Hi, 

I think Steffen provided a good summary of the challenges. 

IMO, the two most promising methods of estimating animal density from acoustic monitoring are the "generalised random encounter model" (gREM) and an extension of spatially-explicit capture-recapture (SECR). These potentially solve Steffen's problems 2 and 3, i.e. linking acoustic counts to abundance (2), and converting abundance to density (3). 

gREM, although theoretically possible, may be tricky in practice (especially obtaining an estimate of how wide an animal's acoustic signal is). See here:

http://onlinelibrary.wiley.com/doi/10.1111/2041-210X.12346/abstract

SECR is potentially very powerful, but depends on an independent estimate of calling rate (a problem Yu Shiu rightly picked up on). I think this would be entirely possible for a species you can find and observe (e.g. a frog or common bird species), but difficult for lots of cryptic, low density species (e.g. tigers!, as Courtney mentioned). See here:

https://www.researchgate.net/publication/272371302_A_general_framework_for_animal_density_estimation_from_acoustic_detections_across_a_fixed_microphone_array

However, gREM and SECR do not help with Steffen's first point (1) "quantifying the number of vocalisations from a stream of acoustic recordings". Others might be able to advise on the best approaches there. Perhaps this is primarily a software / data processing problem...?

In terms of sampling design (as Mariane and Courtney were interested in), it depends what your aims are. For occupancy (which is not equal to abundance/density), similar design principles to camera-trapping are fair (but taking into consideration Yu Shiu's point that the effective sampling area might be MUCH larger for an acoustic sensor than a camera trap, so camera spacing will have to be larger too). For gREM, you can fairly flexible about sampling design (repeated detections of the same individual are not a problem), but your sensors should be set randomly in space (with respect to animal movement), not along trails etc. For SECR, you don't have to set your sensors randomly, but sensors must be close enough together for repeated detections of the same call in multiple sensors simultaneously (this design constrasts, therefore, with an occupancy design). 

Hth,

Ollie

Hi Stephanie,

This is an interesting thread. For those interested in the topic, and forgive me for the blatant self-promotion of work, a Biological Reviews 2013 paper on the topic can be downloaded here   http://www.creem.st-and.ac.uk/decaf/outputs. Additional case studies papers as well as a more general public paper in Acoustics Today are also available for download from the link. This link is the outputs page from project DECAF, dedicated to estimating animal density from passive acoustic fixed sensors, using cetaceans as case studies. The methods have been far more developed and used in the cetacean community, but I suspect a wider use in terrestrial environments will occur in the coming years. A key hurdle is perhaps the dynamics involved in acoustic cue rate production, but the issues will be much easier to tackle in terrestrial environments than in the marine environments we have been working on.

Hope this is helpful,

Tiago