During my Honours research evaluating the factors determing bird diversity in remnants of Buloke woodland, I became aware of the limitations of fixed-effort, area-based methods to estimate bird diversity.  I used 200 x 50 m transects to estimate the number of birds occurring in a woodlands, but would regularly see or hear species outside of the transect boundaries, both during the survey but also while walking to and from the transect area.  During analysis, it became clear that the estimates I was using (based on 12 surveys per patch) were not accurate reflections of who was actually living there; and so began the search for a better approach…

Buloke woodland at Glenlea, near Gerang Gerung

For one aspect of my PhD research, I studied fragments of humind pine-oak forest in the highlands of southern Mexico.  The 17 forests I studied ranged in size from 2 hectares up to almost 160,000 hectares so, as well as minimizing the number of detected but ignored species, the method would need to be scalable (by 6 orders of magnitude–no easy task!).

View across the Oaxaca Valley from Yagul. Remnant hornscrub occurs throughout the valley, but humind pine-oak forest is restricted to the surrounding Sierras

Around this time, my PhD advisor Town Peterson was working with stopping rules and species accumulation curves in biodiversity inventories, so I trialled the use of a results-based stopping rule: “stop sampling only after two consecutive samples in which a total of one or zero new species was added” see Watson 2002 pdf for further details on my Oaxaca study.

I then formalized this approach as the “standardized search” pdf and compared it with other methods used to estimate species richness of birds pdf.

This work was related to relatively low bird diversities in relatively small patches—how well would it work in a large, diverse site?  To find out, I conducted an exhaustive 32 6 hour bird surveys of Barro Colorado Island, Panama, generating a series of predictions about the numbers of species present but missed, then testing those predictions with a further 200 hours of surveys.  This work is described here pdf and is explained in an associated podcast.

Conducting a bird survey on Barro Colorado Island, Panama. In closed canopy forests, most birds are detected by call, rather than visual cues.

This research is ongoing, with current work evaluating the effect of sample completeness on ecological inference, and comparing quantitative and quantitative stopping rules in terms of both accuracy and practicality.  Although this work has been carried out in forests and woodlands, these issues are broadly relevant to other systems, including wetlands.