Marine mammals: methodology for combining data

Project Conclusions and Future Research Directions

This project addressed three main goals:

1. Produce a modelling framework integrating DAS data and PAM data, including the ability to incorporate seasonal and diurnal uncertainty.

2. Produce a test case study on harbour porpoise to validate the methods, producing density maps for a specified site in Scotland.

3. Provide recommendations on standards for static PAM and DAS data collection.

The first two goals were addressed through the review of available data integration methods and selecting a method for implementation in a case study. A calibration method was chosen and applied to a combined DAS and PAM dataset collected in the Moray Firth in 2010. By integrating DAS and PAM data, the different strengths of the data – the broader spatial coverage of aerial surveys and the long-duration, continuous monitoring of PAM surveys – were combined. The immediate benefit of combining the data was the conversion of the timeseries of PAM data into estimated absolute densities with associated uncertainty, including days when the DAS surveys were not operating. This may make it possible to investigate animal densities at a finer temporal resolution than with DAS data alone. In the case study, density surfaces were also estimated from the calibrated PAM data, showing spatial changes in absolute density. It is also potentially possible to apply this method to data already collected (as was demonstrated in the case study), assuming that an estimate of absolute abundance can be estimated from one of the survey platforms.

The key assumption is, however, that the parameters estimated from the combined DAS and PAM data are representative across the time period analysed. For example, in the case study, the DAS surveys were conducted on two dates in August 2010 and two dates in September 2010. It is therefore assumed that the corresponding estimated parameter, vp, combining the average effective detection area of the CPODs and the average probability of an animal clicking in a 1-second time period on those four days is representative of vp throughout all dates in August and September 2010. Further, the application of the combined parameter vp across the daily PAM dataset assumes that the average value of vp is constant across the time period being analysed and does not change from one day to the next. If this assumption is not true, then individual daily PAM estimates could be biased. Two next research steps would be to (1) assess through simulation studies how unaddressed variability in vp would impact the bias and precision of resulting abundance estimates and (2) compare the calibration approach with different data integration approaches (three broad categories were identified in the methods review) to better understand the advantages and disadvantages of the various methods.

Regarding the final goal, survey design considerations and recommendations were outlined for both DAS and PAM data separately, before considering survey design specifically for an integrated survey. The software review also highlighted that tools already exist to (1) aid DAS and PAM survey design and (2) assess the power of survey designs to detect change in trends in abundance and/or density (for both DAS and PAM surveys). These discussions led to identifying two further research steps: (1) there is a need to investigate how a dedicated integrated survey design would differ from the recommendations for individual DAS and PAM surveys, specifically regarding the number of PAM instruments and DAS flights required and (2) there is a need for a software tool to design a combined DAS and PAM survey, which could be an extension of existing tools.

Current survey design recommendations are to:

• Clearly identify the goals of a survey to ensure that the survey design will meet the needs of the survey goals. Goals may need to be prioritised where there are several competing goals and/or target species.
• Follow existing guidance for line and point placement for separate DAS and PAM surveys, though more research is needed to understand survey design requirements for an integrated survey.
• Use existing tools where possible to aid survey design, including assessing the power of the survey to detect changes in density and abundance. More software tool development is required specifically for integrated surveys.

In summary, data integration of DAS and PAM data is possible, providing a time series of absolute abundance estimates (with associated assumptions) that would not be possible using DAS data alone. Therefore, a further recommendation is to:

• Consider the benefits of collecting data from more than one type of surveying platform. Different platform types offer different advantages; in this study combining DAS and PAM data led to a time series of estimated absolute densities that would not have been practically possible from one platform alone. More research is required, however, to determine how many DAS flights are required, and at what intervals, to optimally calibrate the PAM data.

The demonstrated calibration method is flexible and so could be considered for use with other species and other surveying platforms (such as DAS using still images, ship-based surveys or autonomous vehicles). However, there are likely to be specific considerations for each type of surveying platform, which may require adjustments to the data integration method. Ultimately, data integration from several surveying platforms has the potential to impact survey design and data collection recommendations, which in turn may influence required survey effort and, therefore, survey costs.

Future Research Directions

Each stage of the project highlighted future research steps, as summarised here.

• The software review highlighted that there is a need for a software tool to design a combined DAS and PAM survey, which could be an extension of existing tools.
• Several extensions to the case study analysis would be beneficial:
  • Explore variability in the vp parameter as a function of space and time.
  • Assess via simulation how unaddressed variability in vp would impact the bias and precision of resulting abundance estimates.
  • Compare other reviewed methods with the calibration approach to assess the various strengths and limitations of the different methods.
  • Further work via simulation is required to ascertain how the inclusion of acoustic detection probability and cue production rates as informed priors, if available, would affect the precision and accuracy of the estimated parameters.
  • Survey design considerations could be tested via simulation (based on the case study data) by (1) performing a down-sampling analysis to assess how many PAM instruments are required to avoid bias and achieve a suitable level of uncertainty in the resulting abundance estimates and (2) assess how many DAS surveys would be required to adequately calibrate the PAM data.
• Continued research into estimating (1) detection probability and (2) availability parameters for DAS data is important, given the need to estimate absolute density from the DAS data when using the calibration method. In addition, extracting group size information from DAS data might also be useful as it is linked to detection and availability parameters.