Estimating indices of range shifts in birds using dynamic models when detection is imperfect

Abstract

There is intense interest in basic and applied ecology about the effect of global change on current and future species distributions. Projections based on widely used static modeling methods implicitly assume that species are in equilibrium with the environment and that detection during surveys is perfect. We used multiseason correlated detection occupancy models, which avoid these assumptions, to relate climate data to distributional shifts of Louisiana Waterthrush in the North American Breeding Bird Survey (BBS) data. We summarized these shifts with indices of range size and position and compared them to the same indices obtained using more basic modeling approaches. Detection rates during point counts in BBS surveys were low, and models that ignored imperfect detection severely underestimated the proportion of area occupied and slightly overestimated mean latitude. Static models indicated Louisiana Waterthrush distribution was most closely associated with moderate temperatures, while dynamic occupancy models indicated that initial occupancy was associated with diurnal temperature ranges and colonization of sites was associated with moderate precipitation. Overall, the proportion of area occupied and mean latitude changed little during the 1997-2013 study period. Near-term forecasts of species distribution generated by dynamic models were more similar to subsequently observed distributions than forecasts from static models. Occupancy models incorporating a finite mixture model on detection - a new extension to correlated detection occupancy models - were better supported and may reduce bias associated with detection heterogeneity. We argue that replacing phenomenological static models with more mechanistic dynamic models can improve projections of future species distributions. In turn, better projections can improve biodiversity forecasts, management decisions, and understanding of global change biology.

Partitioning global change: Assessing the relative importance of changes in climate and land cover for changes in avian distribution

“…Our results correlating initial occupancy with environmental covariates are, broadly speaking, consistent with biological expectation and a raft of previous studies that have indicated a significant relationship between species distributions and habitat (Robinson, Wilson, & Crick, ; Thogmartin, Sauer, & Knutson, ), climate (Barbet‐Massin & Jetz, ; Bellard, Bertelsmeier, Leadley, Thuiller, & Courchamp, ), or both (Seoane, Bustamante, & Díaz‐Delgado, ; Sohl, ). However, we suggest that estimated relationships between spatial variation in environmental covariates and colonization and extinction rates are of greater ecological interest than relatively phenomenological species distribution models or climate envelope models (Clement et al, ). The estimated vital rates of colonization and extinction bring us closer to understanding the dynamic process underlying the distribution of these species.…”

“…We used correlated‐detection models because the BBS generates spatially replicated surveys, and failure to account for spatial correlation of replicates can bias occupancy estimates (Hines et al, ). Finally, we used a finite mixture model to account for detection heterogeneity because of the variation in habitat and focal species abundance across the study area (Clement et al, ). A finite mixture model approximates the heterogeneity of detection probabilities by positing that the population consists of a mixture of routes which have either a relatively high detection probability or a relatively low detection probability.…”

Use of classical bird census transects as spatial replicates for hierarchical modeling of an avian community

“…Another relevant property of community occupancy models is that they allow occupancy probability to be obtained in relation to environmental covariates and to analyze climate change effects (Clement et al, ). Our results show that a higher percentage of forest cover reduces slightly the mean occupancy of bird communities.…”

“…Occupancy models provide estimates of occurrence probability for species that are corrected for imperfect detection (Bailey, MacKenzie, & Nichols, ; MacKenzie et al, ). These models enable us to rigorously evaluate the effects of environmental variables on occupancy probability, mapping species range dynamics (Kéry, Guillera‐Arroita, & Lahoz‐Monfort, ; Santika, McAlpine, Lunney, Wilson, & Rhodes, ), study the interactions between species (Michel, Jiménez‐Franco, Naef‐Daenzer, & Grüebler, ; Yackulic et al, ) and evaluate the effects of climate change (Clement, Hines, Nichols, Pardieck, & Ziolkowski, ). Multispecies occupancy models are a more complex framework, aimed at estimating total community richness (Dorazio & Royle, ; Dorazio, Royle, Söderström, & Glimskär, ; Kéry & Royle, ) and few studies have evaluated the effects of different habitats (Zipkin, DeWan, & Royle, ) and range shift over two different periods (Moritz et al, ; Tingley & Beissinger, ).…”

Local temperature and ecological similarity drive distributional dynamics of tropical mammals worldwide

“…We modelled population‐specific local colonization and extinction along elevational gradients using dynamic (multi‐season) occupancy models (MacKenzie, Nichols, Hines, Knutson, & Franklin, ). This model is an appropriate and robust tool because it accounts for imperfect detection by camera traps (MacKenzie et al, ; Royle & Dorazio, ) and makes few assumptions about equilibrium or pseudo‐equilibrium (Clement, Hines, Nichols, Pardieck, & Ziolkowski, ). The dynamic occupancy modelling approach has a similar sampling scheme as Pollock’s robust design for mark‐recapture studies (Pollock, ).…”