Q&A: Improving Aerial Surveys of Geese in Alaska with Aerial Imagery
Thousands of geese gather at Izembek Lagoon in southwestern Alaska every fall where they “stage”, meaning that they rest and eat in preparation for migration to lower latitudes. Izembek Lagoon is especially important for Pacific brant geese, as the entire Pacific Flyway population is thought to use the lagoon in fall. This provides an opportunity to efficiently survey the population to track annual abundance and productivity. New aerial imagery and automated counting methods are helping to optimize surveys and improve population estimates.
Return to USGS Alaska Q&A Series
A primary goal of the Migratory Bird Management Office of the U.S. Fish and Wildlife Service (USFWS) is to understand the annual abundance and trends of migratory bird populations across North America. Each year, the USFWS works with numerous partners to survey and estimate abundance. The USFWS recently collaborated with the USGS Alaska Science Center and developed new methods to count geese at Izembek Lagoon in southwestern Alaska. In this Q&A, we talk with Julian Fischer (USFWS Migratory Bird Management Office in Anchorage, Alaska) and Emily Weiser (USGS Alaska Science Center) about these new image and automated counting methods and what they might mean for the future of aerial counting and monitoring bird populations.
Q: How have waterfowl surveys at Izembek Lagoon been conducted in the past?
Julian: Waterfowl at Izembek Lagoon have been surveyed from small airplanes since 1976. This fall staging survey involves a pilot-biologist, who flies the plane and counts the birds on one side of the plane, and a second observer who counts birds on the other side of the plane. The two surveyors identify birds and estimate the numbers that they see out the window as they fly. This is challenging during the fall staging period because the birds are often in mixed-species flocks of hundreds or even thousands of birds, so it’s difficult to accurately estimate the number present. To see the birds well enough to identify and count them, the pilot maneuvers the plane low over the water (45 m, or 150 ft, above the water), which can cause birds to flush in front of the aircraft, making it even more difficult to count them.
Q: How does the new photo survey work?
Emily: The photo survey is flown with the same type of small aircraft but uses cameras instead of human observers. Two cameras mounted in the belly of the aircraft take photos of the lagoon as the plane flies over. Software controls the cameras and automatically triggers them as the plane flies through lines of pre-set photo points. The cameras have 200 mm lenses and high-resolution sensors, so zooming in on a photo provides a clear image of the birds on the water below, even at high altitudes.
Disclaimer statement: Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Dennis Marks USFWS photographer.
Q: How is the photo survey an improvement over the past method of visually counting birds out the window of the plane?
Julian: The photo survey is an improvement in three ways. First, the photo transects represent a known sample of the area of the lagoon. This repeatable design can be replicated annually and provides data analysts the ability to estimate total numbers of birds present and calculate the precision of that estimate. Second, because the cameras capture images at high resolution, birds can be identified and counted even when the aircraft flies at a higher altitude (457 m, or 1500 ft). Flying at higher altitudes prevents birds from moving during the survey, thereby improving accuracy of the survey, as well as minimizing impacts of the survey on the birds. Moreover, the higher altitude provides an increased margin of safety for the aircrew. Third, the photos provide a permanent record of the birds that were present, identified, and counted and provides greater accuracy than visually identifying and counting birds in large flocks.
Q: How are the photos used to develop an estimate of population size?
Emily: Birds in the photos must be identified and counted. And there are a lot of images! We collect 8,000-10,000 images on each flight, so it would be time-consuming to manually count birds in that many photos. We worked with software engineers to develop a computer program to automatically detect, identify, and count birds of each type (Brant, Cackling Geese, Emperor Geese, Gulls, and “Other”, which is mostly ducks). About 95% of photos contain no birds and thus require no further assessment. When the program detects birds in a photo, a person checks the computer identifications and counts to correct any mistakes made by the program. Accuracy of the program is currently being improved, but the fact that it filters out 95% of photos with no birds greatly reduces the workload.
The number of birds of each type in the photos is then translated to the estimated total number on the entire lagoon. The photos sample the lagoon in a regular grid pattern, but miss some of the geese between photos, so the full population is not captured in the photos. However, each photo covers a known area of water, so counts of birds in the photo can be converted into density. Extrapolating that density across the full area of the lagoon provides an estimate of the total number of birds present, very similar to how USGS recently estimated the number of walruses on a coastal haulout using drone imagery. Unlike other waterfowl species, virtually the entire Pacific population of Brant occurs at Izembek Lagoon in fall, thus the survey provides a valuable measure of abundance for this species of management concern. Additionally, the survey counts Cackling geese, the next most abundant species in the area, so that population estimates can be developed.
Q: The new photo survey provided considerably higher estimates of total population size than with the traditional counting method. Do you think applying this method to other areas and other species would also result in higher estimates than determined in the past?
Julian: The greatest improvement in accuracy resulting from photographic methods is expected in surveys of flocked birds because flock size is imperfectly estimated by aerial crews in low-level visual surveys. For Brant, the photo-based estimates are about twice as large as the estimates from the traditional visual surveys. For Cackling Geese, the photo estimates are 30% larger than those from the visual surveys. This was expected because even highly experienced human observers typically underestimate the number of birds present and cannot distinguish all birds by species without some measure of error. The underestimation is more pronounced for larger flocks than for smaller flocks; and Brant typically occur in larger flocks than Cackling Geese at Izembek Lagoon. It therefore makes sense that the visual survey might underestimate Brant numbers to a greater extent than Cackling Geese, explaining the discrepancies with the photo survey. Similarly, photo surveys of other species in other areas would probably reveal that visual surveys have underestimated the numbers present. The severity of the underestimation would depend on how easy it is to visually detect and identify the species of interest (depending on the plumage of the species and whether it is often obscured behind vegetation) as well as the flock sizes in which the species occurs. The photographic methods described here have high potential to be used to document accurate population estimates of waterfowl during other periods of the annual cycle when birds are concentrated including winter, molt, and at breeding colonies.
Thousands of geese gather at Izembek Lagoon in southwestern Alaska every fall where they “stage”, meaning that they rest and eat in preparation for migration to lower latitudes. Izembek Lagoon is especially important for Pacific brant geese, as the entire Pacific Flyway population is thought to use the lagoon in fall. This provides an opportunity to efficiently survey the population to track annual abundance and productivity. New aerial imagery and automated counting methods are helping to optimize surveys and improve population estimates.
Return to USGS Alaska Q&A Series
A primary goal of the Migratory Bird Management Office of the U.S. Fish and Wildlife Service (USFWS) is to understand the annual abundance and trends of migratory bird populations across North America. Each year, the USFWS works with numerous partners to survey and estimate abundance. The USFWS recently collaborated with the USGS Alaska Science Center and developed new methods to count geese at Izembek Lagoon in southwestern Alaska. In this Q&A, we talk with Julian Fischer (USFWS Migratory Bird Management Office in Anchorage, Alaska) and Emily Weiser (USGS Alaska Science Center) about these new image and automated counting methods and what they might mean for the future of aerial counting and monitoring bird populations.
Q: How have waterfowl surveys at Izembek Lagoon been conducted in the past?
Julian: Waterfowl at Izembek Lagoon have been surveyed from small airplanes since 1976. This fall staging survey involves a pilot-biologist, who flies the plane and counts the birds on one side of the plane, and a second observer who counts birds on the other side of the plane. The two surveyors identify birds and estimate the numbers that they see out the window as they fly. This is challenging during the fall staging period because the birds are often in mixed-species flocks of hundreds or even thousands of birds, so it’s difficult to accurately estimate the number present. To see the birds well enough to identify and count them, the pilot maneuvers the plane low over the water (45 m, or 150 ft, above the water), which can cause birds to flush in front of the aircraft, making it even more difficult to count them.
Q: How does the new photo survey work?
Emily: The photo survey is flown with the same type of small aircraft but uses cameras instead of human observers. Two cameras mounted in the belly of the aircraft take photos of the lagoon as the plane flies over. Software controls the cameras and automatically triggers them as the plane flies through lines of pre-set photo points. The cameras have 200 mm lenses and high-resolution sensors, so zooming in on a photo provides a clear image of the birds on the water below, even at high altitudes.
Disclaimer statement: Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. Dennis Marks USFWS photographer.
Q: How is the photo survey an improvement over the past method of visually counting birds out the window of the plane?
Julian: The photo survey is an improvement in three ways. First, the photo transects represent a known sample of the area of the lagoon. This repeatable design can be replicated annually and provides data analysts the ability to estimate total numbers of birds present and calculate the precision of that estimate. Second, because the cameras capture images at high resolution, birds can be identified and counted even when the aircraft flies at a higher altitude (457 m, or 1500 ft). Flying at higher altitudes prevents birds from moving during the survey, thereby improving accuracy of the survey, as well as minimizing impacts of the survey on the birds. Moreover, the higher altitude provides an increased margin of safety for the aircrew. Third, the photos provide a permanent record of the birds that were present, identified, and counted and provides greater accuracy than visually identifying and counting birds in large flocks.
Q: How are the photos used to develop an estimate of population size?
Emily: Birds in the photos must be identified and counted. And there are a lot of images! We collect 8,000-10,000 images on each flight, so it would be time-consuming to manually count birds in that many photos. We worked with software engineers to develop a computer program to automatically detect, identify, and count birds of each type (Brant, Cackling Geese, Emperor Geese, Gulls, and “Other”, which is mostly ducks). About 95% of photos contain no birds and thus require no further assessment. When the program detects birds in a photo, a person checks the computer identifications and counts to correct any mistakes made by the program. Accuracy of the program is currently being improved, but the fact that it filters out 95% of photos with no birds greatly reduces the workload.
The number of birds of each type in the photos is then translated to the estimated total number on the entire lagoon. The photos sample the lagoon in a regular grid pattern, but miss some of the geese between photos, so the full population is not captured in the photos. However, each photo covers a known area of water, so counts of birds in the photo can be converted into density. Extrapolating that density across the full area of the lagoon provides an estimate of the total number of birds present, very similar to how USGS recently estimated the number of walruses on a coastal haulout using drone imagery. Unlike other waterfowl species, virtually the entire Pacific population of Brant occurs at Izembek Lagoon in fall, thus the survey provides a valuable measure of abundance for this species of management concern. Additionally, the survey counts Cackling geese, the next most abundant species in the area, so that population estimates can be developed.
Q: The new photo survey provided considerably higher estimates of total population size than with the traditional counting method. Do you think applying this method to other areas and other species would also result in higher estimates than determined in the past?
Julian: The greatest improvement in accuracy resulting from photographic methods is expected in surveys of flocked birds because flock size is imperfectly estimated by aerial crews in low-level visual surveys. For Brant, the photo-based estimates are about twice as large as the estimates from the traditional visual surveys. For Cackling Geese, the photo estimates are 30% larger than those from the visual surveys. This was expected because even highly experienced human observers typically underestimate the number of birds present and cannot distinguish all birds by species without some measure of error. The underestimation is more pronounced for larger flocks than for smaller flocks; and Brant typically occur in larger flocks than Cackling Geese at Izembek Lagoon. It therefore makes sense that the visual survey might underestimate Brant numbers to a greater extent than Cackling Geese, explaining the discrepancies with the photo survey. Similarly, photo surveys of other species in other areas would probably reveal that visual surveys have underestimated the numbers present. The severity of the underestimation would depend on how easy it is to visually detect and identify the species of interest (depending on the plumage of the species and whether it is often obscured behind vegetation) as well as the flock sizes in which the species occurs. The photographic methods described here have high potential to be used to document accurate population estimates of waterfowl during other periods of the annual cycle when birds are concentrated including winter, molt, and at breeding colonies.