Development of a noninvasive photograph-based method for the evaluation of body condition in free-ranging brown bears

Study area

This study was conducted in the Shiretoko Peninsula (43°50′–44°20′N, 144°45′–−145°20′E), Hokkaido, Japan (Fig. 1). This peninsula has one of the largest brown bear populations worldwide (Hokkaido Government, 2017), and an area from the middle to the tip of the peninsula has been designated as Shiretoko National Park and a UNESCO World Natural Heritage Site. During 1998–2017, we collected body masses and morphometrics from brown bears captured for research purposes, killed for nuisance control, or harvested from the peninsula, including the towns of Shari and Rausu (Fig. 1). In addition, a focal survey was conducted in the Rusha area (44°12′N, 145°12′E; Fig. 1), a special wildlife protection area. Public access is not allowed without permission and there is no human residence except for one fishermen’s settlement. Because the fishermen have not excluded bears from the settlement area in the last few decades, the bears have become habituated to the existence of humans, which enables direct observation at close range. Long-term monitoring of identifiable bears has been conducted in this region since 2006 (Shimozuru et al., 2017). We collected body masses, morphometrics, and photographs of female bears in the Rusha area during 2014–2016. Field experiments were approved by Hokkaido Regional Environment Office and Kushiro Nature Conservation Office (Permit Number: 1606091 and 1705182).

Bear capture and measurements

Bears were sampled each year during 1998–2017 between April and November. Most samples were obtained from bears killed for nuisance control or harvested, and some were obtained from bears captured for research purposes. The variables recorded for each bear included an identification code, date of measurement, location, body mass (kg), and straight-line body length (cm) (Data S1). Body mass was measured with calibrated hanging spring scales. Body length was measured with a non-stretchable tape measure as the straight-line distance from the tip of the nose to the end of the last tail vertebra while the bear was aligned laterally. In addition, we measured torso height (cm) as the distance from the lowest point of the abdomen to the spine in females ≥5 years old during 2014–2017. We also collected tissue (e.g., muscle and liver) from killed bears and blood and hair samples from captured bears for DNA extraction, which allowed us to identify individuals and their sex (Shimozuru et al., 2017; Shirane et al., 2018). Among 503 killed or captured individuals, 22 individuals were sampled more than once during our study period due to repeated capture or killing after capture; we used only the measurement taken at the greatest age in the following analyses.

We estimated the age in years of most bears captured or killed by counting the cementum annuli of the teeth (Yoneda, 1976). For some individuals, we could not determine the exact age due to many cementum-layers developed in old individuals or poor quality of teeth samples. Individuals whose age range could only be estimated were excluded from the growth curve analyses but were included for BCI and subsequent analyses if the growth curve results (detailed below) allowed their classification into an age class. For example, females ≥5 years old were excluded from growth curve analyses but were used as adults for subsequent analyses, whereas males ≥5 years old were excluded from all analyses.

All bears were captured live in accordance with the Guidelines for Animal Care and Use of Hokkaido University and all procedures were approved by the Animal Care and Use Committee of the Graduate School of Veterinary Medicine, Hokkaido University (Permit Number: 1152 and 15009). The protocols for capture received annual approval from the Ministry of the Environment, Japan, and the Hokkaido Government through research permit applications.

Growth curve of body length

To estimate the age at which the growth of body length was completed, growth pattern in body length was examined using a von Bertalanffy curve as previously described in bears (Kingsley, Nagy & Reynolds, 1988; Derocher & Stirling, 1998; Derocher & Wiig, 2002; Bartareau, Cluff & Larter, 2011). The von Bertalanffy size-at-age equation was used in the form A_t = A_∞(1 −e^−K(t−T)), where A_t is body length (in cm) at age t, A_∞ is asymptotic body length (in cm), K is a size growth rate constant (year⁻¹ ), and T is a fitting constant (extrapolated age at zero size; in years). We conducted F tests to determine whether the parameters of the von Bertalanffy growth equation differed significantly by sex. We conducted analyses using FSA package version 0.8.30 (Ogle, Wheeler & Dinno, 2020) and nlstools package version 1.0-2 (Baty et al., 2015) in R (R Core Team, 2019). According to the age reaching 95% of the asymptotic body lengths obtained from this analysis (detailed below), bears were assigned to three age classes for each sex: cubs (0–1 year old), subadults (age 1–4 years and 1–7 years for females and males, respectively), and adults (age ≥5 years and ≥8 years for females and males, respectively).

BCI of killed or captured bears

We calculated BCI as previously described in Cattet et al. (2002). Specifically, body mass and length values were transformed to natural logarithms and a least-squares linear regression analysis was conducted to describe the relationship between the ln-transformed values. The standardized residuals of this linear regression were used as BCI. In addition, as a preliminary experiment for the evaluation of body condition using photographs, we calculated the ratio of torso height to body length (TH:BL) using actual measurement data.

Statistical methods.—To determine if the BCI was independent of body size, we investigated the correlation between BCI and body length that is an indicator of body size (Mahoney, Virgl & Mawhinney, 2001; Cattet et al., 2002). BCI was compared among seasons and age-sex classes using two-way analysis of variance (ANOVA). We used Tukey multiple comparisons (Tukey, 1977) to evaluate differences between the mean values of each comparison. Based on major changes in diet (Ohdachi & Aoi, 1987), we divided the sampling period into three seasons: spring (April to June; main diet of grass), summer (July and August; main diet of grass and ants), and autumn (September to November; main diet of berries and acorns). In addition, we linearly regressed BCI on the TH:BL of the same individuals and calculated the correlation coefficient. We also used correlation analysis between TH:BL and body length to investigate the effects of body size. We conducted all statistical analyses in R (R Core Team, 2019).

Obtaining and filtering of photographs

Periodic surveys (≥1 day/2 weeks) have been conducted since 2011 in the Rusha area, mainly for monitoring the reproductive status of identifiable females (Shimozuru et al., 2017). This area is a narrow estuarine coast stretching south to north for approximately 3 km. Field teams patrolled the area by car and waited for bears to emerge from the vegetation on the mountainside. When bears appeared, we followed individuals, maintaining a distance of about 20–100 m. Individual bears were identified by field staff according to their appearance as described in Shimozuru et al. (2017), and close-up photographs were taken from multiple angles with a digital, single-lens reflex camera (Nikon D800, NIKON Co., Tokyo, Japan; or Canon EOS 5D, Canon Inc., Tokyo, Japan).

For each survey in the Rusha area, lateral photographs of each individual bear were selected and graded based on several attributes: camera focus, camera tilt (vertical), camera angle (horizontal), body/torso height measurability, and body/torso length measurability for photography; and degree of body arch (vertical), straightness of body (horizontal), degree of neck flexing (vertical), and degree of neck bending (horizontal) for bear posture (Table S1, Fig. S1). Each photograph was given a score of 1 (good quality), 2 (medium quality), or 3 (poor quality) for each attribute. Photographs that were given a score of 3 for any attribute were removed from further analyses.

Morphometric measurements from photographs

We used ImageJ version 1.52a (Schneider, Rasband & Eliceiri, 2012) to extract morphometric measurements from lateral photographs of bears. We first adjusted the angle of the photographs according to the ground surface, then measured the torso height in pixels (TH) as the distance perpendicular to the ground from the lowest point of the abdomen to the highest point of the waist (Fig. 2). Length measurements (in pixels) included the following four methods: the horizontal straight-line body length (HBL, Fig. 2) was the straight-line distance from the tip of the nose to the base of the tail; the Euclidean straight-line body length (EBL, Fig. 2) was the Euclidean distance from the base of the tail to tip of the nose; the polygonal-line body length (PBL, Fig. 2) was the sum of the distance from the base of the tail to the highest part of the shoulder parallel to the ground surface, from that point to the base of the ear, and from that point to the tip of the nose; and the horizontal straight-line torso length (HTL, Fig. 2) was the straight-line distance from the base of the tail to the highest part of the shoulder parallel to the ground. For all measurements, any area that could be clearly judged to be only fur was excluded from the measurement range.

Precision of measurements from photographs

To examine the precision of each photograph-based measurement method and the effects of bear posture, we used photographs of one bear (bear ID: HC) that was monitored routinely in the Rusha area during 2016–2018. We classified photographs according to bear posture (Table S1 and Fig. S1): photographs that had a score of 1 for all attributes were assigned to “Good”, those with a score of 2 for body straightness only were assigned to “BS”, those with a score of 2 for neck flexing only were assigned to “NF”, and those with a score of 2 for neck lateral bending only were assigned to “NB”. Photographs that were not assigned to any category were excluded from these analyses.

First, to determine the number of measurements sufficient to reduce measurement error, we assessed measurement precision within photographs by repeatedly measuring (50 times) the body morphometrics from the best photograph taken on September 25, 2017, and assigned to the “Good” category. From these measurements, the coefficients of variation (CVs) for TH, HBL, EBL, PBL, HTL, and the ratio of TH to body/torso length were calculated. In addition, by considering the standard deviation obtained from the 50 measurements as the population standard deviation, we calculated the measurement error at a given number of measurements. We ultimately adopted the minimum number of measurements for which the measurement error had a value that did not affect the second decimal place (i.e., <0.0025). In the following analyses, TH and body/torso length were measured three times, and the TH:body/torso length ratio was calculated from the respective average values according to our results (detailed below).

Second, we assessed measurement precision between bear postures (differences between repeated measures of the same individual taken from photographs with different postures) by taking measurements from photographs in different posture categories. To eliminate the effects of seasonal changes in body condition, we restricted these analyses to photographs taken September 24–26, 2017. The TH:body/torso length ratio was compared among the posture categories for each measurement method with one-way ANOVA. We used Tukey multiple comparisons (Tukey, 1977) to evaluate differences between the mean values of different categories. Then we calculated the CV of each method using all of the photographs applicable to the method to evaluate the measurement precision of each method. We compared CVs among the four methods using an asymptotic test (Feltz & Miller, 1996). From these results, we adopted the method that could be applied to photographs of the most diverse postures while maintaining a sufficiently high measurement precision between photographs (CV <5%). In accordance with these results (detailed below), we used TH:HTL as an indicator of body condition in the following analyses.

Third, to examine whether TH:HTL reflected seasonal changes in body condition, we used photographs taken between late June and early October during 2016–2018. For each half-month, the best two or more photographs were selected and the median TH:HTL obtained from these photographs was considered the evaluation value for that half-month. We compared TH:HTL among half-months using one-way ANOVA and used Tukey multiple comparisons (Tukey, 1977) to evaluate differences between the mean values of each half-month. We conducted statistical analyses using Microsoft Excel® (Microsoft Corporation, 2016) or R (R Core Team, 2019).

Accuracy of measurements from photographs

We examined the accuracy of photograph-based measurement methods using actual measurement data for seven females (≥5 years old) captured in the Rusha area (bear IDs: BE, DR, GI, KR, LI, RI, and WK). We collected photographs of these individuals from within 3 days before and after the days the individuals were captured. After filtering the photographs, we measured TH and HTL and calculated the TH:HTL ratio using two or more of the best photographs. We also calculated BCI using the body mass and length measured at the time of capture.

Statistical methods.—We linearly regressed BCI on the TH:HTL ratio and calculated the correlation coefficient. We conducted statistical analyses using Microsoft Excel® (Microsoft Corporation, 2016).

Body length growth curves

von Bertalanffy curves were successfully fitted to body length data for the 432 individuals with age (in years) assignments (Fig. 3, Table 1, Data S1). The growth curves differed significantly by sex (F_3,426 = 76.63, p < 0.001). Females had achieved 95% of their asymptotic body length at 4.6 years of age, whereas males took 7.6 years to reach the same proportion. In accordance with these results, 476 individuals, including those with known age ranges, were classified into age classes and used in the subsequent analyses: 8 females and 19 males were cubs, 105 females (1–4 years) and 211 males (1–7 years) were subadults, and 92 females ≥5 years old and 41 males ≥8 years old were adults.

BCI of killed or captured bears

Natural logarithmic transformation of the body mass and length data resulted in a linear relationship between mass and length as follows: ln body mass = 3.04 • ln body length − 10.41 (R² = 0.94, residual standard deviation = 0.19, Fig. 4, Data S1). To facilitate estimation of BCI for brown bears, we developed the following model: BCI = (ln body mass − 3.04 ln body length + 10.41)/0.19. There was no correlation between body length and BCI (r = 0.037, p = 0.39), which indicates that BCI was independent of body size (Fig. S2).

An ANOVA of BCI showed that BCI varied significantly by season (F_2,459 = 13.26, p < 0.001; Table 2, Fig. 5), with bears sampled in spring and summer having lower BCI than bears sampled in autumn (both p < 0.001). Differences among age-sex classes were also significant (F_5,459 = 4.20, p < 0.001): Adult males showed higher BCI than adult females (p = 0.002), subadult females (p < 0.001), and subadult males (p = 0.003), whereas BCI did not differ among other age-sex classes (p = 0.35 − 0.99). The interaction between season and age-sex class was not significant (F_9,459 = 0.46, p = 0.90).

We obtained measurements of torso height from 23 adult females. A positive correlation was found between the TH:BL ratio and BCI (r = 0.81, p < 0.001; Fig. 6, Data S1). There was no correlation between body length and TH:BL (r =  − 0.068, p = 0.73), which indicates that TH:BL was independent of body size (Fig. S3).

Precision of measurements from photographs

A total of 220 photographs of the same bear (bear ID: HC) were taken September 24–26, 2017. After filtering based on photographic conditions and the body arch of the bear (Table S1 and Fig. S1), 101 photographs remained. Of these photographs, 15 were assigned to “Good”, 9 to “BS”, 10 to “NF”, and 9 to “NB”.

Based on 50 repeat measurements of the best photograph in the “Good” category, the CV in measurement error within photographs was estimated to be 0.29% for torso height and 0.27%, 0.29%, 0.26%, and 0.45% for HBL, EBL, PBL, and HTL, respectively. For all measurement methods, we reduced the measurement error of the ratio of height to length to less than ±0.0025 by measuring height and body/torso length ≥3 times (Table 3).

The torso height:body/torso length ratio differed among the posture categories for all measurement methods (p < 0.001 for TH:HBL and TH:EBL, p = 0.005 for TH:PBL, and p = 0.002 for TH:HTL, Table 4, Data S2). TH:HBL and TH:EBL obtained from photographs in the “BS”, “NF”, and “NB” categories differed significantly from the results obtained from photographs in the “Good” category (Table 4). TH:PBL measured using “BS” and “NB” photographs were different from those of “Good” photographs (Table 4). TH:HTL differed from “Good” photographs only when we used “BS” photographs (Table 4). When we used all photographs in each category that did not differ from “Good” for each method, the CV was <5% for all methods and did not differ among methods (p = 0.067): 2.47% in TH:HBL (photo n = 15), 2.19% in TH:EBL (n = 15), 3.18% in TH:PBL (n = 25), and 3.93% in TH:HTL (n = 34). Given these results, TH:HTL was adopted as the measurement method with both the largest number of applicable photographs and a CV <5% (i.e., high measurement precision between photographs).

By calculating TH:HTL using photographs of the same bear (bear ID: HC) taken from late June to early October during 2016–2018, we determined that TH:HTL reached its lowest in late August (0.567 ± 0.012; mean ± SE) and its highest in early October (0.714 ± 0.015, Fig. 7). TH:HTL varied significantly among half-months (F_7,16 = 18.41, p < 0.001) and was lower in late August than in late June (p = 0.013), early July (p = 0.007), late July (p = 0.012), early September (p < 0.001), late September (p < 0.001), or early October (p < 0.001).

Accuracy of measurements from photographs

We captured seven adult females in the Rusha area during 2014–2016 and took photographs of each individual within 3 days before and after each capture date (Table S2). There was a positive correlation between BCI calculated from actual morphometric measurements and TH:HTL calculated from photographs (r = 0.77, p = 0.042; Fig. 8).