Tuesday, August 3, 2010: 8:20 AM
325, David L Lawrence Convention Center
Background/Question/Methods
Identifying general principles in movement ecology and integrating studies of different species at different temporal resolutions requires an understanding of how actual animal movement changes across scales. An important advance in animal movement ecology has been the development of models that can assign latent (behavioral) states to different groups of observations. For this class of models, our ability to scale up from observations at fine resolutions to displacement at coarser resolutions may depend on our assumptions about the appropriate temporal resolution of latent states and on the way in which states are defined. Here we rely on movement data for eight individual African forest elephants to investigate how the choice of model structure interacts with the temporal resolution of sampling to affect our understanding of animal movement and ability to scale up to coarser resolutions. We first parameterize four movement models with data collected at four different temporal resolutions and use model comparison to evaluate their fit to observed data. All four models use a novel statistical distribution that allows us to accommodate multimodal distributions of turning angles independent of step length. The simplest model has only one state, while more complex models allow for multiple states determined by either turning angles or step length. After using these models to understand movement patterns at different temporal resolutions, we simulate data using parameters estimated at the finest resolution to determine the relative efficacy of different models in reproducing movement observed at coarser temporal resolutions.
Results/Conclusions
Using a model comparison framework, we conclude that multiple state models do a better job of representing forest elephant movement when steps are measured at finer temporal resolutions while the simplest model, with only one state, is preferred at coarser resolutions. Interestingly, when states are defined by step length, it is preferable to estimate latent states over coarser temporal resolutions. Simulated datasets recreate many aspects of the observed data at 12 and 24 hours, but underestimate the percentage of reversals for all models. The only model to accurately predict displacement at coarser time scales defines states based on turning angles rather than step length. This finding is interesting because past theoretical studies have emphasized the importance of variation in step length in determining scaling in net displacement. We conclude by discussing future directions in the modeling of animal movement across scales, including the need to critically examine the link between latent states and behavior.
Identifying general principles in movement ecology and integrating studies of different species at different temporal resolutions requires an understanding of how actual animal movement changes across scales. An important advance in animal movement ecology has been the development of models that can assign latent (behavioral) states to different groups of observations. For this class of models, our ability to scale up from observations at fine resolutions to displacement at coarser resolutions may depend on our assumptions about the appropriate temporal resolution of latent states and on the way in which states are defined. Here we rely on movement data for eight individual African forest elephants to investigate how the choice of model structure interacts with the temporal resolution of sampling to affect our understanding of animal movement and ability to scale up to coarser resolutions. We first parameterize four movement models with data collected at four different temporal resolutions and use model comparison to evaluate their fit to observed data. All four models use a novel statistical distribution that allows us to accommodate multimodal distributions of turning angles independent of step length. The simplest model has only one state, while more complex models allow for multiple states determined by either turning angles or step length. After using these models to understand movement patterns at different temporal resolutions, we simulate data using parameters estimated at the finest resolution to determine the relative efficacy of different models in reproducing movement observed at coarser temporal resolutions.
Results/Conclusions
Using a model comparison framework, we conclude that multiple state models do a better job of representing forest elephant movement when steps are measured at finer temporal resolutions while the simplest model, with only one state, is preferred at coarser resolutions. Interestingly, when states are defined by step length, it is preferable to estimate latent states over coarser temporal resolutions. Simulated datasets recreate many aspects of the observed data at 12 and 24 hours, but underestimate the percentage of reversals for all models. The only model to accurately predict displacement at coarser time scales defines states based on turning angles rather than step length. This finding is interesting because past theoretical studies have emphasized the importance of variation in step length in determining scaling in net displacement. We conclude by discussing future directions in the modeling of animal movement across scales, including the need to critically examine the link between latent states and behavior.