Browsing by Subject "Spatial"

Pinniped species undergo a life history, unique among marine mammals, that includes discrete periods of occupancy on land or ice within a predominantly marine existence. This makes many pinniped species valuable sentinels of marine ecosystem health and models of marine mammal physiology and behavior. Pinniped research has often progressed hand-in-hand with advances at the technological frontiers of wildlife biology, and drones represent a leap forward in the long-established field of aerial photography, heralding opportunities for data collection and integration at new scales of biological importance. The following chapters employ and evaluate recent and emerging methods of wildlife surveillance that are uniquely enabled and facilitated by drone methods, in applied research and management campaigns with near-polar pinniped species. These methods represent advancements in abundance estimation and distribution modeling of pinniped populations that are dynamically shifting amid climate change, fishing pressure, and recovery from historical depletion.Conventional methods of counting animals from aerial imagery—typically visual interpretation by human analysts—can be time-consuming and limits the practical use of this data type. Deep learning methods of computer vision can ease this burden when applied to drone imagery, but are not yet characterized for practical and generalized use. To this end, I used a common implementation of deep learning for object detection in imagery to train and test models on a variety of datasets describing breeding populations of gray seals (Halichoerus grypus) in the northwest Atlantic Ocean (Chapter 2). I compare standardized performance metrics of models trained and tested on different combinations of datasets, demonstrating that model performance varies depending on both training and testing data choices. We find that models require careful validation to estimate error rates, and that they can be effectively deployed to aid, but not replace, conventional human visual interpretation of novel datasets for gray seal detection, location, age-classification and abundance estimation. Spatial analysis and species distribution modeling can use fine-scale drone-derived data to describe local species–habitat relationships at the scale of individual animals. I applied structure-from-motion methods to a survey of three pinniped species, pacific harbor seals (Phoca vitulina richardii), northern fur seals (Callorhinus ursinus), and Steller sea lions (Eumetopias jubatus), in adjacent non-breeding haul-outs to compare occupancy and habitat selection (Chapter 3). I describe and compare fitted occupancy models of pacific harbor seals and northern fur seals, finding that conspecific attraction is a key driver of habitat selection for each species, and that each species exhibits distinct topographic preferences. These findings illustrate both opportunities and limitations of spatial analysis at the scale of individual pinnipeds. Ease of deployment and rapid data collection make drones a powerful tool for monitoring populations of interest over time, while animal locations, revealed in high-resolution imagery, and contextual habitat products can reveal spatial relationships that persist beyond local contexts. I designed and carried out a campaign of drone surveillance over coastal habitats near Palmer Station, Antarctica, in the austral summer of 2020 to assess the seasonal abundance and habitat use of Antarctic fur seals (Arctocephalus gazella) in the Palmer Archipelago and adjacent regions (Chapter 4). I modeled abundance as a function of date, with and without additional terms to capture variance by site, and used these models to estimate peak abundance near Palmer Station in the 2020 summer season. These findings leverage the spatial and temporal advantages of drone methods to estimate species phenology, distribution and abundance. Together, these chapters describe emerging applications of drone technology that can advance pinniped research and management into new scales of analytical efficiency and ecological interpretation. These studies describe methods that have been proven in concept, but not yet standardized for practical deployment, and their findings reveal new ecological insights, opportunities for methodological advancement, and current limitations of drone methods for the study of pinnipeds in high-latitude environments.

In this thesis, temporal and spatial dependence are considered within nonparametric priors to help infer patterns, clusters or segments in data. In traditional nonparametric mixture models, observations are usually assumed exchangeable, even though dependence often exists associated with the space or time at which data are generated.

Focused on model-based clustering and segmentation, this thesis addresses the issue in different ways, for temporal and spatial dependence.

For sequential data analysis, the dynamic hierarchical Dirichlet process is proposed to capture the temporal dependence across different groups. The data collected at any time point are represented via a mixture associated with an appropriate underlying model; the statistical properties of data collected at consecutive time points are linked via a random parameter that controls their probabilistic similarity. The new model favors a smooth evolutionary clustering while allowing innovative patterns to be inferred. Experimental analysis is performed on music, and may also be employed on text data for learning topics.

Spatially dependent data is more challenging to model due to its spatially-grid structure and often large computational cost of analysis. As a non-parametric clustering prior, the logistic stick-breaking process introduced here imposes the belief that proximate data are more likely to be clustered together. Multiple logistic regression functions generate a set of sticks with each dominating a spatially localized segment. The proposed model is employed on image segmentation and speaker diarization, yielding generally homogeneous segments with sharp boundaries.

In addition, we also consider a multi-task learning with each task associated with spatial dependence. For the specific application of co-segmentation with multiple images, a hierarchical Bayesian model called H-LSBP is proposed. By sharing the same mixture atoms for different images, the model infers the inter-similarity between each pair of images, and hence can be employed for image sorting.

Functional magnetic resonance imaging (fMRI) is a major neuroimaging methodology and have greatly facilitate basic cognitive neuroscience research. However, there are multiple statistical challenges in the analysis of fMRI data, including, dimension reduction, multiple testing and inter-dependence of the MRI responses. In this thesis, a spatial Bayesian variable selection (BVS) model is proposed for the analysis of multi-subject fMRI data. The BVS framework simultaneously account for uncertainty in model specific parameters as well as the model selection process, solving the multiple testing problem. A spatial prior incorporate the spatial relationship of the MRI response, accounting for their inter-dependence. Compared to the non-spatial BVS model, the spatial BVS model enhances the sensitivity and accuracy of identifying activated voxels.