Keywords
nexrad, machine learning, climate change, snow goose
Start Date
1-7-2010 12:00 AM
Abstract
Radar ornithology has provided tools for studying the movement of birds,especially related to migration. Researchers have presented qualitative evidence suggestingthat birds, or at least migration events, can be identified using large broad scale radars suchas the WSR-88D used in the NEXRAD weather surveillance system. This is potentially aboon for ornithologists because such data cover a large portion of the United States, areconstantly being produced, are freely available, and have been archived since the early1990s. A major obstacle to this research, however, has been that identifying birds inNEXRAD data has required a trained technician to manually inspect a graphically renderedradar sweep. A single site completes one volume scan every five to ten minutes, producingover 52,000 volume scans in one year. This is an immense amount of data, and manualclassification is infeasible. We have developed a system that identifies biological echoesusing machine learning techniques. This approach begins with training data using scans thathave been classified by experts, or uses bird data collected in the field. The data arepreprocessed to ensure quality and to emphasize relevant features. A classifier is thentrained using this data and cross validation is used to measure performance. We comparedneural networks, naive Bayes, and k-nearest neighbor classifiers. Empirical evidence isprovided showing that this system can achieve classification accuracies in the 80th to 90thpercentile. We propose to apply these methods to studying bird migration phenology andhow it is affected by climate variability and change over multiple temporal scales.
New software methods in radar ornithology using WSR-88D weather data and potential application to monitoring effects of climate change on bird migration
Radar ornithology has provided tools for studying the movement of birds,especially related to migration. Researchers have presented qualitative evidence suggestingthat birds, or at least migration events, can be identified using large broad scale radars suchas the WSR-88D used in the NEXRAD weather surveillance system. This is potentially aboon for ornithologists because such data cover a large portion of the United States, areconstantly being produced, are freely available, and have been archived since the early1990s. A major obstacle to this research, however, has been that identifying birds inNEXRAD data has required a trained technician to manually inspect a graphically renderedradar sweep. A single site completes one volume scan every five to ten minutes, producingover 52,000 volume scans in one year. This is an immense amount of data, and manualclassification is infeasible. We have developed a system that identifies biological echoesusing machine learning techniques. This approach begins with training data using scans thathave been classified by experts, or uses bird data collected in the field. The data arepreprocessed to ensure quality and to emphasize relevant features. A classifier is thentrained using this data and cross validation is used to measure performance. We comparedneural networks, naive Bayes, and k-nearest neighbor classifiers. Empirical evidence isprovided showing that this system can achieve classification accuracies in the 80th to 90thpercentile. We propose to apply these methods to studying bird migration phenology andhow it is affected by climate variability and change over multiple temporal scales.
