link to AFSC home page

link to AFSC home page link to NMFS home page link to NOAA home page

Polar Ecosystems Program

Accounting for Species Identification Errors from Aerial Surveys of Ice-associated Seals

Research Reports
Oct-Nov-Dec 2012
Contents
Feature
ABL Reports
FMA Reports
NMML Reports
RACE Reports
REFM Reports
Milestones
Complete Rpt. (pdf)
Quarterly Index
Quarterly Home
ice seal
""

Figure 5. The characteristic bands on the coats of ribbon seals are not necessarily clearly visible in an aerial image. The images on the top right and bottom right were taken with a Canon 1DS Mark III digital camera fitted with a Zeiss 100-mm lens from 1,000 ft during a 2012 line-transect survey in the Bering Sea. In the top right image, a species identification expert would likely rely on the clearly visible bands to conclude that the seal is certainly a ribbon seal. In the bottom right image, a species identification expert would likely rely on a combination of body shape, head size, flipper size and shape, and what could be one or more bands to conclude that the seal is probably a ribbon seal.

In April-May 2012, researchers from the National Marine Mammal Laboratory’s (NMML) Polar Ecosystems Program (PEP) conducted abundance and distribution surveys for the four species of ice-associated seals (bearded, spotted, ribbon, and ringed seals) that occur and breed in the Bering Sea. Advanced thermal-imaging technology was used to detect the warm bodies of seals against the background of the cold sea ice, and high-resolution digital images will be used to identify the species of seals detected by the thermal imagers. Ultimately, this project—which is partly funded by the Bureau of Ocean Energy Management and includes a collaboration with Russian colleagues—will provide the first comprehensive estimates of abundance for the four species of ice-associated seals found in the Bering and Okhotsk Seas.

The different characteristics that distinguish these ice-associated seal species can sometimes be difficult to discern from imagery taken at the survey altitude of 1,000 feet. For example, the characteristic bands on the coats of ribbon seals (Fig. 5) or the “red face” of bearded seals (Fig. 6) will not always be visible in a photo, depending on the orientation of the seal and angle of the image. The identifying characteristics of spotted and ringed seals can be even more difficult to discern from aerial photos. Although typically ignored in population estimates, errors can therefore be common when attempting to identify similar-looking seal species from aerial photographs.

red face bearded sea   aerial image of seal

Figure 6. A red face, which is one of the characteristics associated mostly with bearded seals, is not always present, nor is it necessarily visible in an aerial image. The image on the right was taken with a Canon 1DS Mark III digital camera fitted with a Zeiss 100-mm lens from 1,000 ft during a 2012 line-transect survey in the Bering Sea. In this image, a species identification expert would likely rely on the combination of body shape, head size, front-flipper size and shape, and position on the floe to conclude that the seal is probably or certainly a bearded seal.

We are accounting for species misidentification in our abundance model by estimating misclassification probabilities for species identified in the images. Several PEP seal experts are identifying the species of each seal in each image. To learn more about the factors driving the species identification process, our experts are also recording the specific morphological characteristics that are visible in each image. In addition, experts rank their confidence in each species identification as “positive,” “likely,” or “guess,” where it is assumed that a positive species identification is the correct species. Replicating the species identification process with multiple observers for each seal allows the probabilities of correct (and incorrect) species identification to be estimated and accounted for in our final estimates of population abundance for each species.

Our expert observer trials are illustrated in Table 2, where the true species identity is only assumed to be known with certainty when at least one expert ranks the species identification as positive. By “anchoring” on these positive identifications (e.g., seals 1, 2, and 3 in Table 2), statistical models may then be used to estimate the different species misclassification probabilities and assign species identities to images that are not known with certainty (e.g., seals 4 and 5 in Table 2).

Table 2. Example of species identification trials for three expert observers and four species of ice-associated seals (SD = spotted, RN = ribbon, BD = bearded, RD = ringed, UK = unknown). Observers rank their confidence in a species identification as “positive” (red), “likely” (green), or “guess” (purple).

Seal ID

Observer 1

Observer 2

Observer 3

True Species ID

1

RN

RN

RN

RN

2

SD

SD

RD

SD

3

BD

BD

RN

BD

4

SD

SD

SD

?

5

RN

UK

UK

?

Our expert observers are currently conducting species identification trials for more than 600 seals detected by our thermal imagers. Once these trials are completed, not only will we have a better understanding of the frequency of errors when identifying ice-associated seal species from aerial photos, but we also will be able to properly adjust our population estimates for each species accordingly. In addition, we will gain a better understanding of the specific morphological characteristics most commonly used to identify each species from aerial transect surveys.

By Brett McClintock, Erin Moreland, and Peter Boveng

<<<previous

 


            Home | FOIA | Privacy | USA.gov | Accessibility      doc logo