Collaborative Research: ABI Innovation: Dark Ecology: Deep Learning and Massive Gaussian Processes to Uncover Biological Signals in Weather Radar

点击翻译按钮获取中文摘要

Every spring and fall billions of birds migrate across the US, largely under the cover of darkness. Data collected by the US network of weather radars and new analysis methods let us track these migrations. The Dark Ecology Project will develop new resources allowing us to estimate the densities of migrating birds as they have changed in the last 20 years. One outcome will be our better ability to monitor bird populations and their migration systems, and the impacts of various environmental factors. The US network of weather radars has recorded a comprehensive 25-year archive of images of the atmosphere, which provides the baseline information about bird movements. Extracting biological information from the images is not automated currently, making it very slow and inefficient. A team of ecologists and computer scientists will conduct novel research combining methods in computer vision and machine learning to unlock detailed information about bird migration from the entire US archive of weather radar data. The resulting dataset will be freely available, providing an information resource for researchers to estimate the number of birds migrating on any given night, measure the patterns and trends of bird populations, and do hypothesis driven science. The research will advance big data analysis and visualization techniques for large-scale science questions, and will engage scientists, conservation planners, students, and the general public with data, visualizations, and educational material about bird migration.Dark Ecology will leverage large-scale cloud computing and develop novel computer vision, machine learning, and radar analysis methods to measure the densities and velocities of migrating birds across the US. Deep convolutional networks will be trained to discriminate migrating birds from precipitation and other clutter in the radar data. New techniques for domain transfer and weakly supervised training will enable the training of convolutional networks with only modest-sized training sets. Gaussian process (GP) models will be developed to create smooth national maps of migration density and velocity. Novel GP methods and cloud-computing workflows will allow us to scale to massive radar data sets and analyze the more then 200 million archived radar scans. The resulting data and tools will be curated with open access policies, and used by the research team to conduct ecological research about patterns and drivers of continent-scale migration. Project information can be found at http://darkecology.cs.umass.edu.