Ultra-Fine Scale Spatially-Integrated Mapping of Habitat and Occupancy Using Structure-From-Motion

Abstract

Organisms respond to and often simultaneously modify their environment. While these interactions are apparent at the landscape extent, the driving mechanisms often occur at very fine spatial scales. Structure-from-Motion (SfM), a computer vision technique, allows the simultaneous mapping of organisms and fine scale habitat, and will greatly improve our understanding of habitat suitability, ecophysiology, and the bi-directional relationship between geomorphology and habitat use. SfM can be used to create high-resolution (centimeter-scale) three-dimensional (3D) habitat models at low cost. These models can capture the abiotic conditions formed by terrain and simultaneously record the position of individual organisms within that terrain. While coloniality is common in seabird species, we have a poor understanding of the extent to which dense breeding aggregations are driven by fine-scale active aggregation or limited suitable habitat. We demonstrate the use of SfM for fine-scale habitat suitability by reconstructing the locations of nests in a gentoo penguin colony and fitting models that explicitly account for conspecific attraction. The resulting digital elevation models (DEMs) are used as covariates in an inhomogeneous hybrid point process model. We find that gentoo penguin nest site selection is a function of the topography of the landscape, but that nests are far more aggregated than would be expected based on terrain alone, suggesting a strong role of behavioral aggregation in driving coloniality in this species. This integrated mapping of organisms and fine scale habitat will greatly improve our understanding of fine-scale habitat suitability, ecophysiology, and the complex bi-directional relationship between geomorphology and habitat use.

A computer vision for animal ecology

“…Applications of computer vision to describing ecological objects. (1) From McDowall and Lynch (), a three‐dimensional map of the Port Lockroy penguin colony was created by overlaying hundreds of individual photographs (1a) to describe the location of Gentoo penguin ( P ygoscelis papua ) nests (1b). Flags denote occupied penguin nests identified in the images.…”

“…While traditional remote sensing captures coarse changes in habitat quality, animals experience the environment at fine‐scales, in three dimensions, and from a landscape perspective. McDowall and Lynch () generated ultra‐fine scale (<1 cm) maps of penguin colonies by stitching together thousands of overlapping images using a technique called structure‐from‐motion. The resulting three‐dimensional surface allowed fine‐scale mapping of Gentoo penguin ( Pygoscelis papua ) nests and captured variation in slope and aspect that may have been missed by coarser satellite‐based remote sensing (Figure a).…”

When the “selfish herd” becomes the “frozen herd”: spatial dynamics and population persistence in a colonial seabird

“…We therefore approximate continuous space as a discrete hexagonal grid, in which each cell represents a potential nest site. This approximation reduces computational complexity, incorporates the approximately hard‐core repulsion between nest locations at short distance (McDowall and Lynch ), and allows us to model isotropic interactions among nest sites. To separate the effects of habitat quality and intraspecific interactions, we constructed Bayesian auto‐logistic use‐availability models for Adélie Penguins on this hexagonally gridded landscape.…”

“…Fitting this model requires information on both the location of individual nests and the abiotic characteristics of all available nest sites. We use a high‐resolution digital elevation model (DEM), created through a photogrammetric process applied to aerial (UAV) imagery captured at Beagle Island, Antarctica (McDowall and Lynch , Borowicz et al. ), as the basis for this model.…”

Species and habitat mapping in two dimensions and beyond. Structure-from-Motion Multi-View Stereo photogrammetry for the Conservation Community

“…From this, SfM-MVS software is able to output two dimensional orthographic images containing detailed geographical location information, alongside 2.5 dimensional reconstructions (2.5D is used as it relates to the limitations of algorithms to produce true 3D reconstructions from aerial images (8)), such as digital elevation, surface and terrain models (DEM, DSM, DTM) (Figure 2). These data products can be used for mapping habitats (9–11), analysing structure, biomass, topography and change (12–15) but also crucially, we suggest, offer the capability to determine the geographic position of species in both two and three dimensions (16–18), thus being of great value for conservation biologists.…”