Model species: the spur-thighed tortoise in SE Spain
The spur-thighed tortoise T. graeca is a medium-sized long-lived tortoise with slow population dynamics, steady population growth rates (Rodríguez-Caro et al. 2017, 2019), delayed maturation (9–12 years; Rodríguez-Caro et al. 2013), and a clutch size of 1–7 eggs. It is endangered (Vulnerable; IUCN 1996), with main threats in Spain being habitat loss and fragmentation caused by anthropogenic activities (e.g. irrigated lands, greenhouses and tourist urban development; Anadón et al. 2007; Graciá et al. 2020b). In Spain T. graeca inhabits heterogeneous cultural landscapes in the coastal mountains of the most arid region of SW Europe, formed by a shrubland matrix that includes other uses, mainly irrigated and non-irrigated crops (Anadón et al. 2006b). Since the 1950s, this area has undergone intense rural depopulation (De Aranzabal et al. 2008), and since the 1980s the region has been subject to agricultural intensification, the development of urban and touristic areas and infrastructure development (Martínez-Fernández et al. 2015). During the last 60 years these processes have created a polarised landscape where natural and anthropisated areas co-occur, often in close proximity, in an area traditionally inhabited by this tortoise.
We used the spatially explicit individual-based model STEPLAND (Fig. 1), which synthesises data of intense field studies from almost two decades to simulate the species demography, habitat selection and movement of T. graeca in real landscapes. We simulated the population dynamics of T. graeca from the 1950s to the present-day based on time-series data of real landscapes that represent historical land-use changes. We also projected the population trend into the future and recorded changes in several emerging demographic characteristics, including the reproductive rate (RR; i.e., mean number of offspring of reproductive females), population density (number of individuals per hectare), and extinction probabilities after 200 years (Pext). To check the legacy effects of historical land-use changes on population dynamics, we also quantified the extinction debt.
Habitat preference and changed movement patterns impact population dynamics in STEPLAND through a mate-finding Allee effect (Jiménez-Franco et al. 2020a) in two main ways. According to Anadón et al. (2012), tortoises, especially females, respond to a high proportion of intensified areas by moving longer distances and by shifting from a narrow home range to a more unbounded movement pattern, which may lead to increased mating opportunities. Secondly, tortoises very much avoid areas with intensified use (i.e., main roads, dense human infrastructures, intensive agriculture or pine forests) (Anadón et al. 2012), which act as barriers that decrease mating opportunities. However, our extensive empirical work found no evidence for variation in survival (based on age-structure analyses) and clutching between natural and altered landscapes (own data), probably because tortoises never used areas of intensified use (see Table C3 in Anadón et al. 2012). For this reason, and as our goal was to assess the potential impact of land-use changes on population dynamics based on the observed movement patterns, we used the same survival rates and reproduction parameters (i.e., number of clutches, number of eggs per clutch) for the tortoises inhabiting natural and altered landscapes, but different movement parameterizations. Decreasing survival and reproduction will lead to a lower population growth rate, which can be assessed using Leslie matrix models (Rodríguez-Caro et al. 2019; Jiménez-Franco et al. 2020a).
STEPLAND: a spatially explicit individual-based model to simulate spur-thighed populations’ viability
The STEPLAND model (Fig. 1) was developed to integrate demographic processes, i.e., reproduction, mortality and ageing (Graciá et al. 2020a; Jiménez-Franco et al. 2020a), with a previously developed individual-based model of tortoise movement (Anadón et al. 2012). For model description, we followed the Overview, Design concepts and Details protocol (ODD) proposed by Grimm et al. (2010). We summarised the model in the paragraphs below, while a full ODD version is provided in Appendix S1. The model was implemented in Python 2.7, and its code, parameterisation, and the main result files are available in the Figshare repository (Jiménez-Franco et al. 2021).
STEPLAND contains two types of entities: landscape and tortoises. Landscapes are composed of a grid of 10 m × 10 m cells. Each grid cell is characterised by its position (x and y coordinates), and also by its assignment to one of the five habitat quality parameters in STEPLAND. We reclassified the original land-use map that comprised eight land-use cover categories to five habitat types that influence movement (see more details of landscape reclassification in Appendix S2). Tortoises have the attributes sex, age and their location over time, and the model is composed of specific submodels for movement, reproduction and mortality (Fig. 1). The overall time step of the model is 1 year, annual mortality occurs at the end of the year before the age of the surviving tortoises is updated, and newborn tortoise are added at the beginning of the year. However, the movement submodel operates at a finer intraday scale with sex-specific movement probabilities depending on the month.
The movement submodel reflects the empirical finding that the activity and movement patterns of T. graeca individuals strongly differed between “natural” and “altered” landscapes (for a definition of landscape types see the following subsection) as found by Anadón et al. (2012). Movement tracks of tortoise consisted in the model of a sequence of steps to neighbour 10 m × 10 m cells, with probabilities to enter one of the eight neighbouring cells, or to stay in the current cell, depending on three sets of weights that described autocorrelation in movement, home behaviour, and habitat dependence. Each day was divided into four activity intervals, and a probability PMOV decided if a tortoise moved during that interval, and a probability distribution DMOV decided how many steps the individual moved. PMOV and DMOV were determined from the radio-tracking data (see Fig. S4 in Appendix S1), and PMOV followed a seasonal pattern driven by hibernation and aestivation periods, and differed between males and females. To reflect the individual variability in the movement patterns, Anadón et al. (2012) determined in a simulation study 5627 individual movement profiles, which included four types of profiles (female-natural, female-altered, male-natural, and male-altered). They consisted of different combinations of the nine movement parameters that produced trajectories that were in agreement with that observed by radio-tracking data. Following previous studies (Graciá et al. 2020a; Jiménez-Franco et al. 2020a), we randomly assigned one individual movement profile to a tortoise at birth and for all other tortoises we assigned a new movement profile at the end of each year (corresponding to its sex and landscape type). Further details are provided in the ODD protocol in Appendix S1 (Movement section).
The implementation of the demographic processes of reproduction and mortality reflect the species biology with most demographic parameters determined directly from our data (see Tables S1 and S3 in Appendix S3). Reproduction consists of the subprocesses mating, sperm storage and clutching. Mating probabilities in T. graeca emerge from the interaction of movement with landscape structure and are strongly conditioned by landscape connectivity (Fig. 1). STEPLAND assumed that mature females will mate: (i) if during the peak mating season (end of April) at least one reproductive male is located within an encounter distance of 500 m (as determined by Graciá et al. 2020a); and (ii) if females and males were not separated by barriers such as main roads, dense human infrastructures or intensive agriculture (see landscape barriers in Fig. S2, Appendix S2). Sperm storage of T. graeca females can allow them to reproduce for the next 3 years after mating (Cutuli et al. 2013; Jiménez-Franco et al. 2020a). Adult females harboring viable sperm have the opportunity to lay between one to three clutches per year (with 1–6 eggs per clutching), as observed in real populations (Rodríguez-Caro et al. 2021). The eggs were placed at the location of the female in spring and early summer when clutching occurs (from the end of April to the end of July), and the newborns emerge from this location.
Annual survival rates are applied at the end of the year and vary among age classes, including newborns (representing hatching success and survivals of individuals under 1 year), immature individuals (aged 1–3), subadults (aged 4–6) and adults (aged ≥ 7). We did not include a carrying capacity or negative density dependence in demographic rates because the population sizes occurring during model simulations ranged between 0 and 0.4 ind/ha, which correspond to “very low” and “low” densities, considering that the highest density observed in the study area was 13 ind/ha (Anadón et al. 2009).
Landscape scenarios: historical habitat transformations in the Mediterranean ecosystem
The two selected study areas are located in the Almenara mountain range at the core of the species’ range distribution area of T. graeca in SE Spain (Anadón et al. 2006a, 2007) (Fig. 2a) and faithfully represent the history of land-use changes of semiarid Mediterranean landscapes in the past century and, to a certain extent, the history of many rural landscapes in developed countries. The two modelling areas covered 5 × 5 km (Fig. 2b, c). The historical dynamics of land uses were characterised by mapping land-use covers in 1956, 1987 and 2010. One of the study areas, Los Mayorales, has a representative history of the agricultural abandonment with secondary succession of increased Mediterranean scrub. This landscape is composed of a mosaic of natural areas (Mediterranean scrub and pine forest) and traditional agricultural areas (dry-land crops such as cereals, almonds and olives). The other study area, Los Estrechos, represents the anthropisation process, that is, an increase in agricultural intensification, urban areas and infrastructure development with highways (which are no permeable for T. graeca) and roads (which have low permeability).
We built six landscape maps (2 study areas × 3 periods) with eight land-use cover categories (Fig. 2b, c), which were transformed to five habitat types that influence movement (see more details of the mapping process in Appendix S2). We selected these habitat types because they influence T. graeca movement patterns (Anadón et al. 2012; see the ODD for details). Each habitat type is associated with one parameter (i.e., H1W, H2W, …, H5W) that gives the weight in the habitat-biased random walk. The habitat types include (i) with null habitat quality (H5W): urban areas and highways (which are never used); (ii) with very low habitat quality (H1W): intensive agriculture (irrigated crops, citrics and greenhouses), pine forest areas (Mediterranean pine forest of Pinus halepensis) and roads (paved roads except highways); (iii) with good habitat quality (H4W, H3W and H2W, respectively): natural areas on slope (Mediterranean scrub on slope), flat natural areas (flat Mediterranean scrub) and traditional agricultural lands (see more details in Appendix S2).
As the movement parameterizations differ between natural and altered landscapes, we need to assign each of the six landscape maps either to natural or to altered, based on their habitat composition (Anadón et al. 2012). The 2010 landscape in the study area Los Estrechos was classified as altered, whereas all other five landscapes were classified as natural. A landscape with more than 23.8% of cover types (i) and (ii) defined above was defined as altered and natural otherwise.
Design of simulation experiments
We used initial population sizes of 250 individuals in a 5 × 5 km area that represented low T. graeca population densities in natural Spanish populations (0.1 tortoises/ha; Anadón et al. 2009). We selected this density because previous work showed that it is related to an extinction threshold (by assuming sperm storage durability of 3 years): if the initial density is below, spatial effects lead to a mate-finding Allee effect that can cause, especially in altered landscapes, growth rates below 1, but otherwise growth rates exceed 1 (Jiménez-Franco et al. 2020a). Thus, for higher initial densities we would most likely not observe responses to altered movement patterns.
As in previous parameterisations (Graciá et al. 2020a; Jiménez-Franco et al. 2020a), the initial age structure was the same for all the simulations. It was based on a stable age distribution predicted by an age-stage structured deterministic matrix population model using the same parameterisation of survival and reproduction parameters as in the simulation model. For this purpose, we employed the popbio package (Stubben and Milligan 2007) in the statistical R programme, version 3.5.1 (R Core Team 2021; R scripts are shown in Appendix S4). To avoid individuals’ unrealistic initial distribution (e.g., too many individuals placed in low suitability areas), we allowed the initial population to adjust to habitat types. We accomplished this by initially distributing tortoises randomly across all areas of good habitat quality (i.e., habitat types H2, H3, and H4). We then simulated the movement of these individuals (but without demography; i.e., reproduction, mortality or ageing) over 50 years to let them adjust to the initial landscape map. We then took these individual locations as the starting position in the following simulations for the two selected areas with land-use scenarios “without”/“with” change since 1956 (Fig. 2).
We assessed the effect of both, the agricultural abandonment process and the anthropisation process on the reproductive rate (RR), population density (ind/ha) and extinction probability (Pext) of T. graeca over 200 years. To this end, we conducted “impact” scenarios where we simulated the population dynamics in the two study areas: Los Mayorales (representing agricultural abandonment) and Los Estrechos (representing anthropisation). We also conducted “control” simulations by assuming that the landscape had not changed since 1956 (Fig. 2). To account for the stochastic processes, especially for estimating extinction probabilities, we repeated each scenario 256 times. Therefore, the total number of independent model simulations was 1024 (2 study areas × 2 land-use scenarios × 256 replicates).
The simulations in each landscape scenario were analysed to calculate three demographic variables, which were averaged over the 256 replicate trajectories every 10 years. First, the reproductive rate RR at time step t was the mean number of offspring of reproductive females, calculated by dividing the number of individuals born in time step t by the number of reproductive females (those aged 10 years or older) that were present in time step t. Only the non-extinct trajectories in each time step were used to calculate RR. As a second demographic variable we used the population density at time step t. Finally, we calculated the extinction rate Pext for each year t as the proportion of extinct replicates (i.e., having a population size Nt ≤ 1).
To quantify the total legacy effects of the historical land-use changes at the end of the simulation period, we calculated the differences in population densities and Pext-values between the corresponding control vs. impact scenarios. The extinction debt for a given scenario was calculated as the difference between the population viability after the habitat change and the end of the simulation period (Kuussaari et al. 2009). The time-lag for RR and population density was quantified by considering the initial period of the land-use change (in 1987) until the period when the confidence intervals of the control/impact scenarios did not overlap. The time-lag for population viability was quantified by considering a difference in the control/impact scenarios values equal or higher than 0.05.