BIOTROPICA 50(6): 846–849 2018 10.1111/btp.

12608
C O M M E N TA RY

Surveying tropical birds is much harder than you think: a primer of best practices

W. Douglas Robinson1,5 , Alexander C. Lees2,3, and John G. Blake4

1
Oak Creek Lab of Biology, Department of Fisheries and Wildlife, Oregon State University, Corvallis, OR 97331, USA
2
School of Science and the Environment, Manchester Metropolitan University, Manchester, UK
3
Laboratory of Ornithology, Cornell University, Ithaca, NY 14853, USA
4
Department of Wildlife Ecology and Conservation, University of Florida, Gainesville, FL 36211, USA

ABSTRACT
Birds are tempting to include in studies of tropical ecology and conservation. Yet, they are deceptively difficult to detect, identify and, parti-
cularly, count. We briefly review some common challenges of surveying tropical birds, offer guidance on the most important decisions to
consider when selecting methodologies, and recommend best practices to ensure collection of reliable, repeatable, and reviewer-friendly
survey data.

Key words: bird counting methods; community inventory; imperfect detection.

BIRDS ARE THE MOST WELL KNOWN OF ALL TROPICAL BIOTA AND FRE- dark and structurally complex rain forest environments where
QUENTLY A TARGET FOR STUDY BY ECOLOGISTS wishing to evaluate upwards of 95 percent of birds are heard but never seen by an
ecological dynamics of tropical habitats or to compare biodiver- observer. Furthermore, most species utter many sounds aside
sity responses across different land-cover types or treatments from typical ‘loud’ songs, which means that field surveys necessi-
(Remsen 1994, Blackburn & Gaston 1998). The allure of birds tate familiarity with potentially thousands of different bird vocal-
reflects (1) their ubiquity and diversity; (2) their well-documented izations (Remsen 1994). Mastering the suite of sounds requires
responses to environmental variation; (3) the relative ease with months of daily fieldwork in a region (Parker 1991), although the
which most can be identified during field studies; and (4) their ready availability of electronic sound files online can now acceler-
relative cost-effectiveness as a study taxon (Gardner et al. 2008). ate the learning process for bird surveyors.
The availability for researchers of huge data bases on avian phe- Here, we comment on several methodological approaches
notypes, genotypes, functional traits, natural histories, and popula- that can improve estimates of abundance and community-level
tion levels also improves the value of birds as a model taxon for richness, some of which are simple to implement whereas others
studies of ecological questions and conservation challenges. require deeper thought about the specific objectives of each
Yet, identifying and counting birds, particularly many tropical study. Our overview is intended mostly for researchers who wish
species, can be deceptively difficult. The utility of birds in ecolog- to incorporate birds into ecological or conservation studies but
ical studies can be greatly compromised if the frequency of false- who may not yet have sufficient experience with the exceptional
negative and false-positive identification errors is high (Remsen challenges of surveying tropical bird communities and estimating
1994, Lees et al. 2014). From our experience as journal editors abundances. Experienced tropical field ornithologists will know
and reviewers, we have strong reason to believe that the level of that evaluating sources of error and bias in surveys of tropical
error in many tropical avian surveys is a cause for concern. Trop- bird abundances and inventories of richness are active areas of
ical forests, for example, with their extremely high avian diversity investigation (Gale et al. 2009, Anderson et al. 2015, Peele et al.
are particularly difficult to sample adequately as observers need 2015, Gomez et al. 2017).
to be thoroughly familiar with a very large regional species pool For assessments of species richness within tropical bird
containing many rare species (Terborgh et al. 1990, Robinson communities, no field surveys are ever ‘complete’ and thus dis-
et al. 2000). The rate at which observers misidentify or fail to play a trade-off between the time available for the surveys, the
detect birds is influenced by level of expertise, as well as method- methods used during surveys, and expertise of the surveyors. It
ology and duration of surveys (Gaston 1996). Auditory-visual is therefore extremely important to understand how incomplete
surveys in tropical forests pose exceptional challenges in typically or inaccurate any given survey is, so that only comparably com-
plete data sets are used in analyses, or appropriate statistical mea-
Received 16 April 2018; revision accepted 21 August 2018. sures taken into account for differences in completeness. The
5
Corresponding author; e-mail: [email protected] degree to which surveys approach complete inventories is
846 ª 2018 The Association for Tropical Biology and Conservation
 Surveying Tropical Birds 847

affected largely by non-detection of individual species, which can availability and perceptibility by incorporating multiple sampling
be ascribed to two different components of detection (Boulinier methods with appropriate design characteristics, using highly
et al. 1998, Nichols et al. 2009). Availability is whether a species skilled observers, and archiving sound files to allow re-examina-
was vocalizing or visible from a point or a transect. Perceptibility tion of soundscapes and add an element of repeatability to the
is whether a species was detected and identified correctly by the study. We first discuss the use of mist nets as a technique for
observer (Marsh & Sinclair 1989, Johnson 2008). Availability will sampling tropical bird communities, then highlight some aspects
be affected by survey protocols such as timing (across diel and of auditory-visual surveys bird counters should consider carefully.
annual cycles, given the phenology of breeding patterns and Mist nets have been used to sample tropical bird communi-
migration) and length of survey periods in addition to ambient ties for many years but their use has been controversial (Ter-
conditions such as weather and distance from the observer. Per- borgh et al. 1990, Remsen & Good 1996). Much of the
ceptibility is more a function of observer experience and will be controversy has focused on the question of whether or not cap-
influenced by distance to the bird, mediated by signal attenuation ture rates of birds in mist nets can provide a useful index of
in denser habitats and the conspicuousness of the signal. abundance or activity. Given the many factors that influence cap-
Species missed by inexperienced observers tend to be a non- ture rates (e.g., behavior, habitat structure), it is unlikely that mist
random subset of those in an assemblage, leading not only to spu- nets will, at least in most instances, provide a true estimate of
rious assessment of richness but distorted interpretations of abundance. Thus, interpretations of results based on mist-net
macroecological patterns and the conservation value of treatments captures must be done with caution. This may be particularly
(Remsen 1994, Blackburn & Gaston 1998). Best practices for true when comparing results from study sites that differ in habi-
assessing completeness of community surveys include quantifica- tat structure; ground level mist nets will certainly sample a greater
tion of survey completeness using species accumulation curves. To proportion of species in habitats with short and dense vegetation
improve efficiency of work, researchers should also consider use when compared to tall forests. Similarly, care must be taken when
of a priori stopping rules, which are quantitative estimates of sur- comparing rates across species as differences in behavior may
vey completeness determined from predicted shapes of species have a significant impact on the likelihood of capture. Although
accumulation curves (Cam et al. 2002). Software for such analyses mist nets will not provide a complete sample of bird communi-
is freely available (Colwell 2013, Hsieh et al. 2016). Estimators ties in lowland tropical forests, they can be a useful addition to
generated from inadequate community inventories may lead one other sampling methods such as point-counts or territory map-
to conclude sufficient effort has been invested, even when impor- ping. Mist nets do provide the advantage that captured birds may,
tant components of a community have been missed. Furthermore, in most cases, be identified by researchers who lack knowledge to
estimators may predict richness greater than that actually detected, identify birds by sound alone. Another advantage is the opportu-
but provide no information on the ecological traits of ‘missed’ nity to color-mark individuals and facilitate efforts for spot-map-
species. In addition to careful use of statistical estimators, we ping territories, but such approaches require very large
encourage collaboration with local experts whose expertise may investments of effort. In some cases, mist nets may better sample
expedite learning sounds of rarely detected species. Given that some species which vocalize infrequently and that might be over-
accumulation curves in most communities will rarely stop climbing looked with other methods (e.g., Sclerurus leaftossers, some under-
because of detection of vagrants and species with very large home story frugivores), thereby helping to provide a more complete
ranges and low densities, local experts can also provide advice on species list for a given site. When mist nets are deployed in sites
completeness of surveys for expected resident species. with similar habitats, a comparison of capture rates can provide
For example, some tropical bird species are both visually an index with which to compare levels of activity between sites,
cryptic and vocalize infrequently. A recent tracking study of Var- but only for those species likely to be captured in nets. Overall,
iegated Antpitta (Grallaria varia) revealed occurrences of singing mist nets can be useful in studies of tropical bird communities
at just two of 68 locations within its home range (Jirinec et al. but only if their limitations and benefits are recognized.
2018). Detection of such species requires exhaustive sampling One of the most important decisions to be made when
regimes to avoid problems of low availability. Yet, the effort designing auditory-visual surveys of tropical birds involves survey
required to detect and count such rare species is important as radius. The area surveyed influences estimates of abundance as
they are among the most sensitive to environmental change well as probability of detection. Most birds in tropical habitats
(Banks-Leite et al. 2014). On the other end of the abundance are heard and not seen. Although tempting to keep distances
spectrum, we have noticed, as editors, the absence or relative from observer small to reduce chances of missing birds, the
scarcity of some core rain forest species which ought to be abun- shorter the distance the more likely it is that abundances will be
dant in many Neotropical locations, including canopy Zimmerius severely under-estimated because many birds will be missed.
tyrannulets and small understory woodcreepers such as Glyphor- Most birds simply move away in response to observer presence,
ynchus spirurus which often are among the commonest species in while a very few species may be attracted to observers. For
many landscapes (high availability) but easily overlooked because example, during line transect surveys, most birds are rarely
of their unassuming vocalizations (low perceptibility). detected along the transect because they move away as observers
Adequately designed and executed surveys of tropical bird walk toward them. During stationary surveys, birds are also likely
communities need to compensate for this great variability in to detect an observer and move away or stop vocalizing. Most
848 Robinson, Lees, and Blake

tropical field ornithologists recommend including birds detected birds, where count durations of 3, 5, and 8 min are widely used,
at all distances from the observer and recording the distance (ei- there is no standardized length of counts widely used across trop-
ther measured with range-finder, range-finder binoculars or by ical bird communities. In part, this is because there are currently
estimating distance if one is very experienced with distance esti- no continent-wide surveys like the North American Breeding
mation). Use of unlimited distance radii allows one to reduce Bird Survey (using 3-min counts) in Latin America, and the real-
effects of observer proximity on bird behavior while also accu- ization that many tropical birds vocalize more infrequently so
mulating more data than one would gather with short radii. Most count duration often needs to be longer. Researchers must bal-
tropical field ornithologists do not necessarily use all of the data ance efficiency of surveyor effort and completeness of sampling
collected from unlimited distances to estimate abundances as a function of count length. Longer counts increase chances
because ability to hear or see species diminishes with distance, that individual birds may move into range of detection after the
but those data can be used in analyses of community richness. count starts, so the actual area being surveyed becomes difficult
The distance at which detectability declines varies among species. to ascertain. Shorter counts can be too brief to allow opportunity
We often truncate the data at some distance (typically to detect a reasonably high fraction of birds actually present.
75 m–150 m, depending on the species being studied and struc- Commonly used count durations in tropical forests are 8,
tural complexity of the habitat). As editors, we have seen many 10, and 15 min. These durations seem to represent a reasonable
studies where data were collected at radii of 10 or 20 m, presum- balance allowing detection of infrequently vocalizing species and
ably to avoid the issue of missing more distant birds. But our minimizing excessively large rates of movements into detection
experience tells us that the biases appearing from bird response range, but we know of no rigorous assessments of this balance
to observer presence are just too great to justify use of such nor how it varies across a diversity of habitats. An advantage of
short radii. We recommend researchers use unlimited distance 8 or 10 min counts is that such counts can be divided easily into
surveys and note the distances in their data, then truncate data 2-min intervals. Experienced observers can then track detections
later to appropriate distances based on the species-specific dis- of individual birds in each 2-min intervals, creating a ‘capture-
tance histograms. Processes for determining appropriate trunca- recapture’ history and allowing a variety of additional methods
tion distances are explained in the program DISTANCE for estimating detectability and abundance (Farnsworth et al.
(Thomas et al. 2010). Note that DISTANCE requires sample 2002, Alldredge et al. 2007a,b). It is essential that observers
sizes of detections larger than can be normally acquired for many attempt to use modern methods to adjust for detectability issues
rare tropical species. Analytical strategies that compensate for when a goal is to estimate abundance. Use of unadjusted count
such limitations are becoming more common (Gomez et al. data, even when data are sparse, can produce erroneous impres-
2017). sions of abundance differences across species. Yet, tropical bird
Quantifying distances from observer presents substantial species can be so rare that accumulating sufficient data to use
challenges. Most birds are not seen but heard, so determining modern methods is a challenge. Combining data across ecologi-
distance accurately requires extensive experience and an under- cally and vocally similar species to boost sample sizes could be
standing of how sounds of different frequency ranges (high vs. attempted (Dorazio et al. 2013, Iknayan et al. 2014, Gomez et al.
low pitch) attenuate as a function of habitat structure and even 2017). Restricting the set of questions asked may be necessary.
humidity conditions or other ambient environmental noise. Such Sometimes, careful design of study questions and sampling strat-
experience is typically only accumulated after months, sometimes egy can reduce the need for sophisticated and data-hungry sam-
years of survey work. We encourage inclusion of details in manu- ple methods (Banks-Leite et al. 2014). For example, comparing
scripts (at least in supplementary online material) summarizing changes over time in richness and estimated abundances or num-
the experience level of data collectors (Robinson & Curtis in bers of detections per species at a given site might be achieved
review). Observers could maintain and update regularly (daily or simply by ensuring use of identical methods and (ideally) obser-
weekly) a list of species they know they can identify by sound, vers (Blake & Loiselle 2015). More complicated goals such as
which could become part of project metadata. In addition, we comparing variability in abundances of species across guilds
recommend researchers supplement their work by archiving might be confounded by substantial differences in detectability
recordings of bird sounds for review. The recordings can be that are difficult to quantify and control.
‘voucher’ specimens for individual species and archived in sites We have encountered many troubling issues in manuscripts
such as xeno-canto.org or the Macaulay Library (through eBir- that include tropical bird data, only a few of which we have
d.org). We also see value in archiving ambient recordings of the addressed here. The most important challenges to address
entire bird community soundscape within a time frame beginning include: staying current with modern literature on how best to
30 min before dawn and extending to 2 h after dawn, the time adjust count data when a goal is to estimate abundance; objec-
period when most diurnal species vocalize. Such recordings tively assessing completeness of community-level surveys by using
should be at least 15–30 min in duration. The Macaulay Library accumulation curves and objective stopping rules; ensuring use of
at the Cornell Lab of Ornithology is an option for archiving such experienced observers who are very familiar with all sounds of
recordings in a publicly available web space. species in the regional species pool; creating opportunity for
Another commonly variable characteristic of avian surveys is repeatability of assessments by archiving sound files in publicly
duration of counts. In contrast to studies of temperate zone available sites for independent review of identifications; and
 Surveying Tropical Birds 849

archiving in supplementary materials lists of species encountered GARDNER, T. A., J. BARLOW, I. S. ARAUJO, T. C. A  VILA-PIRES, A. B. BONALDO, J.
and numbers of detections of each species so that experts can E. COSTA, M. C. ESPOSITO, L. V. FERREIRA, J. HAWES, M. I. HERNAN-
DEZ, AND M. S. HOOGMOED. 2008. The cost-effectiveness of biodiver-
assess the degree to which species inventories appear to be com-
sity surveys in tropical forests. Ecology Letters 11: 139–150.
plete. Tropical bird communities are rich, important ecologically GASTON, K. J. 1996. The multiple forms of the interspecific abundance-distri-
and are tempting to include in many types of studies. But tropical bution relationship. Oikos 76: 211–220.
birds are deceptively difficult to detect and count accurately, so GOMEZ, J. P., S. K. ROBINSON, J. K. BLACKBURN, AND J. M. PONCIANO. 2017.
great care should be taken when designing and conducting stud- An efficient extension of N-mixture models for multi-species abun-
dance estimation. Methods in Ecology and Evolution 9: 340–353.
ies. Detailed methodology and metadata should be included to
HSIEH, T. C., K. H. MA, AND A. CHAO. 2016. iNEXT: An R package for rar-
allow careful evaluation of potential sampling biases. efaction and extrapolation of species diversity (Hill numbers). Methods
in Ecology and Evolution 7: 1451–1456.
LITERATURE CITED IKNAYAN, K. J., M. W. TINGLEY, B. J. FURNAS, AND S. R. BEISSINGER. 2014.
Detecting diversity: Emerging methods to estimate species diversity.
ALLDREDGE, M. W., K. H. POLLOCK, T. R. SIMONS, J. A. COLLAZO, AND S. A. Trends in Ecology & Evolution 29: 97–106.
SHRINER. 2007a. Time-of-detection method for estimating abundance JIRINEC, V., E. C. ELIZONDO, C. L. RUTT, AND P. C. STOUFFER. 2018. Space
from point-count surveys. Auk 123: 653–664. use, diurnal movement, and roosting of a variegated antpitta (Grallaria
ALLDREDGE, M. W., T. R. SIMONS, K. H. POLLOCK, AND K. PACIFICI. 2007b. A varia) in central Amazonia. Ornitologıa Neotropical 29: 13–20.
field evaluation of the time-of-detection method to estimate popula- JOHNSON, D. H. 2008. In defense of indices: The case of bird surveys. Journal
tion size and density for aural avian point counts. Avian Conservation of Wildlife Management 72: 857–868.
and Ecology 2: 13. LEES, A. C., L. N. NAKA, A. ALEIXO, M. COHN-HAFT, V. D. Q. PIACENTINI, M.
ANDERSON, A. S., T. A. MARQUES, L. P. SHOO, AND S. E. WILLIAMS. 2015. P. D. SANTOS, AND L. F. SILVEIRA. 2014. Conducting rigorous avian
Detectability in audio-visual surveys of tropical rainforest birds: The inventories: Amazonian case studies and a roadmap for improvement.
influence of species, weather and habitat characteristics. PLoS ONE Revista Brasileira de Ornitologia-Brazilian Journal of Ornithology 22:
10: e0128464. 107–120.
BANKS-LEITE, C., R. PARDINI, D. BOSCOLO, C. R. CASSANO, T. PUTTKER, C. S. MARSH, H., AND D. F. SINCLAIR. 1989. Correcting for visibility bias in strip
BARROS, AND J. BARLOW. 2014. Assessing the utility of statistical adjust- transect aerial surveys of aquatic fauna. Journal of Wildlife Manage-
ments for imperfect detection in tropical conservation science. Journal ment 53: 1017–1024.
of Applied Ecology 51: 849–859. NICHOLS, J. D., L. THOMAS, AND P. B. CONN. 2009. Inferences about landbird
BLACKBURN, T. M., AND K. J. GASTON. 1998. Some methodological issues in abundance from count data: Recent advances and future directions.
macroecology. American Naturalist 151: 68–83. Journal of Ecological and Environmental Statistics 3: 201–236.
BLAKE, J. G., AND B. A. LOISELLE. 2015. Enigmatic declines in bird numbers PARKER, T. A. 1991. On the use of tape recorders in avifaunal surveys. Auk
in lowland forest of eastern Ecuador may be a consequence of climate 108: 443–444.
change. PeerJ 3: e1177. PEELE, A. M., P. M. MARRA, T. S. SILLETT, AND T. W. SHERRY. 2015. Combin-
BOULINIER, T., J. D. NICHOLS, J. R. SAUER, J. E. HINES, AND K. H. POLLOCK. ing methods to estimate abundance and transience of migratory birds
1998. Estimating species richness: The importance of heterogeneity in among tropical nonbreeding habitats. Auk 132: 926–937.
species detectability. Ecology 79: 1018–1028. REMSEN, Jr, J. V. 1994. Use and misuse of bird lists in community ecology
CAM, E., J. D. NICHOLS, J. R. SAUER, AND J. E. HINES. 2002. On the estimation and conservation. Auk 111: 225–227.
of species richness based on the accumulation of previously unrec- REMSEN, Jr, J. V., AND D. A. GOOD. 1996. Misuse of data from mist-net captures
orded species. Ecography 25: 102–108. to assess relative abundance in bird populations. Auk 113: 381–398.
COLWELL, R. K. 2013. EstimateS: statistical estimation of species richness and ROBINSON, W. D., J. D. BRAWN, AND S. K. ROBINSON. 2000. Forest bird com-
shared species from samples. Version 9 and earlier. User’s Guide and munity structure in central Panama: Influence of spatial scale and bio-
application. http://purl.oclc.org/estimates. geography. Ecological Monographs 70: 209–235.
DORAZIO, R. M., J. MARTIN, AND H. H. EDWARDS. 2013. Estimating abundance ROBINSON, W. D., AND J. R. CURTIS. In review. Benchmarking bird communi-
while accounting for rarity, correlated behavior, and other sources of ties in tropical forests. Studies in Avian Biology.
variation in counts. Ecology 94: 1472–1478. TERBORGH, J., S. K. ROBINSON, T. A. PARKER, III, C. A. MUNN, AND N. PIER-
FARNSWORTH, G. L., K. H. POLLOCK, J. D. NICHOLS, T. R. SIMONS, J. E. HINES, PONT. 1990. Structure and organization of an Amazonian forest bird

AND J. R. SAUER. 2002. A removal model for estimating detection community. Ecological Monographs 60: 213–238.
probabilities from point-county surveys. Auk 119: 414–425. THOMAS, L., S. T. BUCKLAND, E. A. REXSTAD, J. L. LAAKE, S. STRINDBERG, S. L.
GALE, G. A., P. D. ROUND, A. J. PIERCE, S. N. A. PATTANAVIBOOL, AND W. Y. HEDLEY, J. R. B. BISHOP, T. A. MARQUES, AND K. P. BURNHAM. 2010.
BROCKELMAN. 2009. A field test of distance sampling methods for a Distance software: Design and analysis of distance sampling surveys
tropical forest bird community. Auk 126: 439–448. for estimating population size. Journal of Applied Ecology 47: 5–14.