The authors have declared that no competing interests exist.
The Rajah Brooke's Birdwing,
The Rajah Brooke’s Birdwing,
Despite being protected under Act 76 [
Although there has not been any prior systematic population monitoring, there is a general public perception that numbers of this birdwing have declined over the years. Reasons for a decline in the population could be over-collecting, but more so logging and conversion of forests for agriculture and human habitation. A long-term monitoring program is needed to help understand seasonal variations and longer term population changes. The information gained from such a monitoring program would support conservation efforts [
Before a monitoring program can be implemented, it is essential to develop a monitoring method suitable for the target species. Detectability is a major concern [
In this study, we test the use of counts of puddling Rajah Brooke’s Birdwings as a method of monitoring a local population using an index of abundance, and discuss its suitability and limitations. We determine for the test site the optimum duration of a monitoring session, the best times of day and the effects of weather. In addition, the reliability of the method was tested against a modified transect count. Monthly data are presented for a two-year monitoring period.
The study was conducted in one of the few remaining sites where relatively large numbers of male Rajah Brooke’s Birdwings consistently puddle in single-species groups. The puddling occurs just above a river (Sungai Geroh) at the forest fringe beside the very small indigenous-community village of Ulu Geroh (4° 26' 24.9" N, 101° 15' 01.8" E) in the Malaysian state of Perak. The village is planted with fruit trees and flowering bushes interspersed among small wooden village houses, and is closely flanked by the extensive Bukit Kinta Forest Reserve. Preliminary surveys identified two main puddling sites about 30 m apart that formed the basis for the study (
Visual field counts are impractical and unreliable when large numbers of butterflies puddle. Therefore, we used digital photography to aid in obtaining counts. Two cameras, a Canon EOS 5D Mark II with an EF 70–200 mm f / 2.8 L USM lens, and a Nikon Coolpix P5100, were mounted on tripods 5–10 m from the two main puddling sites and used to capture images of puddling birdwings. The EOS 5D Mark II, which captures images of higher resolution, was used for the larger group. The distance of the cameras and their orientation were adjusted each day and month to frame all the birdwings in each large puddling group and to obtain the best view. Two operators communicated using walkie-talkies and triggered the cameras simultaneously at the different puddles. The numbers of birdwings on the small transient puddle were small and were counted visually. Counts of birdwings in the images were made with the aid of the image analysis software, ImageJ 1.44p.
A comparison of the technique was made with transect counts that were conducted near the puddling sites. Male birdwings in flight patrolling the river were counted along a 180×20 m belt transect by the same observer walking through once at a slow and steady pace for the first 15–20 minutes of each hour from 0900–1700 hrs. The count zone was the full 20-meter width of the transect, up to the height of the tree canopies, and with a forward view as far as the eye could see. The location and direction of the transect walk was such that the observer could avoid disturbing the two puddling sites. It began about 20 m downstream of the largest puddle and moved further downstream parallel and adjacent to the river, which was up to about 20 m wide. This stretch of the river edge encompassed by the belt transect was observed to have the most birdwings in flight, and the birdwings were sufficiently large and distinctive to be recognized easily within the belt. Back and forth flights of what was recognizably the same individual birdwing circling within or across the belt transect were counted as a single sighting. Transect and puddling counts were obtained for three high (January to March 2010) and three low (September, October and December 2010) population months for three days each month but, due to logistical limitations, two days in January 2010. Three photographs were taken one minute apart at the start of every hour from 0900–1700 hrs. The counts of puddling birdwings in the three images at the start of every hour were averaged to obtain hourly averages. Hourly averages of birdwings on puddles and hourly counts of birdwings in flight were averaged to obtain daily averages which were averaged to obtain the monthly averages. Pearson’s correlation was used to examine the relationship between the monthly average numbers of birdwings puddling and in flight.
To determine the sample requirements for counts of puddling birdwings, a few component studies were undertaken. In the first component study, the optimum number of consecutive images that was needed to obtain a reliable count of puddling birdwings at any given hour was determined from two sets of images taken on a single day, beginning 1100 hrs and 1500 hrs. Each set comprised 30 images taken a minute apart, with no disturbances to the birdwings. The 30-image sets were divided into six consecutive blocks of five images each. The averages for the first image, the first three images, and all five images in each block were taken, and the averages of the six blocks were compared for increasing numbers of images. Their degree of similarity was examined using 95% confidence intervals. Based on the results (described in the results section), the average of three images at the start of each hour was used for all subsequent tests. However, if there was a disruption at the start of an hour, such as a sudden gust of wind that caused the puddling birdwings to fly, the three images were taken 10 minutes later. The remaining three component studies utilized the same 6-month puddling dataset that was used in the comparison between transect counts and puddling counts. The second component study examined the number of monitoring hours needed. To determine whether counts at a single hour would be representative of counts throughout the day, the numbers of birdwings puddling at 1600 hrs were correlated against the average numbers of birdwings puddling from 0900–1600 hrs. Counts over a two- or three-day period each month were first averaged for these time periods. The choice of 1600 hrs was based on an observed peak puddling period from 1400–1600 hrs. A third study analyzed whether the numbers of birdwings puddling at a chosen time of day in a monitoring program can be predicted from counts taken at an alternative time of day. This would be important if, for example, rain, heavy cloud or circumstances prevent the count being made at the chosen monitoring period. Though strictly speaking not intended for bivariate data, regression was used to analyze the relationship between the numbers of birdwings puddling at 1100 hrs and 1600 hrs, with the numbers at 1100 hrs taken as an independent variable in the regression analysis to enable a predicted estimate of the number at 1600 hrs. A fourth component study analyzed the number of monitoring days required for a representative count. This was determined by inspecting the correlation matrix for the average counts obtained on the first, second and third day at 1600 hrs. For the single month with only two monitoring days, the third day’s count was treated as a missing value in the correlation.
Monitoring was carried out for a period of two years from January 2010 to December 2011 using a protocol refined from the above-mentioned studies. Three images were captured a minute apart beginning on the hour from 1400 to 1600 hrs for three days each month, except for the first month and last three months in which two days were used due to logistical limitations. The monitoring days were consecutive except on one month in which the three sampling days occurred over a four-day period due to a day of rain. Environmental variables were recorded or scored 30 minutes into each monitoring hour and averaged for each day. Temperature and relative humidity were recorded with a single reading at each hour using a calibrated datalogger (Blue Gizmo BG-DL-01). Brightness and rainfall were scored hourly on a scale of 1–3 (dull, moderately bright, and very bright) and 0–2 (dry, drizzle, and light rain), respectively. Monitoring was not carried out if there was heavy rain during the actual period of monitoring. In addition to hourly scores, a separate score was given each day to generalize brightness and rainfall from early morning to late evening (i.e. all-day brightness and all-day rainfall), using the same scales as the hourly scores. The data was standardized using Z-scores, and analyzed using a General Linear Model (GLM) in Minitab 17® to determine whether environmental variables affected counts of puddling birdwings. Z-scores were used to obtain cross-comparable regression coefficients for the environmental variables. Daily average counts were the response variable in the GLM, and month was treated as a fixed variable, with days nested within months. Daily environmental measurements and scores were treated as covariates. A full analysis was conducted with all variables, and then stepwise variable selection with an α to enter and α to remove of 0.15 was used to obtain a final reduced model.
Computer-assisted counts of large puddling birdwings in the digital images from both types of cameras, and manual field counts of small numbers of puddling birdwings, were all equally manageable and accurate because of the large size of the birdwings, yielding actual numbers rather than estimates. Many birdwings were observed flying out from the trees near the puddling sites in the early morning and flying back to rest on these trees in the evening. Occasionally, when there were disturbances, such as people passing by, birds landing nearby, a leaf fall, and gusts of wind, puddling birdwings became unsettled and flew away, but remained nearby and eventually came back to rest at the puddle. In one such case, when birds flew near to the puddle, all but five of the 20 puddling birdwings flew away and landed on trees nearby, but they returned to their original numbers at the puddle after 10 minutes (
Data shown are monthly means for 6 monitoring months, with standard error bars.
The first of the four component studies to determine the sample requirements for counts of puddling birdwings assessed the optimum number of consecutive images needed to obtain a reliable count. Very little difference was found between the three sets of six five-minute block averages and their 95% confidence intervals for subsamples of one image or three images, or the full five images (
Data shown are the monthly means for six different months, with standard error bars.
The figures show the averages and 95% confidence intervals of the number of puddling birdwings at two different times of the day for one, three and five images.
Number of images per |
Time of day | |
---|---|---|
1100–1130 hours | 1500–1530 hours | |
1 image | 95.3 | 102.7 |
(83.7–107.0) | (92.1–113.2) | |
3 images | 96.2 | 104.2 |
(84.5–107.8) | (94.8–113.5) | |
5 images | 97.0 | 105.1 |
(86.0–108.0) | (96.4–113.9) |
First day | Second day | |
---|---|---|
0.943 | ||
(P = 0.005) | ||
0.722 | 0.831 | |
(P = 0.168) | (P = 0.081) |
When puddling birdwings were monitored monthly for two years from 1400 to 1600 hrs for two to three consecutive days each month, a high degree of population fluctuation between months was observed. Average numbers ranged from only several puddling birdwings at low season to 344 at the peak. Variation between days in the month was small in comparison to the largest variations between months. The lowest population levels occurred in September of the first year and December of the second year. In the first year of population monitoring, two peaks occurred, the highest in May, followed by a much smaller peak in November (
Data shown are mean daily counts at puddles, the standard error and range for each month.
In the analysis for the effects of environmental variables, there was a significant difference between monitoring months (F22, 36 = 41.28, P < 0.001) when controlling for the effect of the former. Birdwing counts also had a significant, positive association with relative humidity (F1, 36 = 6.84, P = 0.013; standardized regression coefficient, β = 0.270) (
Source | DF | Adj SS | Adj MS | F | P |
Monitoring month | 22 | 60.9967 | 2.7726 | 47.66 | 0.000 |
Relative humidity | 1 | 0.4645 | 0.4645 | 7.98 | 0.007 |
Brightness | 1 | 0.5797 | 0.5797 | 9.96 | 0.003 |
Error | 40 | 2.3270 | 0.0582 | ||
Total | 64 | 64 | |||
S = 0.2412; r2 = 96.36%; r2 (adjusted) = 94.18%; r2 (predicted) = 90.14% | |||||
Regression term | Standardized |
SE of coefficient | T | P | |
Constant | -0.0424 | 0.0306 | -1.39 | 0.173 | |
Relative humidity | 0.2017 | 0.0714 | 2.83 | 0.007 | |
Brightness | 0.1767 | 0.0560 | 3.16 | 0.003 |
The significant, positive correlation between the numbers of birdwings puddling and numbers of birdwings in flight showed that the counts of puddling birdwings corresponded with the more traditional method of a transect count. Transect counts have been widely used in monitoring butterfly populations, for example in the studies of Van Swaay [
Like all methods of monitoring, puddling counts may be influenced to some degree by extraneous factors other than population size. Over a longer period of monitoring that examines trends over many years, these factors may be largely inconsequential to the interpretation of population trends. Variations in the availability of the resource over time and variations in attraction are the factors that could influence the counts of birdwings at the puddle. However, since the puddle is caused by a geothermal spring, it is stable year-round and unaffected by periods of dry weather. After rain, its geothermal and chemical properties are restored by underground seepage. The significant, positive correlation between the number of puddling birdwings counted on the ground and the number of patrolling birdwings counted in flight suggests that high and low numbers of puddling birdwings are not a result of temporal changes in puddling behavior such as changing levels of resource attraction, but rather proportional to variations in population size. Furthermore, there was no clear difference in the ratios of puddling to flying birdwings in the high- and low-season months, indicating that the aggregations were not obviously affected by group-size or conspecific visual attraction signals.
The first component study to determine the sample requirements for counts of puddling birdwings demonstrated that, in the absence of events that might disrupt puddling, just one image taken at each hour was sufficient to represent the numbers of birdwings puddling for the hour, because there was little difference between the averages and 95% confidence intervals when one, three or five images were used. The very slight increase in average counts with increasing numbers of images is to be expected, because the averages are centered one minute apart, and numbers of birdwings were on an increase during the 30-minute time periods on which the analyses were based. Where there is a disruptive event that causes the birdwings to take to flight, a delay of 10 minutes is sufficient to allow the birdwings to return before images are captured.
The strong positive correlation between average birdwing counts at 1600 hrs and the average of hourly counts over an eight-hour period (i.e., 0900–1600 hrs) in the second component study showed that just one hour’s sample at this peak puddling period was sufficiently representative of counts for the entire day. The results of the third study showed that the numbers of birdwings puddling during this peak hour of 1600 hrs could be predicted from the number of birdwings puddling at 1100 hours. Ideally, if a single, standardized hour is chosen for the monitoring of the birdwing at this site, it should be between 1400–1600 hours, which was observed to be the peak period for puddling activity. However, the flexibility to predict a standardized hour’s count from an alternative hour’s count is useful during seasons when heavy afternoon rain is expected. In the fourth component study, the lack of a consistent correlation between days indicated that there was a moderate amount of day to day variation in numbers of puddling birdwings within a month. This suggests that counts based on multiple days will yield a better estimate of the population than single-day counts.
In the 24-month monitoring data, brightness and humidity at the time of monitoring had the most significant influence on day to day differences within a month, with higher numbers of birdwings when it was bright and humid. Bright sunshine usually resulted in lower humidity, as could be seen by the negative correlation between these two variables, but where brightness and humidity were both high, this would have enhanced puddling the most, although not greatly, as seen from the low standardized regression coefficients. On the other hand, the population showed great variation that was significant between months over the two-year period, with three peaks of about 250–350 puddling birdwings, and lows of less than 25 individuals. With such high monthly variation, the day to day differences due to daily weather patterns were relatively small. Variations in weather patterns and host plant phenology over the years could be responsible to some extent for the population fluctuation. There is no obvious wet or dry season in the geographical area in which the site is located, and weather patterns are not always consistent from one year to another. All-year-round monitoring is therefore preferable at this site.
One limitation of a population index based on puddling birdwings is that it excludes females. In subspecies
The success of a long-term monitoring program often depends on volunteers. Having a simple, rapid, practical and affordable monitoring protocol that requires the minimum effort to produce significant results encourages long-term volunteer involvement [
One image taken for each puddle at one representative hour would be sufficient as an indicator of the number puddling on a particular day. Three images taken at the start of the hour for three peak hours in a day is not difficult, and should be considered for better population index accuracy in the monitoring program. Having the counts repeated on different days, to average out daily variation due to weather conditions, is more important than replication within each day. For this reason, we would recommend that counts be conducted over three days, which is still very manageable, or at least two days if resources are truly limiting. The method also yields counts that are not excessively affected by small variations in weather. If it rains the whole day, counts can be conducted on a different day. When there is a high likelihood of afternoon rains, usually from the months of March to June, counts can be conducted in the morning and afternoon counts determined by prediction from the regression line.
The method can also be used to monitor populations of
We envisage that the method can also be adapted to monitor other puddling butterfly species, including mixed species groups. While camera trapping is widely used to monitor larger wildlife [
(TIF)
Data shown are the monthly means and standard errors (data on both axes log transformed). The x-axis is taken as an independent variable for the purpose of obtaining a prediction equation.
(TIF)
(DOCX)
We thank Gard W. Otis, Michel Renou and an anonymous reviewer, all of whom asked useful questions and provided suggestions for improvement of this article We are grateful to the local Semai community in the village of Ulu Geroh, especially to Insan, Ah Ngah, Sani, Rohani and the late Aha, as well as to the village head, Ngah Sidin Hamzah, for their support of our study. We are also thankful to staff of the Entomology Branch of FRIM for invaluable field support, and to Tze-Leong Yao (Botany Branch, FRIM), Hin-Fui Lim (Sociology Branch, FRIM), Intan Nurulhani, Norsyakila Yusof, Itam Bahin (Commercialization and Innovation Division, FRIM), and Sonny Wong (Malaysian Nature Society), for their help rendered in various ways.