Page 129 - GIS for Science, Volume 3 Preview

Conserving biodiversity
Biodiversity is under threat globally from anthropogenic pressures such as habitat loss, fragmentation, and degradation. The World Wildlife Fund for Nature’s Living Planet Report 2016 found that populations of vertebrate animals have declined by 58% overall from 1970 to 2012, with freshwater species dropping as much as 81%. The National Audubon Society’s 2014 Birds and Climate Change Report found that habitat loss caused by global warming threatened more than half of the 588 North American bird species in the study. Disturbances such as hurricanes cause significant erosion and extensive land cover change, resulting in loss of critical habitats important to breeding, migratory, and wintering birds.
To conserve coastal birds and other biodiversity, it’s essential to assess disturbance effects quickly and accurately and identify areas in critical need of habitat restoration. Bird watchers are among the most enthusiastic citizen scientists. For years, the National Audubon Society has relied on this volunteer community to help count birds. However, volunteer bird monitoring is often extremely difficult to do well, particularly for sites that have limited accessibility or can only be surveyed during narrow time windows (such as islands with tide-limited beach access). Given the challenges of traditional bird counting, Audubon is augmenting in-person monitoring with drones. Drones, however, produce a large amount of data and require extensive computer resources and human labor to process using traditional analytical methods.
Audubon’s new project to advance the rapid evaluation of bird monitoring and habitat assessment is twofold: documenting changes in bird habitats because of disturbances such as major storms, and more accurately counting the various species of birds living in those habitats. Developing machine learning and cognitive algorithms that can rapidly process remotely sensed data to census birds and wildlife and classify land cover will change how Audubon accomplishes its conservation goals. To this end, high-resolution imagery from drones and aerial surveys, and lidar elevation data for land cover classification was loaded into Azure data science virtual machines for high-throughput processing with ArcGIS Pro. Machine learning algorithms were implemented in Machine Learning Studio using the MicrosoftML package, and additional tools, such as the ArcGIS Pro Spatial Analyst Image Classification tool, were used to help train the algorithms. Once trained, these algorithms enabled Audubon to quickly obtain accurate count and colony size data from remote sites and across large spatial scales that were impractical or impossible to attempt using traditional ground-based surveys. The algorithms developed from this project will vastly increase the speed of image processing, identification, and count estimation for future surveys, empowering Audubon to meet the conservation challenges posed by accelerating global change.
Nesting bird locations can be safely photographed with drones and then mapped and processed with machine learning to arrive at accurate counts.
AI for Geospatial Analysiss 117