Dr. Jessica Bryant
The Zoological Society of London
The team traveled to Bawangling National Nature Reserve on Hainan Island in January to explain the Wildlife Acoustics technology and overall methodology to the local reserve staff, identify sites for Song Meter SM3 deployment, and collect recordings from the Hainan gibbon population to assist later analyses. The Song Meter SM3 recorders were successfully deployed at the start of this month (March), and are now out in the Bawangling landscape, recording the songs of the Hainan gibbon groups. The Song Meters will remain out in the forest for at least four more months during which time they will record from before dawn every morning for several hours, and be replenished with new batteries and memory cards every month to ensure they continue to record the gibbon's daily songs. In the meantime, the team will use the recordings compiled from January to commence the initial stages of analysis to develop recognizers within Song Scope. This approach will permit later analyses using the several months' worth of SM3 recordings.
The Song Meter SM3 recorders remain deployed in the nature reserve, where they are fastidiously collecting call data from one hour before sunrise, at which point the gibbons being to sing their bond-reinforcing songs. This occurs until the early afternoon, when the gibbons are less likely to call as they begin to settle down in their sleeping trees. The recorders are located near to each of the three, well-established Hainan gibbon family groups in the hope that the team might be able to detect between-group song differences at the analysis stage. Dr. Bryant has commenced recogniser development by annotating calls recorded earlier on the year within Song Scope.
Following deployment of our Song Meter SM3 recorders in Bawangling National Nature Reserve (BNNR) for a total of six months (since March 2016), all recorders were successfully retrieved from the forest in early September (after recording ceased in late August). This signals the end of this stage of the project. We have around 5,500 hours of raw recording data, although not all of these will necessarily contain gibbon songs! This will depend on whether the gibbon groups ranged close enough to the recorders for their song to be captured, and the extent of programming issues we encountered during redeployment associated with operation of the recorders by our non-English speaking BNNR Management Office collaborators.
We have now begun the analysis phase of the project using Song Scope. Along with formal analysis of the recordings to determine the capture rate for gibbons in this type of mountainous environment, and whether we can discern the individual gibbon social groups using this approach, a key step will also be to determine the overall success to date and feasibility of this method as a long term monitoring technique for BNNR Management Office staff, given the language and skill barriers encountered with station deployment and maintenance. Together, data on these different aspects will allow us to thoroughly assess the practicability of this new technique as a potential monitoring tool to strengthen the capacity of BNNR Management Office to detect and monitor Hainan gibbons at Bawangling.
Following the conclusion of our field phase of the project, in the last quarter we progressed to analysis. Using a total of almost nine hours of reference recordings, we constructed individual recognisers within SongScope for a repertoire of Hainan gibbon vocalisations: solo male song, male-female duet, alarm call, and where possible, group-specific recognisers for each of these (note: there are currently four Hainan gibbon social groups, but we have recordings for only three of these due to the sensitivity of the newly formed fourth group to human presence).
Using 10 different recogniser configurations, we scanned 12 files of 8 hours duration each known to contain Hainan gibbon calls (so 96 hours of raw recording data in total, and approximately 2 percent of the total hours of raw recording data). We then compared the performance of these recogniser configurations in terms of the number of true positives, false positives, false negatives and their relative proportions. Using the recogniser configuration that performed the best in these aspects, we then further refined the minimum score and quality settings through further testing and performance comparison to determine the most successful recogniser settings for our data.
We were therefore able to create two general (species-specific but non group-specific) recognisers for the Hainan gibbon (solo male song, male-female duet) that successfully detected the species' calls within our Song Meter SM3 recordings, with a suitable balance of true versus false positives and negatives. Due to a lower amount of reference data for the alarm call vocalisations generally, and all vocalisation types for certain gibbon social groups (those that are unhabituated to human presence), the group-specific recognisers did not prove to be as successful as the general species recognisers, with a higher than desired ratio of false positives (and false negatives) to true positives. As such, further testing and refinement of the group-specific recognisers is needed and will be carried out as soon as we can collect additional individual group reference calls, as and when this is possible.
We now plan to liaise with Bawangling National Nature Reserve staff and Hainan Wildlife Conservation Bureau regarding the use of the developed Hainan gibbon recognisers and SM3 recorders to monitor Hainan gibbons within Bawangling, and also hopefully to survey for any possible remnant Hainan gibbon populations outside BNNR (in other key reserves across Hainan) using this tool. However, it is clear that we will need to carry out additional work to streamline the recorder deployment process, and possibly collaborate with appropriate parties to develop a Chinese language version of the Song Meter (SM3) firmware — as we found that deployment success was constrained by the English-only interface. These steps could certainly help to enhance this passive acoustic detection technique as a potential monitoring tool for Hainan gibbons at Bawangling, and possibly beyond, into the future.