Geospatial web services within a scientific workflow: Predicting marine mammal habitats in a dynamic environment

Our ability to inform conservation and management of species is fundamentally limited by the availability of relevant biogeographic data, use of statistically robust predictive models, and presentation of results to decision makers. Despite the ubiquity of presence-only models, where available, survey effort should be included in the modeling process to limit spatial bias. The biogeographic archive therefore should be able to store and serve related spatial information such as lines of survey effort or polygons of the study area, best accomplished through geospatial web services such as the Open Geospatial Consortium (OGC) Web Feature Service (WFS). Ideally data could then be easily fetched by modelers into a scientific workflow, providing a visually intuitive, modular, reusable canvas for linking analytical processes without the need to code. Species distribution model results should be easily accessible to decision makers, such as through a web-based spatial decision support system (SDSS). With these principles in mind, we describe our progress to date serving marine animal biogeographic data from OBIS-SEAMAP (http://seamap.env.duke.edu), and consuming the data for predictive environmental modeling of cetaceans. Using geospatial web services to automate the scientific workflow process, marine mammal observations from OBIS-SEAMAP are used to sample through date-synchronous remotely sensed satellite data for building multivariate habitat models using a variety of statistical techniques (GLM, GAM, and CART). We developed custom scientific workflows using ESRI Model Builder, ArcGIS geoprocessor, R statistical package, Python scripting language, PostGIS geodatabase, and UMN MapServer. These model outputs are then passed to an SDSS with spatial summary capability. Custom products will be open-source and freely available. In the future, we hope to integrate technologies such as OGC WCS, OPeNDAP, and Kepler. The principles and lessons described here can be broadly applied to serving biogeographic data, species distribution modeling, and decision support within the ecological informatics community.