Adaptive Spatio-Temporal Exploratory Models: Hemisphere-wide species distributions from massively crowdsourced eBird data

Authors

  • Daniel Fink Cornell Lab of Ornithology
  • Theodoros Damoulas Cornell University
  • Jaimin Dave Cornell University

DOI:

https://doi.org/10.1609/aaai.v27i1.8484

Keywords:

Species Distribution Models, Big Data, Crowdsourcing, Ensemble Models, Mixture Models

Abstract

Broad-scale spatiotemporal processes in conservation and sustainability science, such as continent-wide animal movement, occur across a range of spatial and temporal scales. Understanding these processes at multiple scales is crucial for developing and coordinating conservation strategies across national boundaries. In this paper we propose a general class of models we call AdaSTEM, for Adaptive Spatio-Temporal Exploratory Models, that are able to exploit variation in the density of observations while adapting to multiple scales in space and time. We show that this framework is able to efficiently discover multiscale structure when it is present, while retaining predictive performance when absent. We provide an empirical comparison and analysis, offer theoretical insights from the ensemble loss decomposition, and deploy AdaSTEM to estimate the spatiotemporal distribution of Barn Swallow (Hirundo rustica) across the Western Hemisphere using massively crowdsourced eBird data.

Downloads

Published

2013-06-29

How to Cite

Fink, D., Damoulas, T., & Dave, J. (2013). Adaptive Spatio-Temporal Exploratory Models: Hemisphere-wide species distributions from massively crowdsourced eBird data. Proceedings of the AAAI Conference on Artificial Intelligence, 27(1), 1284-1290. https://doi.org/10.1609/aaai.v27i1.8484

Issue

Section

Computational Sustainability and Artificial Intelligence