Marine mammals: methodology for combining data

Section 4: Survey design recommendations

The discussions within the final technical meeting covered various aspects of survey design for both DAS and PAM surveys but ultimately considered those that would enable data integration for estimation of absolute abundance using PAM. The main discussion points are summarised below. Many discussion points relate to both DAS and PAM surveys, though points specific to DAS or PAM are highlighted.

Goals of a survey

Firstly, it is important to define the aims of any survey because the ultimate project goal will determine the survey design required. For example, a PAM survey designed to estimate absolute density or abundance from PAM data alone will need to consider detection probability estimation (and cue production information). In PAM surveys, this will often lead to a specific instrument configuration and will likely require more instrumentation overall (to create instrument arrays for animal localisation, for example). More complex data processing will also be required to estimate detection probability from PAM data. Further, if a survey is designed to detect a response to some disturbance, then particular designs e.g., gradient designs (as used in Thompson et al., 2013) may be required. In this project, the role of the PAM data was to supplement the absolute density estimates from the DAS data by providing data at a finer temporal resolution. Therefore, detection probability estimation and cue production information from the PAM data was not required (though see below for an extended discussion about detection probability estimation).

Survey effort

Sufficient replication of transect lines/points and obtaining uniform survey coverage are two important components of survey design (e.g., Buckland et al., 2001). Guidance is given in Buckland et al. (2001; 2015) regarding the number of lines/points required to achieve a desired level of variance (a minimum of 10 – 20 lines or points should be considered for an individual line- or point-transect survey). Tools such as dssd and dsims can also be used to design surveys for both line (for DAS surveys) and point (for stationary PAM surveys) transects. They can help to assess a number of design options, for example, whether the number of lines/points will achieve a reliable estimate of the encounter rate variance, whether the coverage probability within the survey area is uniform or whether stratification can help achieve more precise estimates. Using such tools as part of any survey design exercise is recommended. An example for one of the surveyed areas in the case study is presented in Appendix 1.

A further step would be to conduct a power analysis to investigate whether the survey design in question has enough power to detect changes in density/abundance (if that is the goal of a given project) using tools such as MRSeaPower or AVADECAF.

Required parameters for density/abundance estimation

There are several required parameters for absolute density/abundance estimation. These parameters will fundamentally affect precision and accuracy in density and abundance analyses. Therefore, it is recommended that studies attempt to estimate these parameters where practical, rather than relying on literature values, which may introduce bias due to geographical and temporal variability. More detailed comments on specific parameters are given below.

Detection probability: it is generally assumed that a digital aerial platform detects all animals available for detection within the surveyed strip but this may not always be the case. Detectability can be estimated for DAS data (likely using distance sampling) and this is an active area of research. Regarding PAM data, there are several ways to estimate detection probability (Marques et al., 2013) though, as discussed above, this has implications for survey design. In this project, a method is demonstrated that does not require detection probability to be directly estimated from the CPOD data, though prior information about detection probability could be included in the Bayesian model, if known. Further work via simulation is required to ascertain how the inclusion of detection probability as an informed prior would affect the precision and accuracy of the estimated parameters.

Group size: group size information from the DAS data might lead to insights about demography of the target species but is also linked to estimating availability (see next discussion point). Therefore, extracting group size information from DAS data could be useful.

Availability: estimation of availability parameters of relevance to DAS data is an active research topic. While instantaneous availability estimates are available for harbour porpoise from dive tag data (Teilmann et al. 2013), their application has limitations (e.g. unvalidated assumptions regarding animal visibility at depth), and efforts are underway to address these. Work is also ongoing by HiDef to generate estimates of availability based on a tandem aircraft approach.

Cue production rate (or similar vocal behaviour-linked parameters): again, this is an active area of research (for example ACCCURATE project, University of St Andrews). Similar to detection probability, a future research step would be to assess how the inclusion of prior information about the cue production rate of the target species affects the precision and accuracy of the estimated parameters.

Considerations specific to combining DAS and PAM data

It is important to note that by combining DAS and PAM data in this project, there was no requirement to estimate absolute density directly from the PAM data. Therefore, the survey design principles outlined above may not be strictly required for a combined DAS-PAM survey of this nature. So, while a conservative approach to the PAM survey design should start using the guidance referenced above (e.g., Buckland, 2001; 2015) about number of instruments, it is possible that some of the survey design principles can be relaxed when the PAM data only support the DAS data. However, this would require a dedicated down-sampling analysis, where instruments are removed in a simulation analysis to assess how many are required to still achieve the scientific objectives. This would be a natural next research step for this topic and similar exercises have been undertaken for other data integration methods (e.g., Schliep et al., 2023).

Given the need to estimate absolute density from the DAS data then, when using the calibration method, efforts must be made to collect the most robust DAS data including data required for additional parameters as discussed above. Further, another next step would be to use simulation studies to assess how many DAS surveys would be required to adequately calibrate the PAM data i.e., how many flights, and at what intervals, are required to avoid bias and maintain an acceptable level of uncertainty in the calibrated PAM data.