Table 1 - uploaded by Vincenzo Gervasi
Content may be subject to copyright.
Description of parameter symbols and values used in the individual-based model of reproductive site occupation and detection.
Source publication
Theory recognizes that a treatment of the detection process is required to avoid producing biased estimates of population rate of change. Still, one of three monitoring programmes on animal or plant populations is focused on simply counting individuals or other fixed visible structures, such as natal dens, nests, tree cavities. This type of monitor...
Contexts in source publication
Context 1
... that case, detection probability was fixed to one. The complete list of parameters, used to build the individual-based model, is provided in Table 1, whereas the logical structure of the model is shown in Fig. 1. ...
Context 2
... allowed us to assess the relative importance of each parameter in the model, by estimating the expected variation in model out- puts for a small change in each of the inputs ( McCarthy et al. 1995). For each iteration, we extracted parameter values from a uniform distribution, whose range is pro- vided in Table 1. After completing 1000 runs, we fitted a generalized linear model using all the standardized input parameters as predictors, to allow comparison among the effects of each predictor, and chose the length of the learning phase as response variable, using a Poisson distri- bution. ...
Similar publications
Informed management and conservation decisions for animal populations often require data at sufficient geographic, temporal, and demographic resolutions for precise and unbiased estimates of parameters including population size and demographic rates. Recently developed integrated population models estimate such parameters by unifying population pre...
Citations
... We also modeled individual variation linked with detection in the previous monitoring season Previous i ; a binary covariate which takes the value 1 if individual i was detected in the previous monitoring season and 0 otherwise. During NGS, investigators are believed to have the tendency to prioritize searching in locations where their searches were previously successful, which could positively influence the detection probability of those previously detected wolverine individuals during the focal monitoring season (Gervasi et al. 2014, Milleret et al. 2022. Availability of the monitoring data from the previous year made it possible to account for this potential source of heterogeneity in wolverine detectability. ...
After centuries of intense persecution, several large carnivore species in Europe and North America have experienced a rebound. Today's spatial configuration of large carnivore populations has likely arisen from the interplay between their ecological traits and current environmental conditions, but also from their history of persecution and protection. Yet, due to the challenge of studying population‐level phenomena, we are rarely able to disentangle and quantify the influence of past and present factors driving the distribution and density of these controversial species. Using spatial capture‐recapture models and a data set of 742 genetically identified wolverines Gulo gulo collected over ½ million km² across their entire range in Norway and Sweden, we identify landscape‐level factors explaining the current population density of wolverines in the Scandinavian Peninsula. Distance from the relict range along the Swedish–Norwegian border, where the wolverine population survived a long history of persecution, remains a key determinant of wolverine density today. However, regional differences in management and environmental conditions also played an important role in shaping spatial patterns in present‐day wolverine density. Specifically, we found evidence of slower recolonization in areas that had set lower wolverine population goals in terms of the desired number of annual reproductions. Management of transboundary large carnivore populations at biologically relevant scales may be inhibited by administrative fragmentation. Yet, as our study shows, population‐level monitoring is an achievable prerequisite for a comprehensive understanding of the distribution and density of large carnivores across an increasingly anthropogenic landscape.
... Using this approach, however, no analytical method is used to account for those natal dens that are not detected during the sampling and the uncertainty in registering successful reproduction events (i.e., imperfect and variable detection ;Gervasi et al. 2014, Kellner andSwihart 2014). Nonetheless, the authorities had to base their decisions on these proxies as representatives of the wolverine population, despite the uncertainties and potential sources of errors (Brøseth et al. 2010, Gervasi et al. 2014. By the early 2000s, noninvasive genetic monitoring of wolverines was introduced to improve the estimates of population trends, which is used together with the minimum population size based on the den counts , Brøseth et al. 2010, Gervasi et al. 2016, Bischof et al. 2020, Milleret et al. 2022b). ...
... Many unknown or unrecorded factors may cause additional heterogeneity in detection probability. For example, it is common that during sampling, surveyors do not search randomly, because of their prior knowledge or accessibility (Gervasi et al. 2014), or where a camera trapping or DNA search effort was conducted more intensively in a specific habitat type, inducing some forms of undocumented preferential sampling (Conn et al. 2017). Likewise, equipment failure and human errors can cause detectors to perform in clusters of varying effectiveness unbeknownst to the analyst. ...
1. Wildlife populations live in increasingly human-altered landscapes. Either because of their intrinsic values or due to instrumental values to humans, wildlife populations are monitored to inform about their current status and population trends and to forecast their future status in response to possible changes in their environment. Monitoring wildlife across spatial units and over time is a first step towards adaptive and evidence-based management.
2. This PhD dissertation consists of four articles that are centered on developing new methods, enhancing concepts, and showcasing applications of novel analytical approaches for quantifying landscape-level wildlife population density using noninvasive monitoring data. At the core of this PhD dissertation lie spatially explicit analytical models, namely spatial capture-recapture (SCR), with the ability to yield scale-transcending estimates of population parameters, while accounting for imperfect detection. This PhD dissertation is motivated by applied questions raised during noninvasive genetic monitoring of large carnivores in the Scandinavian Peninsula. However, the methodology and findings have broader implications.
3. The first two articles focus on understanding and mitigating the consequences of spatially variable and autocorrelated detection probability when analyzing wildlife monitoring data. Detection probability – the probability of detecting an individual from the target population, can vary across the study area, because of, for example, certain landscape characteristics or the specifics of the sampling design. Spatial autocorrelation in detection probability occurs when detectability is more similar among neighboring than distant sampling locations or devices. Both articles I and II use simulations to create and test many scenarios that may occur during real-life sampling of wildlife in monitoring studies. Article I evaluates the consequences of not accounting for spatial variation in detectability when analyzing monitoring data with SCR, with a specific focus on the impact on estimates of population size. This study shows that a misspecified SCR model performs reasonably well in many situations, from low to even intermediate levels of spatial variation in detectability. However, Article I identifies problematic cases of highly spatially variable and autocorrelated detection, which can lead to pronounced negative bias in population size estimates. Some of these extreme scenarios are expected in the large-scale monitoring of large carnivores in Scandinavia, which led to a follow-up study in the next chapter of this PhD.
4. Article II describes and tests three novel modeling approaches to account for spatially variable and autocorrelated detection probability in SCR with random effects. This study extends SCR with generalized linear mixed models (GLMMs) and compares the performance of the SCR-GLMMs that do and do not specifically account for spatial autocorrelation in detection probability. Article II then applies the new modeling approaches to Scandinavian brown bear Ursus arctos monitoring data from central Sweden, where the majority of the DNA data was collected opportunistically by volunteers and no reliable measure of sampling effort was available to infer spatially variable detectability. This empirical case study demonstrates the application of the proposed modeling approaches and suggests considerable spatial heterogeneity in the detection of bears, where detectability decreases in an east-to-west direction towards the Swedish-Norwegian border. Further, Article II discusses solutions to identify potential sampling gaps, where variation in the effort is not fully known and highlights computation trade-offs in using such novel SCR analyses in wildlife monitoring studies.
5. The next two articles demonstrate empirical applications of quantifying variation in wildlife population density and its determinants at the population level. Our current understanding of wildlife space use and habitat selection is dominated by geographically limited studies that often make inferences from a few instrumented individuals. Both articles III and IV use noninvasive genetic monitoring data of the wolverine Gulo gulo across the species’ entire range in the Scandinavian Peninsula and assess sex-specific responses of the wolverine density to a suit of historical and present-day environmental covariates. Both these articles predict the Scandinavian wolverine density distribution and provide estimates of population size. Article III, as a prelude to the next chapter, identifies the factors influencing the current density distribution of the wolverine, with a focus on the role of the relict range along the Swedish-Norwegian border, where the wolverine survived intense
human persecution by early 1970s. Article III reveals that distance from this transboundary alpine region is still one of the most important determinants of wolverine density and the highest female and male wolverine densities are expected closer to the relict range. However, current management conditions to limit wolverine expansion, especially in southern Norway, interact with distance from the relict range, and together with other topographic, climatic, and prey-related factors, have shaped the current density distribution of the wolverine in Scandinavia. This study is the first to look into the density determinants of the Scandinavian wolverine population across its entire geographic range.
6. Article IV builds on the findings from Article III and quantifies the dynamics of density determinants of the Scandinavian wolverine over a nine-year monitoring period. This study quantified the change in the impact of the environmental covariates over the past decade, as the wolverine has successfully expanded from the alpine relict range into the boreal forest. Article IV uses recently developed open-population SCR models that provide not only estimates of annual density and its determinants, but also estimates of the demographic parameters (i.e., recruitment and survival) needed to predict changes in the population dynamics. This study reveals that, on the one hand, whereas the role of the relict range is still important for determining the wolverine’s density distribution, its significance is diminishing over time. On the other hand, forest is appearing more and more as a significant predictor of today’s wolverine density. Article IV tracks temporal trends in the main determinants of male and female wolverine densities and it discusses the results in relation to the population recovery of the wolverine in Scandinavia in the presence of ongoing human pressure.
... In fact, if a species has a low detection probability, it becomes challenging to estimate whether its populations are stable or declining [10]. Additionally, detection probability can vary across study sites and over time, increasing the risk of biased inference [11][12][13]. As a result, repeated surveys are often required, should we want to obtain reliable trend estimates, particularly when detection probability is low [10], while a species with a high detection probability is less impacted by these issues. ...
... Cont.Animals 2022,12, 2085 ...
... Animals 2022, 12, 2085 ...
Most animal species are detected imperfectly and overlooking individuals can result in a biased inference of the abundance patterns and underlying processes. Several techniques can incorporate the imperfect detection process for a more accurate estimation of abundance, but most of them require repeated surveys, i.e., more sampling effort compared to single counts. In this study, we used the dependent double-observer approach to estimate the detection probability of the egg clutches of two brown frog species, Rana dalmatina and R. latastei. We then simulated the data of a declining population at different levels of detection probability in order to assess under which conditions the double counts provided better estimates of population trends compared to naïve egg counts, given the detectability of frog clutches. Both species showed a very high detection probability, with average values of 93% for Rana dalmatina and 97% for R. latastei. Simulations showed that not considering imperfect detection reduces the power of detecting population trends if detection probability is low. However, at high detection probability (>80%), ignoring the imperfect detection does not bias the estimates of population trends. This suggests that, for species laying large and easily identifiable egg clutches, a single count can provide useful estimates if surveys are correctly timed.
... The detectability of many species varies with environmental conditions; annual variation in weather can strongly affect the optimal time of year to survey, and the overall likelihood of detecting the target species (Field, Tyre & Possingham, 2005;Jackson et al., 2006;McConville et al., 2009;Rizzo et al., 2017;Shaffer, Roloff & Campa, 2019). If the factors influencing detectability are not known, one can never be confident that failure to detect constitutes a true absence, and the development of standardized monitoring schemes will be severely hampered (Penteriani et al., 2005;Bried & Pellet, 2012;Gervasi et al., 2014;Bellier, Kéry & Schaub, 2016;Crone, 2016). ...
... Owing to the high labor cost of continually running a drift-fence however, monitoring of adult populations is often sporadic or entirely absent. Population assessments based solely on dip-net data are common, yet fraught with uncertainty (Penteriani et al., 2005;Bried & Pellet, 2012;Gervasi et al., 2014;Bellier, Kéry & Schaub, 2016;Crone, 2016). Because larval development occurs only in years with suitable weather and hydrological conditions, it is possible for a breeding population of adults to occur but for the detection probability of larvae to be zero. ...
Local extinction and undetected presence are two very different biological phenomena, but they can be challenging to differentiate. Stochastic environments hamper the development of standardized monitoring schemes for wildlife, and make it more challenging to plan and evaluate the success of conservation efforts. To avoid reintroductions of species at risk that could jeopardize extant populations, managers attempting translocation events require a higher level of confidence that a failure to confirm presence represents a true absence. For many pond breeding amphibians, monitoring of the breeding population occurs indirectly through larval surveys. Larval development and successful recruitment only occurs after a sequence of appropriate environmental conditions, thus it is possible for a breeding population of adults to exist at a site but for detectability of the species to be functionally zero. We investigate how annual variability in detection influences long-term monitoring efforts of Reticulated Flatwoods Salamanders ( Ambystoma bishopi ) breeding in 29 wetlands in Florida. Using 8 years of historic dip net data, we simulate plausible monitoring scenarios that incorporate environmental stochasticity into estimates of detection probability. We found that annual variation in environmental conditions precluded a high degree of certainty in predicting site status for low-intensity monitoring schemes. Uncertainty was partly alleviated by increasing survey effort, but even at the highest level of sampling intensity assessed, multiple years of monitoring are required to confidently determine presence/absence at a site. Combined with assessments of habitat quality and landscape connectivity, our results can be used to identify sites suitable for reintroduction efforts. Our methodologies can be generally applied to increase the effectiveness of surveys for diverse organisms for which annual variability in detectability is known.
... To estimate the population size of threatened species, many biologists use direct counts (O'Shea et al. 2003). However, direct counts are biased due to imperfect detection (Thomas et al. 1989, Kunz 2003, meaning this method provides only indices of a population size, potentially leading to erroneous conclusions regarding population trends when detection probability is not constant over time (Kunz et al. 2009, Archaux et al. 2012, Gervasi et al. 2014). The issue of imperfect detection has been the focus of methodological developments over the last 50 yr (Buckland et al. 2001, Williams et al. 2002. ...
Establishing effective wildlife conservation measures requires accurate demographic information such as population size and survival probability: parameters that can be extremely difficult to obtain. This is especially the case for threatened species, which are often rare and sometimes occupy inaccessible areas. While noninvasive genetic sampling (NIGS) techniques are promising tools for providing demographic data, these methods may be unreliable in certain situations. For instance, fecal samples of frugivo-rous species in tropical areas degrade rapidly, affecting the usability of the genetic material. In this study, we compared (1) NIGS capture-mark-recapture (NIGS-CMR) with conventional CMR to determine their potential in estimating demographic parameters of fruit bats, and (2) the precision of these demographic parameters and the associated costs given varying sampling designs through simulations. Using Living-stone's fruit bats (Pteropus livingstonii) fecal samples, microsatellite markers were tested and genotyping success and error rates were assessed. The average genotyping success rate was 77%, and the total geno-typing error rate for all loci was low (allelic dropout rate = 0.089, false alleles rate = 0.018). Our results suggested that five loci were required to identify individuals. Simulations showed that monitoring the species over a 9-yr period with a recapture rate of 0.20 or over a 6-yr period with a recapture rate of 0.30 seems appropriate to obtain valuable demographic parameters. Overall, in comparison to conventional CMR, NIGS-CMR offers a better method for estimating demographic parameters and subsequently for conducting long-term population monitoring in flying foxes due to the fact that (1) sample collection is easy and the level of genotyping errors in the laboratory is low and (2) it is cheaper, less time-consuming, and less disturbing to individual animals. We strongly advocate an approach that couples a pilot study with simulations as done in this study in order to choose the most efficient monitoring method for a given species or context.
... v www.esajournals.org 6 July 2021 v Volume 12(7) v Article e03571 influence the probability of being detected at subsequent occasions (Gervasi et al. 2014). To account for this potential traphappiness (Williams et al. 2002), we used a binary indicator of whether an individual was detected or not during the previous monitoring season (TrapResponse it ). ...
... Under this specification, a spatial intensity function describes movement as a series of isotropic Gaussian random walks (see also Gardner et al. 2018) weighted by the spatial covariate considered (Zhang et al. 2020). We used the average number of known wolverine dens per habitat cell coming from independent den surveys (Gervasi et al. 2014, Bischof et al. 2020a) as a covariate of wolverine density and estimated β dens , the effect of the number of dens in a given habitat cell on the probability that an individual has its AC located in this same cell. ...
Integrating dead recoveries into capture–recapture models can improve inference on demographic parameters. But dead‐recovery data do not only inform on individual fates; they also contain information about individual locations. Open‐population spatial capture–recapture (OPSCR) has the potential to fully exploit such data. Here, we present an open‐population spatial capture–recapture–recovery model integrating the spatial information associated with dead recoveries. Using simulations, we investigate the conditions under which this extension of the OPSCR model improves inference and illustrate the approach with the analysis of a wolverine (Gulo gulo) dataset from Norway. Simulation results showed that the integration of dead recoveries into OPSCR boosted the precision of all demographic parameters. In addition, the integration of dead‐recovery locations boosted the precision of the inter‐annual movement parameter, which is difficult to estimate in OPSCR, by up to 40% in case of sparse data. We also detected a 139–367% increase in the probability of models reaching convergence with increasing proportion of dead recoveries when dead‐recovery information was integrated spatially, compared with a 30–107% increase when integrating dead recoveries in a non‐spatial way. The analysis of the wolverine data showed the same general pattern of improved parameter precision. Overall, our results highlight how leveraging the demographic and spatial information contained in dead‐recovery data in a spatial capture–recapture framework can improve population parameter estimation.
... Monitoring plans that collect relative abundance indices, unadjusted for the detection process, have been criticized in the literature (Anderson 2001, Williams et al. 2001, Gervasi et al. 2014; however, for a variety of historical and logistic reasons they remain common for many species and in many jurisdictions (Rabe et al. 2002, Keegan et al. 2011). Although collecting annual information on the detection process (e.g., via distance sampling or double-observer protocols) should be encouraged wherever feasible (Williams et al. 2001), in their absence state-space modeling techniques can help account for the imperfect detection process when time series length is adequate (De Valpine andHastings 2002, Buckland et al. 2004). ...
Monitoring annual change and long‐term trends in population structure and abundance of white‐tailed deer (Odocoileus virginianus) is an important but challenging component of their management. Many monitoring programs consist of count‐based indices of relative abundance along with a variety of population structure information. Analyzed separately these data can be difficult to interpret because of observation error in the data collection process, missing data, and the lack of an explicit biological model to connect the data streams while accounting for their relative imprecision. We used a Bayesian age‐structured integrated population model to integrate data from a fall spotlight survey that produced a count‐based index of relative abundance and a volunteer staff and citizen classification survey that generated a fall recruitment index. Both surveys took place from 2003–2018 in the parkland ecoregion of southeast Saskatchewan, Canada. Our approach modeled demographic processes for age‐specific (0.5‐, 1.5‐, ≥2.5‐year‐old classes) populations and was fit to count and recruitment data via models that allowed for error in the respective observation processes. The Bayesian framework accommodated missing data and allowed aggregation of transects to act as samples from the larger management unit population. The approach provides managers with continuous time series of estimated relative abundance, recruitment rates, and apparent survival rates with full propagation of uncertainty and sharing of information among transects. We used this model to demonstrate winter severity effects on recruitment rates via an interaction between winter snow depth and minimum temperatures. In years with colder than average temperatures and above average snow depth, recruitment was depressed, whereas the negative effect of snow depth reversed in years with above average temperatures. This and other covariate information can be incorporated into the model to test relationships and provide predictions of future population change prior to setting of hunting seasons. Likewise, post hoc analysis of model output allows other hypothesis tests, such as determining the statistical support for whether population status has crossed a management trigger threshold. © 2020 The Wildlife Society. We developed a Bayesian integrated population model for white‐tailed deer populations using 2 streams of low‐cost monitoring data: a road‐based spotlight count survey and a volunteer‐collected recruitment index. The model allows for a coherent picture of populations status, accounting for several sources of uncertainty and allowed us to estimate winter severity effects on deer demography.
... Using simulation studies can reveal important assumptions about monitoring programs [5]. Our individual-based model allowed us to better incorporate stochasticity into the underlying process governing population change based on individual survival and reproduction, especially when paired with an observation process [35]. However, our model was not spatially explicit, and not a true representation of the complexity of the system. ...
Researchers and managers are often interested in monitoring the underlying state of a population (e.g., abundance), yet error in the observation process might mask underlying changes due to imperfect detection and availability for sampling. Additional heterogeneity can be introduced into a monitoring program when male-based surveys are used as an index for the total population. Often, male-based surveys are used for avian species, as males are conspicuous and more easily monitored than females. To determine if male-based lek surveys capture changes or trends in population abundance based on female survival and reproduction, we developed a virtual ecologist approach using the lesser prairie-chicken (Tympanuchus pallidicinctus) as an example. Our approach used an individual-based model to simulate lek counts based on female vital rate data, included models where detection and lek attendance probabilities were <1, and was analyzed using both unadjusted counts and an N-mixture model to compare estimates of population abundance and growth rates. Using lek counts to estimate population growth rates without accounting for detection probability or density-based lek attendance consistently biased population growth rates and abundance estimates. Our results therefore suggest that lek-based surveys used without accounting for lek attendance and detection probability may miss important trends in population changes. Rather than population-level inference, lek-based surveys not accounting for lek attendance and detection probability may instead be better for inferring broad-scale range shifts of lesser prairie-chicken populations in a presence/absence framework.
... In addition, great methodological advances in the field of capture-mark-recapture (CMR) studies have put much focus on the benefits and potential for insight from marked individuals [13] even if they are not radio collared. While radio collaring has been used mainly for research purposes, the opportunities offered by CMR methods has been gradually implemented in monitoring programs [14,15]. In particular, many large carnivore monitoring programs are currently implementing CMR analysis based on non-invasive sampling (e.g. of scats; see [16]). ...
... Group size is then determined from the photographs. However, as it is well known that population estimates which are not based on a sampling scheme that allow observation probability to be estimated separately are often biased low [14], we considered models that both include and exclude this data set. ...
We developed a model for estimating demographic rates and population abundance based on multiple data sets revealing information about population age- and sex structure. Such models have previously been described in the literature as change-in-ratio models, but we extend the applicability of the models by i) using time series data allowing the full temporal dynamics to be modelled, by ii) casting the model in an explicit hierarchical modelling framework, and by iii) estimating parameters based on Bayesian inference. Based on sensitivity analyses we conclude that the approach developed here is able to obtain estimates of demographic rate with high precision whenever unbiased data of population structure are available. Our simulations revealed that this was true also when data on population abundance are not available or not included in the modelling framework. Nevertheless, when data on population structure are biased due to different observability of different age- and sex categories this will affect estimates of all demographic rates. Estimates of population size is particularly sensitive to such biases, whereas demographic rates can be relatively precisely estimated even with biased observation data as long as the bias is not severe. We then use the models to estimate demographic rates and population abundance for two Norwegian reindeer (Rangifer tarandus) populations where age-sex data were available for all harvested animals, and where population structure surveys were carried out in early summer (after calving) and late fall (after hunting season), and population size is counted in winter. We found that demographic rates were similar regardless whether we include population count data in the modelling, but that the estimated population size is affected by this decision. This suggest that monitoring programs that focus on population age- and sex structure will benefit from collecting additional data that allow estimation of observability for different age- and sex classes. In addition, our sensitivity analysis suggests that focusing monitoring towards changes in demographic rates might be more feasible than monitoring abundance in many situations where data on population age- and sex structure can be collected.
... Moreover, for species that have large movement capacities, some individuals can move between counting sites and can thus be counted twice if counts are not conducted simultaneously De Barba et al., 2010;Garel et al., 2010;Hedges et al., 2013;Marchandeau et al., 2006). Direct count method provides only indices of the population size that might however lead to erroneous conclusion regarding population trends when detection probability is not constant over time (Archaux et al., 2012;Gervasi et al., 2014). ...
The Comoros Islands are known for their important biodiversity with a high endemism rate for each taxonomic group. This natural richness face huge anthropogenic pressures due to a high rate of habitat loss and fragmentation estimated to be the highest in the word. Wild endemic mammals are the most threatened fauna in these islands. These species are often characterized by small population sizes making them highly vulnerable to disturbances. Indeed, small population size makes populations prone to allee effect, genetic drift or inbreeding depression, which subsequently conducts to a decrease of species’ evolutionary potential, thus diminishing their long term viability. In order to understand the effect of habitat disturbance on the Comoros Islands natural fauna, I studied two endemic and highly threatened flying fox species (Pteropus livingstonii with a population size estimated of 1300 individuals and P.seychellensis comorensis whose population is estimated to few thousands of individuals). For that, I combined different approaches including spatial distribution and ecological niche modeling, as well as population demography and socio-economic approaches. This integrated approach is crucial to identify the different causes of mammals’ population loss and propose relevant conservation measures. In a first part of this thesis, I show the results of the spatial distribution modeling and habitat selection of the two flying fox species as well as their geographic distribution ranges using Species Distribution Modelling (SDM) and Ensemble of Small Models approach specifically adapted to rare and threatened species. This first part allowed me to assess which ecological variables and anthropogenic pressures are determinant for the distribution of both species as well to characterize the degree of threat of the two species. In the second part of this thesis, I studied the genetic diversity and population structure of both species among the four islands of Comoros with the aim to look for possible gene flow breaks between sub-populations but also to uncover which species face a high risk of extinction. This study highlights that these two phylogenetic and morphologically related species show different genetic structures among islands. In a third part, I explored the feasibility and costs of a non-invasive genetic monitoring protocol to obtain accurate population size, demographic parameters and develop a long-term monitoring of P. livingstonii. Due to the sensitivity of this species to capture and handling but also because of its rareness, a direct monitoring using classical capture-recapture method was not possible. This study showed that this approach is realistic but involves a high cost that seems to be unsuitable with the budgets available for conservation of the species in the Comoros Islands. In a fourth part of my thesis, I characterized the anthropogenic pressures that impacts both species using a socio-economic characterization of these islands (forest exploitation and hunting pressures among others) by using semi-structured interviews and a Q-methodology approach. This allows me to understand the relationship of local communities with the local biodiversity as well as to interpret the ongoing natural habitat evolution and to predict its future. In the last chapter (fifth), I combined the results of all the different but complementary approaches used along the thesis with the aim to propose a management plan appropriate for these two species.
