The role of habitat configuration in shaping animal population processes: a framework to generate quantitative predictions

By shaping where individuals move, habitat configuration can fundamentally structure animal populations. Yet, we currently lack a framework for generating quantitative predictions about the role of habitat configuration in modulating population outcomes. To address this gap, we propose a modelling framework inspired by studies using networks to characterize habitat connectivity. We first define animal habitat networks, explain how they can integrate information about the different configurational features of animal habitats, and highlight the need for a bottom–up generative model that can depict realistic variations in habitat potential connectivity. Second, we describe a model for simulating animal habitat networks (available in the R package AnimalHabitatNetwork), and demonstrate its ability to generate alternative habitat configurations based on empirical data, which forms the basis for exploring the consequences of alternative habitat structures. Finally, we lay out three key research questions and demonstrate how our framework can address them. By simulating the spread of a pathogen within a population, we show how transmission properties can be impacted by both local potential connectivity and landscape-level characteristics of habitats. Our study highlights the importance of considering the underlying habitat configuration in studies linking social structure with population-level outcomes. Supplementary Information The online version contains supplementary material available at 10.1007/s00442-021-04967-y.

animal populations. Yet, we currently lack a framework for generating quantitative 23 predictions about the role of habitat configuration in modulating population outcomes. For 24 example, it is well known that the social structure of animal populations can shape spreading 25 dynamics, but it remains underexplored to what extent such dynamics are determined by the 26 underlying habitat configuration. To address this gap, we propose a framework and model 27 inspired by studies using networks to characterize habitat connectivity. We first define animal 28 habitat networks, explain how they can integrate information about the different 29 configurational features of animals' habitats, and highlight the need for a bottom-up 30 generative model that can depict realistic variations in habitat structural connectivity. Second, 31 we describe a model for simulating animal habitat networks (available in the R package 32 AnimalHabitatNetwork), and demonstrate its ability to generate alternative habitat 33 configurations based on empirical data, which forms the basis for exploring the consequences 34 of alternative habitat structures. Finally, we use our framework to demonstrate how Introduction 43 Animals rarely move unrestrictedly, as the physical habitat environments they depend on are 44 often heterogeneous and uneven (Fahrig, 2007;Lovett et al., 2007;Kovalenko et al., 2012). By 45 shaping individual movements, the physical configuration of habitats can have implications 46 for population and community dynamics, including ecological interactions (Plitzko and  represented in the population-can increase the local spread (among immediate contacts) but 67 decrease the speed and global reach of pathogen transmission (Read and Keeling, 2003; 68 Keeling, 2005;Sah et al., 2018). However, to unravel the role of social structure in shaping 69 ecological and evolutionary dynamics, we need to also understand the mechanisms that shape 70 animal social structure. Alongside social decisions, features of the physical habitat 71 environments can play a major role in shaping where animals move, who they (re-)encounter, 72 and how often they interact with one-another . For example, a study in sleepy 73 lizards (Tiliqua rugosa) found that habitats with more barriers increased the rates of encounters  80 Farine, 2020). However, we also require quantitative tools to enable us to explicitly link 81 configurational properties of habitats to social structures, and therefore generate testable 82 hypotheses on the role of the physical habitat environments on socially-mediated population 83 outcomes. 84 The features of animal habitats are typically multi-faceted-they can be described by the 85 heterogeneity, sizes, abundance and spatial arrangements of habitat components (Tokeshi 86 and Arakaki, 2012). For a given animal species, these features determine habitat structural 87 connectivity, indicating where individuals can move and, thereby, the behaviours that they

Habitat structural (or physical) connectivity and habitat networks
Structural connectivity captures the physical features of habitat components and their spatial arrangement, which defines where animals of a given species can move (Baguette and Van Dyck, 2007). Habitat networks fundamentally describe the features of the physical habitat environments that determine habitat structural connectivity for a species, rather than the actual movements of animals through their physical habitat environments. For example, the features and spatial distribution of the chambers in an ant colony (Perna and Theraulaz, 2017) determine the structural connectivity of the colony, which can then structure the movements-and other processes, such as recruitment-of ants (Vaes et al., 2020). The actual movements of individuals can also be the products of other non-habitat components, such as the instant social environment the individuals are exposed to. Thus, in contrast to functional connectivity and movement networks (see below), structural connectivity describes the habitat components and their spatial arrangements for a given species-the habitat networks-that determine the movement potential of individuals (Fig. 2).

Habitat functional connectivity and movement networks
Animal movement networks characterize the observed movements of animals, with nodes anchored in physical landscapes and links quantifying the movement flows of animals among them (Pasquaretta et al., 2020). The resulting functional connectivity then aims to describe the extent to which habitats facilitates or impedes movements (Taylor et al., 1993;Goodwin and Fahrig, 2002).
However, movement networks are the products of the spatial, social, and stochastic components that shape individual movements (Jacoby and Freeman, 2016), and habitat functional connectivity therefore reflects the interplay between individual behavioural decisions of animals and their physical habitat environments (Baguette and Van Dyck, 2007) as well as other underlying drivers of movement decisions, which can be multiple and complex (Nathan et al., 2008). For example, the movements of individuals from one patch to another could be affected by both the habitat features between these patches (i.e. permeability of landscape matrix), as well as the social (or biological) environment (Armansin et al., 2019) and an inherent drive to move in a specific direction (e.g. migration). Thus, functional connectivity inferred from observed movements may not completely estimate the underlying landscape features that facilitate or impede animal movements. The correlation between habitat structural connectivity and functional connectivity will further be influenced by the sampling effort-more samples (e.g. many individuals across many years) are Making generalizable predictions about the consequences of habitat configuration requires 149 being able to generate a range of alternative networks depicting a plausible diversity of habitat 150 configurations, while at the same time being able to maintain certain key aspects of 151 configurational features (e.g. Fig. 1-b, 1-c). One way to overcome the limitation of empirical 152 networks is to simulate networks using a generative model. Generative network models have 153 been instrumental to our understanding of the importance and consequences of network 154 structure for a range of phenomena (Granovetter, 1973;Watts and Strogatz, 1998) Here, we address the need for a generative model of animal habitat networks by extending 179 the random geometric network model. We first define animal habitat networks and outline 180 the key configurational features of animal habitats that can be captured by such networks. 181 Next, we describe a modelling framework (available in the R package AnimalHabitatNetwork) 182 for simulating animal habitat networks that is explicitly tailored to depict the diverse physical 183 configurations of animal habitats. We show that our network simulation algorithm can be 184 tuned to efficiently capture the structural connectivity of real habitats. Doing so is important, 185 as making predictions requires producing realistic scenarios as well as contrasting these 186 against alternative scenarios. Finally, we illustrate how our framework can be used to simulate concave (more likely to be linked) to convex (less likely to be linked). We define λ > 0 so P(Dij) 288 consistently increases over Dij (Fig. A1). This function enables us to generate a wide spectrum 289 of curves to cover the diverse and evolving relationships between Dij and P(Dij) by tuning λ 290 and µ (Fig. A1-A3)-which depict the propensity for physical features that inhibit individual 291 movements (e.g. barriers) to exist in the landscape (λ) and the species-level movement 292 characteristics that determine individuals' ability to exploit existing physical connectivity (µ).
293 Table 1 Parameters of the AHN model for depicting habitat physical configuration. for capturing the diverse landscape geometry), therefore it inherits spatial properties of the 317 landscape. Next, (ii) the algorithm removes the link between node i and j (i ≠ j) from the 318 network with probability P(Dij); in this example it results in a disconnected habitat network. 319 Then, (iii) the (disconnected) network components are rewired with minimal number of links. 320 Finally, (iv) the habitat network can be transformed to unweighted, if so desired. 321 In some cases, P(Dij) fragments the network into (disconnected) network components (e.g. Fig.   322 3-ii). The more µ approaches 0, the more links will be filtered out, and the more likely it is for and η = 1). 463 We initiated simulations by randomly allocating individuals to the nodes of each habitat 464 networks, and modelled individual movements and transmission dynamics for 500 timesteps. 465 When modelling individual movements, we defined the probability for an individual to stay 466 at the current node i at each timestep consistently as ps = 0.5, and the probability of moving 467 from node i to j as (1ps) × wij / Σjwij (where wij is the weight of the link between node i and j). 468 We simulated pathogen transmissions in the population using an SIR epidemic model 469 (Kermack and McKendrick, 1927;Keeling and Eames, 2005), where each individual is either