Testing photo-identification techniques for cataloging

Transcription

1 1 Testing photo-identification techniques for cataloging pygmy blue whales in Sri Lanka Melanie Good 1, Asha de Vos 1, 2, Donald A. Croll 1 1 Centre for Ocean Health, University of California Santa Cruz, 100 Shaffer Road, Santa Cruz, CA 95060, USA 2 The Sri Lankan Blue Whale Project, 131 W.A.D. Ramanayake Mawatha, Colombo 2, Sri Lanka Abstract The pygmy blue whale (Balaenoptera musculus indica) found off the southern coast of Sri Lanka is a poorly studied, non-migratory subspecies of blue whale for which there are no estimates of population abundance. Since 2008, the Sri Lankan Blue Whale Project has worked to build a photographic catalog to estimate population parameters such as population abundance using photo-identification methods. This study tested the efficacy with which inexperienced judges used a classification method to sort photographs from the Sri Lankan blue whale photo-id catalog, as well as their ability to identify the number of individual whales in a set of photographs. The judges classifications were then compared with those made by experienced researchers. Judges agreed 59% with the photographic classifications made by experienced researchers and on average underestimated the number of whales. The adoption of a classification method of photographs is important to facilitate future photo-identification efforts and comparisons between catalogs. Key Words: photo-id, assessment, blue whale, Indian Ocean

2 2 Introduction Over the last few decades, photographic-identification has been the backbone of many long-term research projects on abundance, distribution, and migration patterns of several species of marine mammals, particularly cetaceans (Sears et al. 1990). Photo-identification is a technique used to identify individuals over time by photographing distinctive characteristics, and relies on recognizing individual whales using natural markings. This is beneficial for a number of reasons: 1) the study subjects are not physically handled, 2) appropriate natural markings are permanent and not liable to fall off or wear away like artificial markings, and 3) natural markings do not alter the animal s behavior (Hammond 1986). However, this method requires more time than using artificial marks to recognize individuals and it is easier to make mistakes due to the marks being natural patterns and not numbered or color coded tags (Hammond 1986). For cetaceans of all species, researchers are confident that natural markings do not change dramatically over time, and any slight changes do not interfere with their ability to identify individuals (Hammond, Mizroch & Donovan 1990). It is important for databases of a given species to have unified methods of cataloging photographs, identifying individuals, and collecting data in order to facilitate collaboration and the sharing of information between databases. The IUCN Red List of threatened species lists blue whales (Balaenoptera musculus) as an endangered species. In 2008, it was estimated that over the past three generations, assuming that generations last an average of 31 years, the total blue whale population has decreased by at least 70% (Reilly et al. 2008). After the end of commercial whaling, efforts were made to establish long-term monitoring programs for conservation purposes. These programs rely heavily on photo-identification techniques to mark individuals for estimating population size using

3 3 mark-recapture models. Individuals from populations of blue whales in the Eastern Pacific Ocean (Sears et al. 1990; Gendron & De la Cruz 2012) and the Southern Hemisphere (Olson 2009) are photographed and stored in photo-identification databases, and the populations are routinely monitored. The pygmy blue whale is a poorly studied subspecies of blue whale in the Indian Ocean that was subjected to illegal Soviet pelagic whaling in the 1960s (Mikhalev 1997). The impact of harvesting pygmy blue whales during the 1960s is unclear due to a lack of pre-exploitation and current population estimates; however, a sizeable decrease in catch size indicates that population size was reduced to low levels (Mikhalev, 1996). Although there has been no commercial harvest since 1967, the pygmy blue whale still faces anthropogenic threats in the form of ship strike, incidental catch, pollution, and disturbance caused by whale-watching (de Vos et al., in preparation). Since 2008, the Sri Lankan Blue Whale Project (SLBWP) has built and maintained a photo-identification catalog for this population using images from field seasons spanning January through March and through the collection of photographs from whale-watch tourists. The standard method of photo-identification involves taking photographs of a large portion of the left and right flanks of the animal at a perpendicular angle. The dorsal fin is used as a reference point for the comparison of photographs that have been assigned to pigmentation categories (Sears et al. 1990). The classification method adopted by this project simplifies the process and allows for comparisons between photographs in which a smaller portion of the whale is visible (Gendron & De la Cruz 2012). This method sorts photographs into seven categories by dorsal fin shape and five categories by pigmentation pattern (Gendron & De la Cruz 2012). Inexperienced judges were used to test the subjectivity with which photographs are assigned to

4 4 the dorsal fin and pigmentation categories of this classification method, as well as the accuracy with which matches of individual whales are made. Methods The Sri Lankan Blue Whale photo-identification catalog maintained by the Sri Lankan Blue Whale Project contains photographs taken off southern Sri Lanka between 2008 and 2013 using a Canon 7D camera with a mm lens. When possible, photographs of the dorsal regions were taken from both sides of the whale with the camera at a 90⁰ angle to its flank. For each whale sighting, the best photographs are selected based on angle, clarity, and contrast relative to the other sighting photographs. Sightings with only poor quality photographs are kept on record but are excluded from population parameter estimations (Hammond 1986; Hammond et al. 1990). Judges selected for the following tests had little to no experience with the SLBWP s photographic catalog or photo-identification to reflect the project s reliance on volunteers to maintain the catalog. Two judges had four months of experience working with the catalog, one judged was an artist who illustrated blue whales from the catalog, and three additional judges had no prior exposure to the catalog. Photographs of dorsal fins and associated pigmentation patterns were classified according to the seven dorsal fin shape categories and five pigmentation categories described in Gendron and De la Cruz (2012). An additional Undefined classification included in both dorsal fin shape and pigmentation categories accounts for photographs that are too low in quality to be confidently classified. To test the subjectivity of this classification method, we selected a total of 33 photographs of similar quality from the SLBWP s photographic catalog and asked six judges to independently assign categories to each photograph according to the classification method

5 5 using a reference sheet for guidance (Figures 3 and 4; Gendron and De la Cruz, 2012). The selected photographs represented all 5 pigmentation categories and 5 of the 7 dorsal fin shape categories. No undefined dorsal fins were included in the photographs selected for the test, however judges were given the option to assign undefined dorsal fins if they felt image quality or camera angle biased their decision. We compared the results using a chi-squared test to classifications made by the collaboration of the two researchers most experienced with blue whale photo-identification. As we found significant differences, we used a subdivision of the test to identify the most subjective categories by removing the category that contributed the most to the previously calculated chi-square. To test the accuracy with which judges discerned the number of whales in a set of photographs, we selected 22 flank photographs taken from the right side of the animals. All photographs were of similar quality and depicted whales with triangular, falcate-triangular, and falcate dorsal fin shapes spanning the various pigmentation categories. The same two researchers from the previous test collaboratively identified 19 whales from the set; three whales were represented by two photographs each while all others were represented by a single photograph. The six judges from the previous test were asked to independently determine the total number of whales represented in the set of photographs. Test efforts were limited to 2 hour sessions to minimize the error due to fatigue (Sears et al. 1990). These results were then resampled (n = 6) 10,000 times using the bootstrap method. We compared the true number of whales against the sample mean using a 95% confidence interval generated through bootstrap analysis to determine whether judges with little photo-identification experience can identify the correct number of whales in a set of photographs.

6 6 Results Applying the classification method to whale sightings with dorsal fin photographs of sufficient quality in the SLBWP s photographic catalog showed an uneven distribution among the dorsal fin categories (Table 1). The falcate dorsal fin was the most abundant, comprising 28.0% of the 239 sightings catalogued. The smallest dorsal fin category, mutilated fins, only includes 0.8% sightings. None of the sightings currently documented have been classified as hooked fins. For the pigmentation pattern categories, the most commonly observed was the striped pattern (32.6%) and the least observed was the balanced pattern (4.6%). Table 1 Number of blue whale sightings classified per dorsal fin and pigmentation categories within the Sri Lankan blue whale photo-identification catalog. Dorsal fin categories Pigmentation patterns Triangular Falcatetriangular Falcate Hooked Straight Marked Mutilated Undefined Total (%) Light Lightmottled Striped Balanced Dark Undefined Total (%) There was no significant difference between dorsal fin classifications made by the judges and those assigned by the experienced researchers (χ 2 = 8.23, d.f. = 4, p = 0.083). The triangular category contributed the most to the chi-square value, and after it was removed in a subdivision of the test no significant difference was found. The most ambiguous category was straight fins with only 20% of judge s responses matching the researchers. Marked fins were the least ambiguous (83.3% matched; Table 2). Undefined dorsal fins made up 10.6% of the responses but could not be included in the chi-square calculation.

7 7 As in Gendron and De la Cruz (2012), the balanced and light-mottled pigmentation patterns were the most ambiguous categories with a significant difference between judges and researchers (χ 2 = 17.86, d.f. = 4, p = ). Balanced pigmentation contributed the most to the chi-square value, but the significance was lost only when balanced and light-mottled were both removed in the subdivision of the test. No judge assigned an undefined pigmentation to a whale. Matches between judges and the researchers ranged from 29.2 to 91.7% (Table 3), with lightmottled as the most ambiguous and light as the least. Table 2 Percentages of matches between dorsal fin category assignments made by judges compared to the researchers classification of 33 photographs from the Sri Lankan blue whale catalog, and the percentages of non-matching classifications. Dorsal fin category Matches (%) Classifications of mismatched photographs (%) Triangular 50.0 Falcate-triangular (29.6) Straight, Undefined (9.3) Falcate-Triangular 47.2 Straight (20.8) Falcate (16.7) Falcate 63.3 Falcate-triangular (23.3) Straight (10.0) Straight 20.0 Falcate-triangular (43.3) Undefined (30.0) Marked 83.3 Falcate-triangular (16.7) Table 3 Percentages of matches between pigmentation category assignments made by judges compared to the researchers classification of 33 photographs from the Sri Lanka blue whale catalog, and the percentages of non-matching classifications. Pigmentation category Matches (%) Classifications of mismatched photographs (%) Light 91.7 Dark (8.3) Light-mottled 29.2 Balanced (50.0) Light (12.5) Striped 58.3 Dark (22.9) Balanced (10.4) Balanced 70.0 Striped (23.3) Dark (6.7) Dark 76.2 Balanced (10.7) Striped (6.0) In our second test, the average number of whales identified from the set of 22 photographs was The bootstrapped median was 17.5 whales with a 95% confidence interval of (16.17, 18.5). The probability of correctly identifying 19 whales lies outside the confidence interval. Therefore, if the responses of the six judges are representative of the average volunteer, people tend to underestimate the number of blue whales by identifying false matches between

8 8 photographs. The distribution of responses was skewed due to the number of photographs available for each whale, making it unlikely that judges would overestimate the number of whales represented. If all whales were equally represented in the set, the average number of whales determined by the judges may fall closer to the true number. Discussion As the Sri Lankan Blue Whale Project s photo-identification catalog grows it will become increasingly important to use a classification method that reduces the number of photographs compared and time needed for photographic matching. The classification method tested in this study accomplishes this by dividing the catalog into 48 combinations of dorsal fin and pigmentation categories. A photograph is first compared to the most representative dorsal fin and pigmentation categories, a process described in detail in Gendron and De la Cruz (2012). This avoids comparisons with photographs that have a low likelihood of matching. Like other classification methods, the categories are based on fin shapes and pigmentation patterns that change progressively which may lead to uncertainty when assigning categories. However, Gendron and De la Cruz (2012) have found that with time and experience classification becomes less ambiguous. Our test detected no significant difference between the dorsal fin categories assigned by the judges and those assigned by the researchers, however, on average the judges only matched the researchers assignments 52.8% of the time. This may be because Sri Lankan blue whales appear not to possess the exaggerated falcate and hooked dorsal fins found in the Gulf of Mexico population for which this classification method was developed. For many of the judges, this test

9 9 was their first time looking at blue whales and attempting to classify dorsal fin shapes with only a reference sheets for guidance. Creating reference sheets with more detailed descriptions and photographs from the SLBWP s photographic catalog should also decrease new volunteers initial uncertainties. We expect greater consensus as new volunteers gain more experience and greater familiarity with the whales in the SLBWP s photographic catalog. Although no undefined dorsal fins were included in the test, 10.6% of the judges responses were listed as such. This suggests judges will avoid classifying difficult to judge dorsal fins, thereby reducing the number of incorrectly categorized photographs. When assigning pigmentation categories we found that some whales did not fit any of the listed categories. In those cases the researchers chose the category that was the best match, however this may indicate that the pigmentation categories suggested by Gendron and De la Cruz may not be transferable to the Sri Lankan population. In some cases, whales appeared to have different pigmentation patterns anterior and posterior to the dorsal fin, which makes classifying the overall pattern difficult with the current categories. This problem arose most often with whales considered to have characteristics of striped, balanced, or dark pigmentation patterns. These whales would be best described by a dark-mottled category which does not exist in the Gendraon and De la Cruz classification method (Fig. 1). Figure 1. Flank photograph of a blue whale from the SLBWP s catalog that exemplifies dark-mottled pigmentation. There is no uniform distribution of light and dark pigment or a distinct dark band along the back as described by the balanced and striped categories.

10 10 Judges underestimated the number of individual whales in the 22 photographs presented. All six judges correctly identified the three whales represented by two photographs each, but only two judges assigned the correct number of whales. Out of all the photographs of different whales incorrectly matched together, only two judges made the same error. The two whales incorrectly matched in this case had similar pigmentation patterns, but different dorsal fins. It is possible that these pictures may be of the same whale whose dorsal fin was damaged between the times the photographs were taken, but without a higher quality photograph we cannot claim the photographs are of the same whale. Because the standard photo-identification process requires photographic matches to be confirmed by at least two individuals (Agler, Beard & Bowman 1990; Sears et al. 1990; Gendron & De la Cruz 2012), it is unlikely for photographs to be incorrectly matched in a photo-identification database even if two relatively inexperienced volunteers work together. Despite the relatively high levels of disagreement between inexperienced judges and experienced researchers, it is important to adopt a classification method to save time in future photo-identification efforts. Classification methods also facilitate comparisons between other photographic catalogs (Gendron & De la Cruz 2012). Inaccuracies in dorsal fin and pigmentation categorization will be corrected during periodic reviews of the SLBWP s photographic catalog by experienced researchers. New volunteers are asked to sketch a number of whales from the photographic catalog in order to increase their familiarity with the animals. This in turn should also increase their ability to accurately categorize photographs and identify matches. Now that individual blue whales within the Sri Lankan population are being identified and added to the SLBWP s photo-identification database, the next goal for the project is to develop a grading system of photograph quality. By only using photographs above a quality

11 11 threshold, we can reduce recapture bias when estimating population abundance (Hammond 1986). Additionally, the SLBWP plans to encourage tourists to contribute photographs to the Sri Lankan blue whale catalog. These initial efforts are important steps towards long-term monitoring and management of a unique subspecies of blue whale we still know very little about. Acknowledgements We thank Pete Raimondi and Mark Carr for their help designing this study and their continued support. Thanks to Kelly Newton for guidance in building the photo-identification database. Finally we thank the Costal Conservation Action Lab for the use of their lab space and resources. Literature Cited Agler, B., Beard, J. & Bowman, R. (1990) Fin whale (Balaenoptera physalus) photographic identification: methodology and preliminary results from the western North Atlantic. Report of the International Whaling Commission Special Issue, 12, Gendron, D. & De la Cruz, A.U. (2012) A new classification method to simplify blue whale photo- identification technique. North, 12, Hammond, P.S. (1986) Estimating the size of naturally marked whale populations using capturerecapture techniques. Reports of the International Whaling Comission Special Issue, 8, Hammond, P.S., Mizroch, S.A. & Donovan, G.P. (1990) Individual recognition of cetaceans: use of photo-identification and other techniques to estimate population parameters. Report - International Whaling Commission Special Issue, 12, Mikhalev, Y.A., (1996). Pygmy blue whales of the northern-western Indian Ocean. Paper SC/48/SH30 presented to IWC Scientific Committee, June 1996, Aberdeen, UK (unpublished). 30pp. [Paper available from the Office of this Journal]

12 12 Mikhalev, Y.A. (1997) Additional information about the catches of Soviet whaling fleet Sovetskaya Ukraina. Report of the International Whaling Commission, 47, Olson, P. (2009) Blue whale photo-identification from IWC IDCR/SOWER surveys. IWC Sc/61/Sh19, Reilly, S.B., Bannister, J.L., Best, P.B., Brown, M., Brownell Jr., R.L., Butterworth, D.S., Clapham, P.J., Cooke, J., Donovan, G.P., Urbán, J.& & Zerbini, A.N. (2008) Balaenoptera musculus. The IUCN Red List of Threatened Species. Version < Sears, R., Williamson, M.J., Wenzel, F.W., Bérubé, M., Gendron, D. & Jones, P. (1990) Photographic identification of the Blue Whale (Balaenoptera musculus) in the Gulf of St. Lawrence, Canada. Report of the International Whaling Commission Special Issue, 12,