Taking a Big Data Approach to Estimating Species Abundance
Date: 2016-12-07 04:15 PM – 04:30 PM
Last modified: 2016-10-15
Knowledge of species distributions is the foundation of biodiversity conservation. While estimates of occurrence are typical of distribution estimates, incorporation of abundance can provide critical information about the status and trends of a species.
Many species have dynamic distribution patterns that change seasonally and distribution visualizations attempt to account for this. Additionally, the sampling of species distributions are rarely uniformly distributed across a species range, and other data sources are often necessary to generate smooth surfaces of distributions.
Here we describe a big data workflow using data from the citizen science project eBird which we combine with NASA earth imagery to estimate patterns of seasonal bird abundances across a species entire life history. Methods. Our goal is to train a model that learns associations between eBird observations and environmental features estimated using NASA earth imagery. eBird is an online database of bird observations where participants enter when, where, and how they went birding, then fills out a checklist of all the birds seen and heard during the outing. eBird provides various options for data gathering including point counts, transects, and area searches. NASA MODIS provides information on local-scale ecological processes (i.e., habitat) at a comprehensive global coverage allowing us to fully use eBird data. We interpolate predicted patterns of abundance across geographic regions based on the environmental conditions within those regions. The model creates weekly abundance estimates in a 3x3 km grid across the Western Hemisphere. Results. We have generated models estimating the patterns of abundance and habitat preferences across a species’ full life cycle. This allows us to explore the dynamic patterns of habitat preference; estimate regional and seasonal abundance; calculate the percent of the annual cycle a species occurs anywhere across their full annual distribution based on weekly estimates of relative abundance. Conclusions. Our modeling approach is the first to document distributional dynamics of migratory birds across the annual cycle and, in doing so, highlight the strong spatiotemporal variation that has been obscured by traditional range and distribution maps. The combination of fine-scale and broad extent of the NASA data make it possible to develop comprehensive biodiversity visualizations that can be integrated across a range of spatial and temporal scales. Our success is based on open access to data; developing data interoperability techniques that link heterogeneous data. The outcome is species distributions that are temporally explicit across broad spatial areas at high resolution.