss of lateral and longitudinal connectivity, aquifer overexploitation, instream structure simplification, eutrophication, water pollution, recreational uses and presence of invasive species. The biogeographical conditions vary from the biological communities associated with oligotrophic and cold streams in the upper reaches to the biological communities associated with intermittent and anastomosing Mediterranean rivers in the lower reaches. The average flow is between 4 and 6 m³/s in the main courses of the two basins and varies spatially and seasonally, with severe droughts in summer and regular floods during spring and autumn. Average annual precipitations exceed 1000 mm in the upper reaches and goes down to around 600 mm near the coast.

The Eurasian otter became extinct from the study area during the second half of the 20th century, at the same time as it disappeared from many rivers in western Europe mostly due to industrial pollution and direct persecution (Roos et al. 2015). Dispersive otters have been detected in the study area regularly again since 2010, and the first instance of reproduction after local extinction was documented in 2018 (Tolrà and Ruiz-Olmo unpublished data). Thanks to a population monitoring program, we know that currently the two river basins together hold a population of about 25 otters, with an average of 3 family groups per year (2019–2021).

Significant differences in habitat characteristics between otter-present and otter-absent river reaches and between breeding and non-breeding river reaches were tested using two-sample Wilcoxon tests in order to identify which habitat variables from the complete dataset were relevant for otter breeding and presence. Afterwards we tested for correlation in habitat variables using the Spearman coefficient to reduce redundancy in the dataset (r ≥ 0.65), because the results of regression models may be affected by correlation among the covariates (Dormann et al. 2013).

To model otter habitat selection, we took two different but complementary approaches. The first approach aimed to model the relationships between otter and the main, a priori hypothesized habitat gradients or factors in anthropized rivers, i.e., productivity, habitat structure and human disturbance, using dimensionality reduction techniques to obtain appropriate latent scores for each of these three groups of variables, which were then used in a regression framework. The second approach aimed to analyse the relationships between otter and predictor variables in a structure-agnostic way, i.e., without imposing an a priori structure to the candidate variables. To do this, we entered all predictors into a variable selection procedure to obtain sets of the most parsimonious models using the information-theoretic approach (Burnham and Anderson 2002).

For the first modelling approach, principal component analysis (PCA) was used to reduce the dimensionality of the three groups of environmental variables to obtain appropriate scores for the latent habitat factors. Our interest was in the additive effect of the main human disturbance and habitat structure variables expected to have an impact on otter habitat selection. Thus, PCA was performed separately on the three data matrices: human disturbance (3 variables), habitat structure (5 variables), and biogeographical (11 variables). The broken stick model, which was used to identify the optimal number of principal components of each PCA (Jackson 1993), suggested that one component was appropriate in all three cases. The three habitat factors thus obtained were used for otter habitat selection modelling. Relationships between otter data and habitat factors were analysed using mixed-effects generalized linear models (GLMM; McCullagh and Nelder 1989) with transect and year random effects as needed to control for pseudoreplication. Two-way interactions were included in the models when found statistically significant in likelihood ratio tests. Models were fitted to the negative binomial distribution (after overdispersion was detected) to evaluate the relationship of the habitat factors to otter habitat use intensity (i.e. spraint counts). The variable «Period» (cold vs. warm) was included in all GLMM models for occurrence and habitat use intensity. GLMMs with binomial error distribution and logit link were used to estimate the strength of the associations between habitat factors and otter occurrence and between habitat factors and otter family groups. Due to the small number of otter breeding sites, this subcategory was not included in the habitat factor analyses.

For the second modelling approach, multimodel inference based on the Bayesian information criterion (BIC) was used to identify which habitat variables best explained otter habitat selection. The BIC selects more parsimonious models compared to Akaike’s information criterion (AIC) and helps reduce the number of variables for interpretation purposes (Johnson and Omland 2004; Grueber et al. 2011). We developed a set of candidate models for otter occurrence and for otter habitat use intensity using the uncorrelated habitat variables (r ≤ 0.65). We ranked candidate models using BIC weights. Models with the highest weight were better supported and explained more variance. Only candidate models with ≤ 2 BIC value compared to the best model were considered (Burnham and Anderson 2004). GLMM with negative binomial distribution was used to assess the association between covariates and otter habitat use intensity data, whereas GLMM with binomial error distribution and logit link function was used to assess the associations between covariates and otter occurrence. As above, all models included the variable «Period» (warm vs. cold) and transect random effects on the intercept. Model selection was done by exhaustive search among all model with up to five predictors (including Period) and no interactions. Due to the limited number of otter reproduction cases, the associations between covariates and both breeding sites and family groups occurrence were evaluated using univariate logistic models. For each type of breeding data, we represented and tested the significance of the predictor variables by means of a chi-squared test and assessed the goodness of fit of the models with McFadden’s pseudo-R². McFadden’s pseudo-R² values from 0.2 to 0.4 indicate good model fit (Hensher and Stopher 1979). Spatial autocorrelation of model residuals was assessed by visual inspection of residuals on spatial plots, and with Moran’s I, permutation tests, and Moran scatterplots using package spdep. We did this on separate GLM models per period and year, and generally found no convincing evidence of spatial autocorrelation to warrant more complex models.

All statistical analyses were performed in R version 4.3 (R Development Core Team, 2010) using the packages lme4 (Bates et al. 2015) and vegan (Oksanen et al. 2019).

Key habitat factors that were extracted from each data matrix were human disturbance (explained 68.33% of the matrix variance), habitat structure (explained 55.43% of the matrix variance), and habitat productivity (explained 39.78% of the matrix variance). High values of the human disturbance factor corresponded to river reaches with high human accessibility close to urban centres in areas with high human population densities. High habitat structure values corresponded to river reaches with a heterogeneous river form and abundance of large woody debris, pools, riverbank refuges, and islands and arms. High values of habitat productivity corresponded to slow flow velocity river reaches with high water temperature and conductivity and with abundant aquatic vegetation. Loadings are shown in Table 3.

Multiple regression models revealed the importance of these three habitat factors (Fig. 3; Table 4). Both otter occurrence and habitat use intensity increased in environments with high habitat productivity, low human disturbance, and high habitat structural complexity (p-value: <0.01 in all cases), with particularly strong responses at low disturbance and high productivity, as shown by the negative interaction of these two factors in both models. Both otter habitat use intensity and occurrence were significantly (p-value:<0.001) higher in the cold period (autumn and winter). The probability of family groups responded to the same habitat characteristics, with a particularly significant association to the habitat structural complexity (p-value:< 0.01, z = 2.79). These results indicated that, in addition to the productivity natural gradient, otter habitat use was mostly limited by human disturbance and structural simplification.

Four habitat structure variables were the best-supported predictors (p < 0.05 & McFadden’s pseudo-R² > 0.2) of the otter breeding probability, both for breeding sites and for presence of family groups (Fig. 5). These were riverbank refuges, large woody debris, and river form. In this light, otter breeding habitats were characterized by structurally well-preserved river reaches with a high amount of large woody debris within the riverbed, abundance of pools, presence of riverbank refuges (also including amounts of large woody debris), and a complex river form close to those configured by river natural dynamics (meandering). Pool abundance had a more prominent role on the presence of family groups, as judged by its R².

In addition to these four key breeding variables, there were other significant (p < = 0.05) but less well-supported (McFadden’s pseudo-R² < 0.2) variables. Both human accessibility and the resident human population significantly affected otter breeding, which even when selecting productive river reaches with abundant aquatic vegetation, moved away from the more human-populated and accessible river reaches. The availability of wetlands and relevant tributaries was also linked to both breeding sites and otter family groups occurrence, while water permanence and exotic vegetation were slightly associated with family groups, with these selecting river reaches with sufficient habitat stability and avoiding areas with a higher predominance of exotic vegetation (Fig. 6).

The strong influence of habitat structure on otters revealed by our analyses probably is related to the fact that a high habitat structural complexity offers more diverse foraging and resting opportunities and is associated with higher biodiversity than homogeneous environments (MacArthur 1970). Otters mainly use the interface between aquatic and terrestrial habitats (Kruuk 1995). Therefore, they require that both riparian and riverbed areas have appropriate structural characteristics within their home range. Within the riparian area, riverbank refuges played a major role in otter distribution and habitat use intensity and, together with large woody debris and river form, were also the most relevant habitat features for the selection of breeding sites in our study area. Thus, consistent with other studies that emphasise the relevance of riparian quality habitat for otters (Elmeros et al. 2006; Kruuk 2006; Weinberger et al. 2019), our results suggest that otters require a sufficient extent of well-structured riparian habitats, providing secure resting sites, protection from high floods, natal den substrates and complementary trophic opportunities. Regarding the riverbed area, otters preferred river morphologies closer to those generated by natural physical dynamics, avoiding channel incision and human-induced simplification. River hydrogeomorphic processes and river-floodplain connectivity are linked with the river form and instream structure (Frisell et al. 1986; Newson and Large 2006), which in turn are connected with diversity and abundance of ecological niches and freshwater biodiversity (e.g., Harvey and Clifford 2008) and therefore with greater accessibility to trophic resources for otters. Although hydro-geomorphologic integrity has widely recognised effects on biodiversity and functioning of river ecosystems (e.g., Elosegi et al. 2010) surprisingly little research has previously suggested associations between riverbed structural complexity and otter habitat selection at the reach scale level (but see Ruiz-Olmo and Jiménez 2008; Scorpio et al. 2016). The strong association between otters and well-structured habitats in our study area might be enhanced by the intensive channel straightening and structure simplification that occurs in large proportions of the lower-middle river reaches in the Besòs and Tordera basins, suggesting that otters tend to concentrate their activity in local, well-structured habitats patches within a less-suitable, structurally simplified habitat matrices.

As integral elements of instream structure, stream pools and large woody debris contributed to explaining all aspects of otter habitat selection and were particularly relevant for breeding site selection. Large woody debris is a recognised key component of river aquatic habitats since it promotes stepped-channel profiles, pool habitats, energy flow dissipation and organic matter accumulation, and overall provides high levels of physical diversity (Bilby and Likens 1980; Brooks et al. 2004; Roni et al. 2015), and are associated to increases in river fish, amphibian and invertebrate populations (Thevenet and Statzner 1999; Dolloff and Warren 2003; Kail et al. 2007; Schneider and Winemiller 2008; Thompson et al. 2017; Dalbeck et al. 2020), which are the main prey for otters (Mason and Macdonald 1986; Krawczyk et al. 2016). The link between otters and large woody debris could be particularly relevant in low and medium-flow river reaches, where this feature has an even greater role in shaping habitat structure and local ecosystem functioning (Dominguez and Cederholm 2000; Anlanger et al. 2022).

For its part, the major role of pools in our study area is consistent with Delibes et al. 2000 and Ruiz-Olmo et al. 2007, who suggested an association between otters and pools in Mediterranean ecosystems during the dry season. Stream pools are a relevant feature for freshwater biodiversity as their availability increases the heterogeneity of depth, flow velocity, and riverine habitats, especially in fast-flowing areas, which contribute to increased biological productivity and prey populations (e.g., Matthews 1998; Pollock et al. 2003; Cunningham et al. 2007; Smith and Mather 2013). Moreover, especially in intermittent streams, stream pools increase the abundance and resilience of aquatic and semi-aquatic fauna in low-water level scenarios (Magoulick and Kobza 2003; Davey and Kelly 2007; Beesley and Prince 2010) and increase habitat stability during the dry season (Magalhães et al. 2002), which was identified as critical for otter breeding in Mediterranean rivers by Ruiz-Olmo and Jimenez 2008. Therefore, the relevance of pools for otters could be particularly prominent in low-flow or intermittent rivers, which are progressively spreading in Europe due to drought intensification and aquifers overexploitation for irrigated agriculture (Dudgeon et al. 2006; Rupérez-Moreno et al. 2016; Marx et al. 2018).

River habitat features can vary considerably on a fine-scale (Gostner et al. 2013), shaping a river reach-scale mosaic of simple and more complex habitat structure. Our analyses suggest that, if sufficient longitudinal connectivity is maintained, otter home ranges in human-dominated riverscapes might consist of well-structured habitat patches interspersed among poorer-quality habitats. In this light, even though our results show that the highest occurrence and, above all, the highest activity and breeding probability were concentrated in well-structured habitats, otters occurred regularly in suboptimal habitats. This may partially explain the incongruences with studies that found otters in apparent low-quality areas, such as feeding grounds in heavily simplified river reaches, and even the exploitation of human-made niche opportunities found for Weinberger et al. 2016 in the Alps, or the use of poor-structured habitats by inexperienced and low-fitness individuals suggested by Ruiz-Olmo and Jimenez 2008. On the other hand, considering that the success of source populations in well-preserved habitat patches may trigger an expansion into sink populations in poorer habitats (Baltrūnaitė et al. 2009; Delibes et al. 2009; Clavero et al. 2010; Romanowski et al. 2013) it must take into account that the time of recolonisation and the source-sink population dynamics can be relevant factors in the spatial habitat exploitation by the species (Pulliam 1988). In this regard, although otter recolonisation in our study area started more than 15 years ago and the population numbers, abundance and distribution have stabilised (Tolrà and Ruiz-Olmo unpublished data), if the most structured habitat patches allow good individual recruitment, it is possible that in the future some of the less-fit individuals may be displaced, and even attempt to breed, in more poorly structured habitats. Future work is needed to disentangle interactions between otter habitat selection, population size and recolonisation time.

We found a general pattern in which otters selected areas furthest from human settlements and with lower human population density within high-order river reaches. Our results are consistent with some studies (e.g., Baltrūnaitė et al. 2009; Romanowski et al. 2013; Weinberger et al. 2019) that suggested otter sensitivity to human disturbance in addition to the factors related to environmental gradients, but contrast with other studies that found no significant relationships (e.g., Madsen and Prang 2001). Inconsistencies between studies are most likely due to poor representativeness of low anthropized areas and the application of different scales or proxies to assess human disturbance, which may bias results because each type of disturbance may have unique effects (Suraci et al. 2021). Focusing only on one proxy of human disturbance may lead to erroneous conclusions (Nickel et al. 2020). For example, distance to roads or houses was used as the only proxy for human disturbance in some studies (Durbin 1998; Weinberger et al. 2016; Juhász et al. 2013), whereas otters were not affected by distance to roads on our study, but were influenced by other human disturbance-related variables. Thus, although roads are currently the principal cause of human-induced mortality for otters (Grogan et al. 2001; Hauer et al. 2002), our results suggest that fine-scale otter habitat selection is not affected by infrastructures that do not lead to increased habitat frequentation or modification. However, we found that noticeably affected by human accessibility, which was the most relevant human disturbance-related variable for otters in our study.

Avoidance of high human-accessible river reaches suggests that otters, like other apex carnivores (Ordiz et al. 2021), are noticeably affected by outdoor recreational activities in human-dominated landscapes. Although high availability of adequate resting and breeding structures may increase otter tolerance of human disturbance (Macdonald and Mason 1994), our analyses suggest that high levels of human accessibility might prevent otter reproduction and establishment regardless of habitat quality because of their risk perception. This is consistent with Weinberger et al. 2019, who demonstrated that the availability of non-disturbed resting sites is a fundamental requirement for otters. The importance of human accessibility to otters may have been overlooked until recently because otters shape their space use by balancing the costs and benefits of the available habitats (Travis and Dytham 1999) and may use different river reaches with different characteristics for feeding grounds and resting (Sulkava 2007) so that otters exploit large areas and can regularly visit high human accessibility reaches where trophic resources are abundant, but have stronger selection against this risk at breeding and resting sites. This is analogous to other carnivores such as wolves, lynxes, and bears, which also avoid human areas especially during daytime (Ordiz et al. 2017; Ripari et al. 2022; Salvatori et al. 2023), and consistent with the fact that human disturbance can promote spatiotemporally varying habitat selection (Richter et al. 2020), in which the nocturnal activity resulting from temporal segregation would allow for spatial coexistence to some extent (Gaynor et al. 2018). In that sense, at the population level, otters might be unaffected by the existence of localised high human-accessible river reaches (e.g., near villages or fishing places) that they would avoid for resting and breeding, and instead be strongly affected by large-scale human accessibility (e.g., extensive riverwalks).

As mentioned above, due to the species high seasonal and daily mobility (Sulkava et al. 2007), otter data occurrence does not discriminate between river reaches used recurrently by floating individuals or constant transit between different habitat patches, and the otter core areas. Therefore, especially if we consider habitat requirements are more stringent for breeding than for non-breeding individuals, the conservation measures aimed at enhancing otter occurrence need not be useful for promoting otter breeding. Females with cubs have high energetic demands (Elmeros and Madsen 1999), requiring high accessibility to trophic resources (Ruiz-Olmo et al. 2001), and are very vulnerable to disturbance and predation (Durbin et al. 1996), thus being more food-limited and refuge-dependent than other individuals. Our findings show how breeding habitat selection by otters is strongly influenced by human pressures in human-dominated landscapes, resulting in a trade-off between preference for highly productive areas, situated in the lower and middle river reaches, and avoidance of structural habitat simplification and human-made disturbance. Thus, despite otters can inhabit heavily anthropized areas at coarse scales and have relative habitat plasticity for foraging (Mason and Macdonald. 1986; Kruuk 1995 and Durbin et al. 1996), have strict fine-scale habitat requirements for cubbing and den establishment area selection.

Although Weinberger et al. 2019 indicated that otter resting site selection is strongly associated with high riparian vegetation cover, our analyses revealed that otters might be more flexible in their requirements for vegetation cover, which could have masked the association with high structural complexity in previous studies. In our study females with cubs were associated with river reaches with riverbanks harbouring numerous refuges, riverbeds with abundant large wood debris and pools and with channel morphologies closer to those generated by natural physical dynamics. The fact that habitat stability and abundance of stream pools appeared to be more relevant for river reaches with family groups presence than in family core areas could indicate that females tend to carry their cubs outside natal den river reaches in areas with lentic habitats and permanent water availability, where trophic resources are more accessible and abundant throughout the year. This is consistent with studies carried out in less anthropized areas (e.g., Ruiz-Olmo et al. 2005), suggesting a general pattern.

Otters avoided river reaches close to urban centres and densely populated areas for reproductive activities but displayed no explicit aversion when dispersing or foraging. Therefore, we suggest that otter-perceived interaction risk with humans shapes their breeding habitat selection in human-dominated landscapes. The preference for low human disturbance river reaches for reproduction is consistent with the results of Beja 1996. Otters were more deterred by distance to urban centres than by roads, adapting their fine-scale spatial behaviour to their perception of the landscape of fear, showing an evident avoidance of human-accessible areas, but being indifferent to infrastructures that do not involve impacts on habitat or increased human frequentation. According to the predation risk allocation hypothesis (Lima and Bednekoff 1999), roads could act as a predictable risk that, once built, has no added impacts within the habitat, whereas human accessibility poses a recurrent unpredictable risk within the breeding habitat. Suggesting that otters could breed relatively close to human infrastructures if sufficiently secure and well-structured habitat patches are available, so that localised human accesses to habitat (e.g., fishing points) might impact otter breeding habitat selection less than extensive riverwalks, which generate large-scale disturbances. Human disturbance effects on otter reproduction might be intensified by the increasing presence of domestic dogs, numerous in our study area, which impact has been widely demonstrated for several species (e.g., Banks et al. 2007; Hughes et al. 2013) but requires further studies to properly assess its effect on otters. In this light, we encourage future studies to further investigate breeding habitat selection on a small-scale involving other anthropized river landscapes and larger numbers of breeding females.

Our findings indicate that increasing habitat structural simplification and outdoor recreational activities, although not the main factors of otter decline in the past century (Clavero et al. 2010; Roos et al. 2015) and still secondary role in some low anthropized areas (Delibes et al. 2009), may be emerging as threats for otters in lowland riverscapes situated in heavily anthropized areas. However, efforts to preserve European river habitats have so far focused above all on water quality and concentrated on oligotrophic and headwater environments (Schindler et al. 2016) leaving floodplains and their riparian habitats largely unprotected (e.g., McCluney et al. 2014; Globevnik et al. 2020). Thus, for otter recovery and prevalence, it is necessary to provide instruments that enable and encourage governmental institutions to establish novel conservation measures to protect and restore the lowland river processes and biodiversity. We believe that our study can contribute to this by guiding river management focused on the conservation of otters in human-dominated scenarios, as well as to prevent future declines in currently less anthropized riverscapes.

The preference of otters for well-structure river reaches underlines the importance of preserving riverbanks, instream structure and natural geomorphological dynamics. This requires avoiding the river straightening and bank stabilisation that are detrimental to the multiple benefits provided by lateral connectivity, which induces the creation of riverbank refuges and promotes complex riverbed forms through processes of erosion, sedimentation and meandering (e.g., Paillex et al. 2009). Furthermore, river management should rule out the removal of instream structures (e.g., large woody debris) from the riverbed and riverbanks, which is still promoted by some European river management agencies, as these elements have direct benefits for otter foraging, by constituting habitats with abundant and accessible prey (Anlanger et al. 2022), and as refuges, by providing resting and breeding sites. Drawing on this insight, habitat creation or restoration to enhance sinuosity and floodplain reconnection, reintroduce instream structures, or recover wetlands well-connected to the river systems will have relevant positive effects on otter populations, even though more superficial actions such as the construction of artificial refuges or the planting of riparian vegetation will have vague repercussions since which do not address the root causes of habitat degradation. Moreover, due to the role of wetlands as refuges and their importance for breeding (Juhásk et al. 2013), their maintenance and restoration could also be decisive for the otters in these contexts. On the other hand, our results suggest that the promotion of new riverwalks and recreational activities sites, a now usual practice in European anthropized rivers due to their attractiveness for human leisure activities (Winter et al. 2019), could lead to a drastic reduction of suitable otter resting and breeding areas through increased human frequentation and loss of refuge structures in the riverbank. Considering these, the construction of extensive riverwalks should be limited in anthropized areas, where without regulation some local authorities may extend them along the entire middle and lower river reaches.

In human-dominated landscapes, comprehensive river restoration is often not feasible due to the existence of human activities and infrastructures that disrupt ecological processes (e.g., Monk et al. 2019) and the societal demands to recreationally enjoy the natural areas (e.g., Michel et al. 2021). Our findings showed that in heavily anthropized areas otter persist may not be compatible with human activities uniformly distributed in the riverscape. Nevertheless, we demonstrate that otters can persist if they have access to habitat patches that meet their specific requirements. In this light, to make river conservation and human activities compatible in heavily anthropized basins, we suggest that a feasible formula could be to promote segregation and mosaic of river section roles. The functional mosaic could combine areas with concentrated human disturbance with river reaches with management measures to restrict outdoor recreation, such as complete closure to the public or road closures in specific time windows (Whittington et al. 2019), together with management schemes that promote habitat structural complexity and natural river morphodynamics. These protected river reaches, which could be called otter micro-reserves due to the flagship character of the species (Kruuk 2006), would comprehensively benefit the riverine biodiversity because otter is subject to common threats with many riverine biodiversity representatives to lowland river reaches, being considered an umbrella species (Bifolchi and Lodé 2005).