The study of population dynamics has long depended on methodological progress. Among many striking examples, continuous time models for populations structured in age (Sharpe & Lotka, 1911) were made possible by progress in the mathematics of integral equations. Therefore the relationship between population ecology and mathematical and statistical modelling in the broad sense raises a challenge in interdisciplinary research. After the impetus given in particular by Seber (1982), the regular biennial EURING conferences became a major vehicle to achieve this goal. It is thus not surprising that EURING 2003 included a session entitled “Methodological advances”. Even if at risk of heterogeneity in the topics covered and of overlap with other sessions, such a session was a logical way of ensuring that recent and exciting new developments were made available for discussion, further development by biometricians and use by population biologists. The topics covered included several to which full sessions were devoted at EURING 2000 (Anderson, 2001) such as: individual covariates, Bayesian methods, and multi–state models. Some other topics (heterogeneity models, exploited populations and integrated modelling) had been addressed by contributed talks or posters. Their presence among “methodological advances”, as well as in other sessions of EURING 2003, was intended as a response to their rapid development and potential relevance to biological questions. We briefly review all talks here, including those not published in the proceedings. In the plenary talk, Pradel et al. (in prep.) developed GOF tests for multi–state models. Until recently, the only goodness–of–fit procedures for multistate models were ad hoc, and non optimal, involving use of standard tests for single state models (Lebreton & Pradel, 2002). Pradel et al. (2003) proposed a general approach based in particular on mixtures of multinomial distributions. Pradel et al. (in prep.) showed how to decompose tests into interpretable components as proposed by Pollock et al. (1985) for the Cormack–Jolly–Seber model. Pledger et al. (in prep.) went on in their thorough exploration of models with heterogeneity of capture (Pledger & Schwarz, 2002; Pledger et al., 2003), by considering the use of finite mixture models for the robust design. Given the level of details in demographic traits presently addressed by capture–recapture, the problem of heterogeneity, once apparently settled by fairly reassuring messages (Carothers, 1973, 1979), is becoming again a central issue, with potential disastrous consequences if improperly handled. Heterogeneity models, that bear also a relationship to “multi–event models” (Pradel, in press), will thus certainly be increasingly useful. Pollock, Norris, and Pledger (in prep.) reviewed the capture–recapture models as applied to community data (Boulinier et al., 1998) and developed general removal and capture–recapture models when multiple species are sampled to estimate community parameters. Because of unequal delectability between species, these approaches bear a clear relationship to heterogeneity models, which will be more and more a reference for comparative studies of communities and “macroecology” (Gaston & Blackburn, 2000). Bonner & Schwarz (2004) proposed a capture–recapture model with continuous individual covariates changing over time more fully developed in Bonner & Schwarz (2004). The difficulty here is to set up a sub–model predicting the covariate value when an individual is not captured. While multi–state models permit an ad hoc treatment by categorizing the covariate, Bonner and Schwarz bring a sound answer by considering the covariate obeys a Markov chain with continuous state–space. Otis & White (2004) presented a thorough, simulation–based, investigation of two approaches used to test the contrasting hypotheses of additive and compensatory hunting mortality based on band recovery data. The two approaches are the usual ultra–structural model and a new one based on a random effects model. Thispaper can be viewed as part of a revival of studies of the dynamics of exploited populations, in the broad sense, including the study of man–induced mortality in the framework of conservation biology (Lebreton, in press). This revival is a direct consequence of the increasing impact of man on the biosphere and of continuing methodological progress (Ferson & Burgman, 2000). The use of random effects models (see also Schaub & Lebreton, 2004) directly builds upon the seminal work by Anderson and Burnham (1976). Stauffer presented a Winbugs implementation of the Cormack–Jolly–Seber model that complemented other presentations in the conference and the short course. Finally, Morgan, Besbeas, Thomas, Buckland, Harwood,Duck and Pomery, proposed a thorough and timely review of integrated modelling, i.e., in our context, of models considering simultaneously capture–recapture demographic information and census information. These methods were covered in other sessions, in relation to Bayesian methodology. Integrated modelling appears indeed to be the logical way of combining all pieces of information arising from integrated monitoring, and as one of the great methodological challenges for our community in the years to come (Besbeas et al., 2002).

In the Southern Hemisphere, humpback whales (Megaptera novaeangliae) migrate along the extended continental coastlines of Australia, South America, and South Africa. This study reports on photo-identification capture–recapture data from a long-term survey conducted in Hervey Bay, Queensland, where a substantial proportion of the population stop over early in the southern migration. Photo-identification data were collected over 10 weeks per year from 1997 to 2009. The migration through Hervey Bay is dominated and led by females with high fidelity to the site. Mature females, yearlings, and immature whales use the Bay during August, while mature lactating females with calves dominate during September and October. Complex social behaviours occur throughout the season and differ between the early and late cohorts. We argue that the composition of the two cohorts and their distinctively different behaviours indicate that Hervey Bay is not simply a resting site but an area of aggregation that serves important social and biological benefits. A multistate open robust design model was fitted to capture–recapture data to estimate the annual number of whales visiting the Bay, the permanent emigration rate, proportions of the visiting population that do not enter the Bay each year, the number present during each week, and their residency times. The number of annual visitors to the Bay increased approximately linearly from 857 in 1997 to 2175 at the end of sampling in 2009 with two-thirds migrating through during the first half of each season. The population rate of growth may have been slowing by 2009, but there was considerable uncertainty in the trajectory and little basis for projection into the future. While it is desirable to know the current status of the Hervey Bay population and what has occurred since 2009, the cost and effort required make further manual collection and matching of images unlikely. The development of AI algorithmic matching software may enable further research in future.

Effective management of the threatened Antillean manatee (Trichechus manatus manatus) in Puerto Rico requires reliable estimates of population size. Estimates are needed to assess population responses to management actions, and whether recovery objectives have been met. Aerial surveys have been conducted since 1976, but none adjusted for imperfect detection. We summarize surveys since 1976, report on current distribution, and provide population estimates after accounting for apparent detection probability for surveys between June 2010 and March 2014. Estimates in areas of high concentration (hotspots) averaged 317 101, three times higher than unadjusted counts (104 +/- 0.56). Adjusted estimates in three areas outside hotspots also differed markedly from counts (75 +/- 9.89 versus 19.5 +/- 3.5). Average minimum island-wide estimate was 386 +/- 89, similar to the maximum estimate of 360 suggested in 2005, but fewer than the 700 recently suggested by the Puerto Rico Manatee Conservation Center. Manatees were more widespread than previously understood. Improving estimates, locally or island-wide, will require stratifying the island differently and greater knowledge about factors affecting detection probability. Sharing our protocol with partners in nearby islands (e.g., Cuba, Jamaica, Hispaniola), whose populations share genetic make-up, would contribute to enhanced regional conservation through better population estimates and tracking range expansion. El manejo efectivo del manati antillano amenazado en Puerto Rico requiere estimados de tamanos de poblaciones confiables. Dichas estimaciones poblacionales son necesarias para evaluar las respuestas a las acciones de manejo, y para determinar si los objetivos de recuperacion han sido alcanzados. Se han realizado censos aereos desde 1976, pero ninguno de ellos han sido ajustados para detecciones imperfectas. Aqui resumimos los censos desde 1976, actualizamos la distribucion, y reportamos los primeros estimados poblacionales ajustados para la probabilidad de deteccion aparente en los censos de Junio 2010 a Marzo 2014. Las estimaciones poblacionales en areas de mayor concentracion del manati promedio 317 +/- 103, tres veces mas abundante que los conteos sin ajuste (104 +/- 0.56). Las estimaciones poblacionales en tres areas fuera de las areas de mayor concentracion del manati tambien fueron marcadamente diferentes (75 +/- 9.89 vs 19.5 +/- 3.5). El estimado minimo poblacional en la isla entera fue de 386 +/- 89, similar al estimado maximo de 360 sugerido en el ano 2005, pero menor a los 700 sugeridos recientemente por el Centro de Conservacion de Manaties de Puerto Rico. Documentamos que el manati tiene una distribucion mas amplia de lo que se sabia con anterioridad. El mejoramiento de los estimados poblacionales locales o a nivel de isla requerira que se estratifique a la isla en forma diferente y que se investiguen los factores que influencian a la probabilidad de deteccion. Compartir protocolos como este con colaboradores de islas vecinas (por. ej., Cuba, Jamaica, Espanola), cuyas poblaciones de manaties comparten material genetico, contribuiria a la conservacion regional mediante mejores estimaciones poblacionales y monitoreo de la expansion de su ambito domestico.

Accurate estimates of population abundance are essential to both theoretical and applied ecology. Rarely are all individuals detected during a survey and abundance models often incorporate some form of imperfect detection. Detection probability, however, consists of three components: probability of presence during a survey, probability of availability given presence, and probability of detection given availability and presence. We develop an integrated model to separate these three detection components and provide abundance estimates for the available, present, and superpopulation of individuals. Our framework integrates several common survey methods for unmarked populations: spatially and temporally replicated counts, distance sampling data, and time-of-detection data. Simulations indicated relatively unbiased estimates for detection and availability probabilities. Negative bias in estimated superpopulation abundance was present with three temporally replicated surveys, but greatly reduced with six surveys. In a case study of Island Scrub-Jays (Aphelocoma insularis), posterior modes for presence, availability, and detection probabilities were 0.78, 0.96, and 0.26, respectively, from 10-min point counts repeated at 97 sites on three occasions, with noticeable differences among available, present, and superpopulation abundance estimates. This generalizable framework integrates common sampling protocols and provides joint inferences on the components of detection probability, spatial and non-spatial temporary emigration, and abundance in unmarked populations.

Population parameters of poorly marked gregarious species are difficult to estimate. This is the case for common dolphins (Delphinus sp.), a genus known for its lack of distinctive marks resulting in a low mark ratio. Furthermore, the widespread nature of common dolphins results in low recaptures. We developed reliable photo-identification protocols to ensure accurate identification of individuals in the Hauraki Gulf, New Zealand. These protocols combined the use of nicks and notches and pigmentation patterns for identification and included the development of a distinctiveness threshold. The data were further stratified by the level of distinctiveness of each individual (as distinctive or highly-distinctive). Photo-identification surveys were conducted from January 2010 to December 2013. Mark-recapture techniques were implemented through a POPAN super-population approach to estimate seasonal apparent survival, capture probability and abundance of dolphins. A total of 2,083 unique adult common dolphins were identified, 51.3% were classified as D1 (highly distinctive; n = 1,069) and 48.7% as D2 (distinctive; n = 1,014). Of all individuals identified, 34.3% (n = 704) were re-sighted over subsequent years. The proportion of marked dolphins (when compared to unmarked dolphins) was 26.3% for D1 and 46.4% for D1 & D2, respectively. Apparent survival was estimated at 0.767 (CI = 0.694-0.827) for D1 animals, and 0.796 (CI = 0.729-0.850) for D1 & D2 combined. For D1 only, seasonal abundance varied from 732 (CI = 460-1,177) in autumn 2010 to 5,304 (CI = 4,745-5,930) in spring 2013. While the inclusion of D2 individuals may offer a more precise estimate of total abundance, the inability to determine additional sources of bias (for example, arising from under or overestimated mark ratios) meant that estimates for D1 individuals were deemed the least biased for this population. The photo-identification protocol, stratification of the data and steps taken to eliminate potential model violations provided a useful and novel approach to estimate population parameters for common dolphins. These approaches could be implemented for other large gregarious populations (≥500 individuals) of animals with poor natural markings.

Species often exhibit preferences for certain habitat features such that species occurrence is a function of a unit's habitat. However habitat on the landscape will often be dynamic, changing through time via processes such as vegetation succession, human activities, and environmental variation. When habitat is dynamic, it will frequently be useful to partition species occurrence dynamics into components associated with habitat versus other factors. Solely focusing on species occurrence is likely to yield an inadequate descriptor of the system's state, as similar patterns may have very different meanings with respect to system well-being. In this chapter, we discuss methods that model system dynamics using two interrelated components, species occurrence dynamics that depend on habitat and habitat dynamics that (may) depend on species occurrence. Utilizing the multi-state, multi-season occupancy model framework described in Chapter 9, we highlight the flexible nature of this model to identify important factors for the separate but interrelated processes of habitat and occupancy dynamics. This enables a number of biologically interesting questions about these interrelationships, including species response to ecological succession, natural disturbance frequencies, climate change, and habitat loss and fragmentation associated with human activities. This modeling approach also provides a framework enabling predictions to be made about the future distribution of a species in the face of such dynamics.

Shorebird populations face increasing challenges as rising sea levels and growing human populations constrain their breeding habitats. On recreational beaches, the nesting season often coincides with a season of high visitor use, increasing the potential for conflict, which may negatively influence beach-nesting shorebird species. We designed a field experiment to study the responses of nesting American Oystercatchers (Haematopus palliatus) to off-road passenger vehicles (ORVs) at Cape Hatteras and Cape Lookout National Seashores in North Carolina, USA. We used continuous video and heart rate recordings to assess changes in the behavior and physiology of incubating oystercatchers. We conducted driving experiments affecting 7 nesting pairs in 2014 and 19 nesting pairs in 2015, between April and July of each year. Experimental treatments were repeated throughout the incubation period for each nest. Although responses were highly variable within and among pairs, paired randomized permutation tests indicated that, overall, oystercatcher pairs spent a greater proportion of time with their heads up and exhibited slower heart rates during driving treatments. Pairs also left their nests more frequently and attended their nests for a lower proportion of time during driving treatments, although these responses diminished over time. Higher nest attendance and lower departure rates late in incubation may have reflected a stronger attachment to nests closer to hatching or habituation to the driving treatment, although individuals continued to exhibit physiological responses to passing vehicles throughout incubation. Beach-nesting birds may benefit from reduced vehicle traffic at their nesting sites, allowing parents to spend more time attending the nest and less time on defensive behaviors.

The two-state and multi-state occupancy models described in Chapters 4 and 5, require a number of assumptions to be made about the data and biological system for inferences from the models to be valid. There will be situations where these assumptions are not reasonable and the modeling must be extended to provide a more accurate description of the system and collected data. In this chapter we detail a number of extensions that have been developed for this purpose, particularly for violations of the closure and independence assumptions. Topics covered include, estimating occupancy for a finite population or small area, false positive detections, multi-scale (or multi-method) occupancy, correlated detection process (autocorrelated survey outcomes), staggered entry-departure, and spatially correlated occupancy.

Species occurrence at locations of interest is often defined in terms of a dichotomous variable, e.g., species presence/absence, but a greater number of categories or states could be used. Single-season multi-state occupancy models have been developed to investigate and model patterns of species occurrence for two or more occupied states. As in the simpler case, the true occupancy state may not always be observed with certainty due to imperfect detection. In this chapter we detail the development of the single-season multi-state occupancy model, and provide examples of its application. This is a key chapter of this book as many of the more complex extensions presented later utilize this multi-state modeling framework.

Building on design recommendations from the previous chapter, we highlight additional issues that must be considered when designing a multiple-season occupancy study. Specifically, we discuss the time intervals between seasons, rotating panel designs, and trade-offs between the number of units and the number of seasons. We emphasize the importance of a probabilistic sampling scheme for selecting units to survey (i.e., the probability of a unit being selected from the population of units can be defined). Targeting units that are suspected to be occupied (e.g., based on recent presence records) may lead to apparent trends in occupancy, that could be misleading in some circumstances. In general, we recommend longitudinal studies (i.e., survey the same units over multiple seasons) where the interval between sampling seasons matches the temporal resolution of the motivating biological questions. As with single season studies, pilot studies and computer simulations are particularly valuable to assess alternative field methods, and to gain realistic expectations about the limits on what inferences could be drawn from the resulting data. Refining methods and understanding potential limitation before embarking on the full data collection phase, will enable resources to be used more efficiently, and may avoid future disappointments.

So far in this book, we have focused on scenarios where the occupancy status of only a single species is of interest. As outlined in Chapter 2, there are a number of ecological applications where inference is desired about the presence or absence of multiple species across a set of sampling units. In this chapter we turn our attention to investigations of co-occurrence for two or more species, from detection/nondetection data. Specifically, we demonstrate how imperfect detection can lead to misleading conclusions about the nature of any co-occurrence interaction between two species, and describe modeling approaches that have been developed to examine patterns, and the underlying dynamic processes, of species co-occurrence. These modeling approaches enable detection probabilities to be incorporated directly into our inferential framework, and are very flexible: allowing detection probabilities to depend on the number of species present at a unit, and for a potential lack of independence of species detections in the same survey occasions. Our presentation of the modeling approaches draws heavily from the methods described in previous chapters. Examples are provided of their application, investigating co-occurrence patterns for two salamander species in Great Smoky Mountains National Park, USA, and dynamics of northern spotted owl and barred owl interactions in western Oregon, USA.

Designing a good study is often as much an art as a science. It is a combination of taking the study objectives, knowledge about the intended statistical methods and species biology (the 'science'), with practical considerations because of the logistical limitations to obtain a design for collecting the required data (the 'art'). In this chapter, we discuss the importance of defining key occupancy components (e.g., unit, season, survey, etc.). We describe optimal designs that provide the most precise estimates of occupancy for a given set of resource constraints, and provide some general recommendations. The ideas in this chapter should not be regarded as an instruction book for how to design your study, but a play book of options that may be appropriate in some situations, but not in others. Studies should be designed on a case-by-case basis, as the details of each design (goals, species, environment, etc.) will often be different. Once a satisfactory design has been developed, we recommend that simulation and pilot studies are invaluable tools for assessing whether the design will provide the type and quality of information required to meet the studies' objectives. We remind readers of the 'Garbage In, Garbage Out' principle.

Species presence/absence or occurrence metrics are widely used in many areas of ecology, such as species range or distribution, habitat relationships, metapopulation dynamics, and monitoring. In this chapter we review many of the ecological, and some non-ecological, settings where such measures have been used, noting that in many cases imperfect detection, or false absences, have the potential to result in misleading inferences. We give examples of how occupancy modeling that accounts for imperfect detection could be, or has been, applied in these situations.

This chapter provides important background information about many of the fundamental statistical principles and methods used throughout this book. It is intended for those who may have had little formal training in statistics, however should be a valuable reference for any practitioner. Topics covered include random variables, likelihood functions, expected value and variance calculation, maximum likelihood and Bayesian methods of estimation, incorporating and interpreting predictor variables, hypothesis testing and model selection.

Another type of multi-species studies, are those focused on community-level metrics such as species richness. In this chapter we detail how some of the single-species occupancy models described in earlier chapters have been applied, or extended, for use in such studies, while accounting for imperfect detection. We highlight how Bayesian methods using MCMC are particularly useful in such settings to easily calculate relevant community-level summaries based on presence/absence data. These modeling approaches can be used to assess richness at a single point in time, or to investigate changes in the species pool over time.

Patterns in species occurrence may be caused by a variety of factors, which influence how a species is distributed across a region at any point in time. Nondetection of the species when it is present at a survey location may cause misleading inferences to be made about these patterns, particularly when detection probabilities are affected by some of the same factors that may influence species occurrence. In this chapter, methods that can be used to investigate patterns in species occurrence while accounting for imperfect detection are described, beginning with descriptions of older ad hoc approaches and then focusing on a more modern, model-based approach. Details are given on the underlying model structure, how to incorporate predictor variables or covariates, unequal sampling effort and assessing model fit. Examples of these methods being applied using maximum likelihood and Bayesian methods of statistical inference are also given.

[This corrects the article DOI: 10.1371/journal.pone.0198167.].