Studying biodiversity at sub-pixel scales in Landsat imagery

The Landsat satellite archive provides an extraordinary time series of environmental change, and can be used to either observe biodiversity directly or variables related to biodiversity, but many key processes occur at spatial scales too small to be studied at its native resolution. Here we combine a model for the detection of key landscape features, modified from occupancy modelling, with an automated classification approach that can detect the spectral signature of penguin guano and thus map the location and size of penguin colonies throughout Antarctica. This approach permits sub-pixel resolution inference on the size of penguin colonies and is easily modified for other biodiversity applications that evolve at a slower time scale than the repeat frequency of Landsat. We discuss some of the technical challenges of tracking very small landscape features over several decades, such as the need for precise georectification (in a region with few ground control points), and highlight future opportunities for fusing such sub-pixel inference with spatial spread models that would permit a more nuanced understanding of a target species' spatial ecology.