Probabilistic inference in high-dimensional state-space models is computationally challenging. For many spatiotemporal systems, however, prior knowledge about the dependency structure of state variables is available. We leverage this structure to develop a computationally efficient approach to state estimation and learning in graph-structured state-space models with (partially) unknown dynamics and limited historical data. Building on recent methods that combine ideas from deep learning with principled inference in Gaussian Markov random fields (GMRF), we reformulate graph-structured state-space models as Deep GMRFs defined by simple spatial and temporal graph layers. This results in a flexible spatiotemporal prior that can be learned efficiently from a single time sequence via variational inference. Under linear Gaussian assumptions, we retain a closed-form posterior, which can be sampled efficiently using the conjugate gradient method, scaling favourably compared to classical Kalman filter based approaches.

The analysis of trajectories has become an important field in geographic visualization, as cheap GPS sensors have become commonplace and, in many cases, valuable information can be derived either from the data themselves or their metadata if processed and visualized in the right way. However, showing the "right" information to highlight dependencies or correlations between different measurements remains a challenge, because the technical intricacies of applying a combination of automatic and visual analysis methods prevents the majority of domain experts from analyzing and exploring the full wealth of their movement data. This paper presents an exploration through enrichment approach, which enables iterative generation of metadata based on exploratory findings and is aimed at enabling domain experts to explore their data beyond traditional means.

Studying animal movement and distribution is of critical importance to addressing environmental challenges including invasive species, infectious diseases, climate and land-use change. Motion sensitive camera traps offer a visual sensor to record the presence of a species at a location, recording their movement in the Eulerian sense. Modern digital camera traps that record video present new analytical opportunities, but also new data management challenges. This paper describes our experience with a year-long terrestrial animal monitoring system at Barro Colorado Island, Panama. The data gathered from our camera network shows the spatio-temporal dynamics of terrestrial bird and mammal activity at the site-data relevant to immediate science questions, and long-term conservation issues. We believe that the experience gained and lessons learned during our year long deployment and testing of the camera traps are applicable to broader sensor network applications and are valuable for the advancement of the sensor network research. We suggest that the continued development of these hardware, software, and analytical tools, in concert, offer an exciting sensor-network solution to monitoring of animal populations which could realistically scale over larger areas and time spans.