Mist-nets are widely used to survey bird presence, abundance, and survival (Dunn and Ralph 2004), for example at bird banding stations (Spotswood et al. 2012) and in the Monitoring Avian Productivity and Survivorship (MAPS) and el Monitoreo de Sobrevivencia Invernal (MoSI) Programs (DeSante et al. 1999, DeSante and Kaschube 2009). Common, widely used mist-net deployment methods facilitate comparability between datasets collected by different researchers (Karr 1981, Remsen and Good 1996, DeSante et al. 2020). That consistency, however, amplifies the risk of bias because the broad usage of a method may cause any biases inherent to that method to be replicated across many datasets, potentially leading to large errors (Dufour and Weatherhead 1991, Remsen and Good 1996, Larsen et al. 2007, Bonter et al. 2008). Mist-netting data can underlie population trend detection, range shift monitoring, species assemblage detection, the design and evaluation of conservation strategies, and other important ornithological goals (Dunn and Ralph 2004). Any biases in mist-net data may therefore have far-reaching consequences; and as Biro and Dingemanse (2009:66) observe, “most sampling techniques possess inherent sampling inefficiency and bias.”
Mist-netting avoids some biases present in other survey methods: compared to visual or aural surveys, mist-netting is less likely to miss small, quiet, and secretive species (Wang and Finch 2002, Dunn and Ralph 2004) and likely depends less upon the field identification skills of individual researchers (Karr 1981). However, mist-nets may not capture a representative sample of the local avifauna for several reasons. For example, (1) variable external conditions may impact capture rates (e.g., rainfall, Silkey et al. 1999; vegetation, Pagen et al. 2002, Wang and Finch 2002); (2) species vary in their capturability (MacArthur and MacArthur 1974, Silkey et al. 1999, Wang and Finch 2002); (3) sex, feeding strategy, or personality may drive inter-individual differences in capturability (Dufour and Weatherhead 1991, Vanderkist et al. 1999, Biro and Dingemanse 2009); and (4) individuals may vary over time in their susceptibility to capture due to changes in age, condition, molting status, status as a territory-holder or a transient, or previous capture experience (MacArthur and MacArthur 1974, Fitzgerald et al. 1989, Dufour and Weatherhead 1991, DomèNech and Senar 1997, Insley and Etheridge 1997, Bart et al. 1999, Silkey et al. 1999, Nur et al. 2004, Marques et al. 2013, Roche et al. 2013, LaBarbera and Scullen 2021). Quantifying potential biases such as these allows researchers to consider them when designing studies and interpreting data, and to eliminate or moderate their impacts on study results (Biro and Dingemanse 2009). There is, however, an additional potential bias in mist-netting that has thus far been studied little, rendering it difficult for researchers to account for: capture height bias.
Mist-nets are commonly set only at ground-level, where they capture birds no higher than ~3 m above the ground (Bonter et al. 2008, Vecchi and Alves 2015). Methods for achieving alternative net heights exist, e.g., Stokes et al. (2000), but appear to be rarely used. Mist-net height is such a consistent aspect of standard procedure that it goes generally unremarked; for example, in the widely used MAPS Manual, net height is only briefly mentioned (DeSante et al. 2020). However, concerningly, Remsen and Good (1996) demonstrated that even slight differences in the vertical activity patterns of birds could result in considerable differences in capture rate, as the birds became more or less likely to fly over rather than into the net. Because these vertical activity patterns are relative to the top of the net, even small differences in net height could produce differences in capture patterns.
Available evidence suggests that net height does influence capture patterns. It has been documented that ground-level mist-nets alone fail to adequately assess community composition, as they bias samples toward understory-favoring species (Fitzgerald et al. 1989, Remsen and Good 1996, Wang and Finch 2002, Derlindati and Caziani 2005, Bonter et al. 2008). For example, in the forest of Orongorongo Valley, New Zealand, height biases were found in 7 of 14 species captured in net rigs spanning 1.5–13.5 m above the ground (Fitzgerald et al. 1989). In the Atlantic rainforest of Ilha Grande, Brazil, “vertically-mobile” mist-nets that were positioned 0–17 m above the ground captured nearly 50% more species than standard ground-level mist-nets and resulted in an estimate of species richness nearly twice that of ground-level nets (Vecchi and Alves 2015). In NY state, USA, mist-nets elevated to ~3 m above ground level during migration captured 12 species not captured in ground-level nets, and > 50% of species were more likely to be captured at one net height than the other, with more species captured in ground-level nets (Bonter et al. 2008).
The results of these studies underscore the need to better understand the effects of mist-net height on capture patterns. They demonstrate that net height matters in different habitats (rainforest with a mean canopy height of 23 m vs. early secondary-growth habitats with mean vegetation heights of 5 m), species assemblages (tropical vs. temperate and mainland vs. island avifauna), and life history periods (tropical winter vs. temperate migrations), and that the specific effects of mist-net height on capture patterns vary with at least some of these variables (Fitzgerald et al. 1989, Bonter et al. 2008, Vecchi and Alves 2015). To understand the possible biases in data collected using only ground-level nets, research in other habitats and during more life history periods is needed. Crucially, it is not well-studied whether ground-level nets introduce biases into uses of banding data other than characterizing community composition, such as demographic studies, migration phenology, and population trends, although theoretical analysis suggests that they may (Remsen and Good 1996).
It is important to understand not only how sampling height impacts captures generally, but specifically how those impacts differ among species, individual attributes, and external conditions, so that researchers can determine whether sampling height may affect their specific research question. We expect that capture height bias will vary with other factors because capture height biases should be driven by avian behavior (MacArthur and MacArthur 1974, Remsen and Good 1996), which can vary by age, sex, personality, experience, territorial vs. non-territorial status, foraging guild, weather, and many other factors (Bart et al. 1999, Silkey et al. 1999, Nur et al. 2004, Marques et al. 2013, Roche et al. 2013, LaBarbera and Scullen 2021). Such interactions have been documented in comparisons of capturability among different trap types, where, for example, different trapping methods may capture different age or sex ratios (Dufour and Weatherhead 1991, DomèNech and Senar 1997). These effects can interact with yet additional factors, as when abundance estimates vary both with trap type and habitat (Pagen et al. 2002, Wang and Finch 2002), and when Insley and Etheridge (1997) observed that mist-nets under-captured adult Redshanks Tringa totanus relative to juveniles, but only in early autumn, because of age-specific molt phenology. The few data that exist on mist-net capture height biases support the existence of these interactions: multi-species studies found instances of the effect of mist-net height on capture probability interacting with foraging guild (Vecchi and Alves 2015), time of day, season, capture history (Fitzgerald et al. 1989), age, and sex (Bonter et al. 2008).
We examined a year-round, 26-year capture dataset from riparian woodland in the temperate Mediterranean climate of the south Bay Area, California, USA, to test for differences in avian capture probabilities between paired ground-level mist-nets and mist-nets elevated ~3 m above ground level. We further explored whether capture height bias interacted with species, capture history, age, sex, foraging guild, net location, and season. To our knowledge, this work represents the first test of a relationship between foraging height and capture height bias and the largest exploration of capture height biases across the full annual cycle.
We used constant-effort bird banding data from Coyote Creek Field Station (hereafter CCFS) in Milpitas, CA, USA, from 1993 to 2020. CCFS is located at the southern end of the San Francisco Bay (37.4381°N, 121.9285°W). The 12-ha area is divided among four habitats: a remnant riparian corridor that borders the creek, two areas of restored riparian woodland (replanted in 1987 and 1993, respectively), and a large grassy area that is managed as an overflow channel for flood control. Although ground-level mist-nets were operated in all these habitats, elevated mist-nets were operated only in the remnant riparian habitat. Within 5 m of the mist-nets, the dominant overstory plant species were box elder (Acer negundo), western sycamore (Platanus racemosa), Fremont cottonwood (Populus fremontii), and California buckeye (Aesculus californica), and the dominant understory plant species were coyote brush (Baccharis pilularis), California blackberry (Rubus ursinus), and poison hemlock (Conium maculatum). Vegetation surveys conducted in 1997, 2005, and 2013 found no changes over that time in the remnant riparian habitat in understory abundance or diversity, overstory abundance or diversity, tree height, tree diameter-at-breast-height, or canopy cover (San Francisco Bay Bird Observatory, unpublished data).
Three net arrays, composed of 14, 14, and 19 mist-nets, were operated on Wednesdays, Saturdays, and Sundays, respectively, year-round. Most of the nets in these regularly operated net arrays were standard ground-level nets; however, there were also two paired net rigs with two nets stacked on top of each other (one ground-level and one elevated), with one paired net rig operated on Saturdays and the other operated on Sundays. A pulley system was used to raise and lower the nets and safely retrieve birds from the upper net panels. Trammel spacing was standardized at 0.56 m such that the ground-level net spanned 0.50–2.74 m above the ground and the elevated net spanned 2.74–4.98 m above the ground. An additional paired net rig was operated 1993–1998 on multiple days each week, after which its use was discontinued. All nets were 12 m x 2.6 m made of polyester with 4 tiers and 30 mm mesh. Net sites were titled by their distance in meters upstream along Coyote Creek from the San Francisco Bay. The two regularly operated paired net sites were titled 9280 and 9590, and the discontinued paired net site was titled 8735 (Table 1). All three paired net rigs were in the mature riparian corridor bordering the creek.
Nets were opened ~30 min before sunrise and closed after five hours. Effort was recorded for each day where 1 net hour represented one 12 m x 2.6 m net open for 1 hr. Nets were checked every 30 min. Birds were banded with USGS aluminum bands and we recorded the capture date, time, and net for each bird. Age and sex of birds were determined when possible following Pyle (1997), and standard body condition, plumage, and morphological data were collected before birds were released. Recaptured individuals were similarly processed. Hummingbirds were banded only from 1993 to 1999, so the analyses of hummingbirds includes only those years.
Taxa were classified into the following height-based foraging guilds, similar to those used by DeGraaf and Wentworth (1986), based on a review of the “Diet and Foraging” sections of their Birds of the World accounts (Billerman et al. 2020): aerial (flycatchers), hovering (hummingbirds), ground (foragers on ground substrates), lower strata (foragers above the ground, in shrubs and subcanopy but below midcanopy), middle strata (foragers in tall shrubs and midcanopy), and upper strata (foragers in the canopy; see Table 2). Many species utilized multiple strata, in which cases all of those strata were included in that species’ classification; e.g., the Warbling Vireo Vireo gilvus “forages from just above ground to near top of canopy” (Gardali and Ballard 2020), so was classified as “lower to upper strata” (Table 2). We kept our classifications broad to avoid making overly specific determinations where significant variability may exist, or detailed data is lacking. When possible, we used foraging information specific to coastal California populations to make these classifications.
Bayesian analysis is well-suited to analyzing small sample sizes (Hox et al. 2012). However, determining the desired minimum sample size for Bayesian analyses is an area of ongoing research (Joseph and Bélisle 2019). For our analyses we aimed to balance conservatism with the value of utilizing data when possible. We targeted a credible interval length of ≤ 0.50, because that was the maximum interval length for which it would be possible to detect a height bias. For single proportion binomial models, we used a minimum sample size of 12, which was determined by examining the minimum sample size calculated using the three main approaches to Bayesian binomial sample size determination (average coverage criterion, average length criterion, and worst outcome criterion), with α = 0.05 and maximum credible interval length of 0.50, and then selecting the largest of those three calculated sample sizes (Joseph et al. 1995). For difference between two proportions binomial models, we used a minimum sample size of 28 for each group, which was similarly determined by calculating the minimum sample size using the three methods named above and with the same alpha and credible interval settings, and then choosing the largest of the three calculated sample sizes. In both cases, the largest sample sizes proved to be those calculated using the worst outcome criterion, which is generally found to be the most conservative (Joseph et al. 1995). There is no established method for calculating the minimum sample size for a Bayesian fixed effects analysis, so for the analysis of net location we used the minimum sample size for single proportion binomial models for each net location, i.e., each analyzed net location had a minimum sample size of 12. We additionally required a total minimum sample size of (12n*1.25), where n was the number of net locations analyzed, resulting in total minimum sample sizes of either 30 or 45, depending on whether the analysis included two or all three net locations.
We filtered our data to include the first capture of each individual at any of the three paired net rigs. This included newly banded birds as well as recaptures that were originally captured at non-paired (ground-level-only) net locations, which we refer to as “new recaptures.” We treated height of capture net (ground-level or elevated) at the paired net rig as a Bernoulli random variable. Single proportion binomial models were used to estimate the probability of ground-level capture for taxa that met the sample size criterion of n ≥ 12, where n = number of first captures. We added fixed effects for each net location (8735, 9280, and 9590) for all taxa with sufficient captures to assess whether capture height biases varied between locations. Difference between two proportions binomial models were used to estimate differences by capture history (newly banded or new recapture), age, sex, and season for all taxa that met the sample size criterion for each group of size n ≥ 28. To assess differences between sexes, only data from known-sex individuals, as determined by reliable plumage characters and/or breeding physiology, were included. Our assessment of age differences focused on captures from June through December where individuals of most taxa could reliably be separated into hatch year (HY) and after-hatch year (AHY) age classes. To examine seasonal patterns, captures were partitioned among winter (December–February), spring (March–May), summer (June–August), and fall (September–November).
All models, except for that concerning the feeding guilds, were built in a Bayesian framework using the BUGS language (Lunn et al. 2000). We estimated each parameter’s posterior distribution with three parallel Markov Chain Monte Carlo (MCMC) chains using Gibbs sampling in JAGS v. 4.3.0 (Plummer 2017) called from the package ‘jagsUI’ (Kellner 2019) in R v. 3.4.2 (R Core Team 2017). For the single and difference between two proportion binomial models, uninformative priors of the form were used. For binomial models assessing the fixed effects of each net location, uninformative priors with a normal distribution with a mean of 0 and variance of 1000 were used for the intercepts and slopes. Inspection of the trace plots and Gelman-Rubin diagnostic confirmed successful convergence of parameters (Gelman et al. 2004). For single proportion models, we considered results significant if the 95% credible interval did not overlap with 0.5. For the difference of two proportions models we considered results significant if the 95% credible intervals did not overlap with 0.
To test whether observed capture height biases were related to foraging guilds, we ran a generalized linear model of the proportion of captures that were in the elevated net for each species, weighted by the total number of captures for each species, with foraging guild as a fixed effect, and with a binomial distribution and a logit link.
We did not correct for multiplicities among our tests. Multiplicity corrections in Bayesian statistics is a growing field with no consensus on the appropriate analytical techniques (Westfall et al. 1997, Berry and Hochberg 1999, Sjölander and Vansteelandt 2019). Some have argued that such corrections are not necessary for many Bayesian applications (Gelmen et al. 2012, Sjölander and Vansteelandt 2019).
From 1993 to 2020, operation of the three paired net rigs resulted in 7312 captures of 81 taxa. At all three rigs, the elevated net captured slightly more birds than the corresponding ground-level net, but these differences were not significant (Table 1). Forty-three taxa met the sample size criteria for our single species models (Table 2). Capture height biases were detected in 20 of these taxa (Fig. 1), with seven more likely to be captured in ground-level nets and 13 more likely to be captured in elevated nets. The taxa with ground-level biases included two thrushes and three sparrows, while those with elevated net biases included three hummingbirds and two flycatchers. Interestingly, while both Yellow-rumped Warbler subspecies analyzed were biased toward the elevated nets, their 95% credible intervals for the probability of ground-level capture did not overlap. Audubon’s Yellow-rumped Warblers Setophaga coronata auduboni were more likely to be captured in the elevated nets than Myrtle Yellow-rumped Warblers S. c. coronata.
Of the 43 taxa analyzed for overall capture height biases, there were sufficient sample sizes to evaluate location effects at all three net locations (8735, 9280, and 9590) for 16 taxa and at the two regularly operated net locations (9280 and 9590) for an additional 12 taxa (Fig. 2, see Table A1.1 for sample sizes). There were no instances of taxa having opposite biases at different net locations. Of the 20 taxa with overall capture height biases, four (Mourning Doves Zenaida macroura, Black-chinned Hummingbirds Archilochus alexandri, Rufous Hummingbirds Selasphorus rufus, and Willow Flycatchers Empidonax traillii) did not have sufficient sample sizes to evaluate net location effects. Eight of the 16 taxa with overall capture height biases and sufficient sample size (Anna’s Hummingbirds Calypte anna, Bewick’s Wrens Thryomanes bewickii, Swainson’s Thrush Catharus ustulatus, Hermit Thrush Catharus guttatus, American Goldfinches Spinus tristis, Fox Sparrows Passerella iliaca, Song Sparrows Melospiza melodia, and Audubon’s Yellow-rumped Warblers) exhibited consistent biases at all net locations evaluated. For the other eight taxa with overall capture height biases and sufficient sample size (Downy Woodpeckers Dryobates pubescens, Western Flycatchers Empidonax difficilis/occidentalis, Chestnut-backed Chickadees Poecile rufescens, Golden-crowned Kinglets Regulus calendula, Golden-crowned Sparrows Zonotrichia atricapilla, Common Yellowthroats Geothlypis trichas, Yellow Warblers Setophaga petechia, and Myrtle Yellow-rumped Warblers), significant effects at 1–2 net locations drove the overall effect. Additionally, there were four taxa (Bushtits Psaltriparus minimus, Ruby-crowned Kinglets Regulus satrapa, Orange-crowned Warblers Leiothlypis celata, and Wilson’s Warblers Cardellina pusilla) that did not have overall capture height biases but had at least one net location with a significant effect on capture height bias.
We assessed differences in the probability of ground-level capture between newly banded and newly recaptured birds for 15 taxa (see Table A1.2 for summary statistics and Table A1.3 for sample sizes). We found significant differences for only the Myrtle Yellow-rumped Warblers, which had a higher probability of ground-level capture for newly banded birds than for newly recaptured birds. Both the newly banded and newly recaptured Myrtle Yellow-rumped Warblers were biased toward elevated nets, but the bias was significantly greater for recaptured birds.
Of five taxa with sufficient known age captures in the summer and fall (see Table A1.4 for summary statistics and Table A1.5 for sample sizes), only Swainson’s Thrush showed an age-related capture height bias: AHY birds were more likely to be captured in ground-level nets than HY birds. Of 11 taxa with sufficient known sex captures (see Table A1.6 for summary statistics and Table A1.7 for sample sizes), only Anna’s Hummingbirds showed a sex-related capture height bias, with females more likely to be captured in ground-level nets than males.
Capture height biases differed significantly among foraging guilds and broadly followed the expected height patterns (Fig. 3, Table 3). Ground-level biases were detected only in taxa belonging to the ground and ground to lower foraging guilds. Biases toward capture in elevated nets were detected in most taxa belonging to the aerial, hovering, and middle to upper foraging guilds, and zero taxa belonging to the ground to lower foraging guild. However, contrary to expectation, biases toward capture in elevated nets were also detected in two taxa belonging to the ground foraging guild (American Goldfinches and Mourning Doves). The ground and ground to lower guilds were less likely to be captured in elevated nets than the lower to upper guild, and the middle to upper and hovering guilds were more likely to be captured in elevated nets than the lower to upper guild. The aerial guild did not differ from the lower to upper guild.
We assessed seasonal differences for 18 taxa that met the sample size criteria for at least two seasons (see Table A1.8 for summary statistics and Table A1.9 for sample sizes). Only seasons for which there was sufficient sample size were assessed. We found significant differences for five taxa: (1) Anna’s Hummingbirds were more likely to be captured in ground-level nets in the winter compared the fall; (2) Ruby-crowned Kinglets were more likely to be captured in ground-level nets in the winter compared to the spring; (3) Swainson’s Thrush were more likely to be captured in ground-level nets in the spring compared to the fall; and both (4) Song Sparrows and (5) Common Yellowthroats were more likely to be captured in ground-level nets in the spring compared to the summer.
Almost half the taxa in this study (46.5%) exhibited capture height biases, with 13 taxa biased toward capture in elevated nets and seven taxa biased toward ground-level nets. We found evidence that foraging height impacted capture height bias, and that biases differed among seasons. There was less support for differences in capture height bias by net location, capture history, age, or sex, although a few taxa did exhibit these effects. Together, these results suggest that using only ground-level mist-nets has greater potential to introduce bias than previously thought.
Although other investigators have grouped species in feeding guilds when analyzing methodological biases (see Vecchi and Alves 2015 on capture height, Gilbert et al. 2021 on visibility bias in waterbird surveys, and Vold et al. 2017 on comparisons between visual and acoustic surveys), to our knowledge this is the first study that has tested for capture biases based on foraging height guilds. As expected, the probability of capturing individuals of each guild in ground-level nets broadly reflected the guilds’ vertical foraging strata. The middle to upper and hovering guilds were significantly more biased toward elevated capture than the lower to upper guild, which in turn was significantly more biased toward elevated capture than the ground to lower and ground guilds. The aerial guild was not different from the ground to upper guild, which may be related to the fact that its component species—all flycatchers—often forage from low perches (Billerman et al. 2020). Unexpectedly, two ground foraging taxa, Mourning Doves and American Goldfinches, had biases toward capture by elevated nets. This discrepancy highlights the limitations of attributing capture height biases to foraging habits alone: foraging is only one of many circumstances that may lead to capture in a mist net (Bonter et al. 2008, LaBarbera and Scullen 2021). Nevertheless, we found that foraging guilds do reflect the general patterns of capture height biases, suggesting that foraging guild classification could be a good approximation for the height at which a taxon most often encounters a mist net. Researchers may be able to use known foraging height as a predictor of overall capture height bias, which could be valuable when designing mist-net sampling schemes.
Comparison of our species-level results to those of Bonter et al. (2008) shows greater consistency in ground-level biases than biases toward elevated nets. Of 14 species analyzed in both papers, we detected opposite biases for only two: Willow Flycatchers (analyzed by Bonter et al. 2008 as Traill’s Flycatcher Empidonax alnorum/traillii, which may also include Alder Flycatchers, a sister species difficult to distinguish from Willow Flycatcher in the hand and that share similar habitat preferences and life histories) and Yellow Warblers were biased toward elevated nets in our study but ground-level nets in Bonter et al. (2008). We detected the same ground-level biases for four taxa (Swainson’s Thrush, Hermit Thrush, Common Yellowthroats, and Song Sparrows). None of the five taxa identified in our study as having biases toward elevated nets that were also analyzed by Bonter et al. (2008) had such a bias in that paper. Unfortunately, none of the taxa with elevated biases in Bonter et al. (2008) were analyzed in our study. Although limited by small sample size, these comparisons suggest biases toward elevated net capture may be more variable than those for the ground-level, at least between these two study sites. This may reflect habitat structure: in temperate woodland, the height of the ground and ground-level features (such as herbaceous vegetation) are probably more consistent than overstory vegetation structure. Bonter et al. (2008) report that the top of their elevated net rigs extended above the canopy at their site, whereas the canopy at our study site extended above the tops of our elevated net rigs by ~4 m. Bonter et al. (2008) speculate that the resulting increased sun and wind exposure of their elevated nets, in comparison to their ground-level nets, contributed to increased net avoidance behavior at the elevated net level. In contrast, our elevated nets remained under the protection of the canopy vegetation. The data are consistent with this proposed effect of relative canopy height: although the captures in Bonter et al. (2008) were substantially biased toward the ground-level nets, we did not find a difference in overall captures between our two net heights. Comparison between only two studies is inherently limited, however. The variation between these studies highlights the need for wider usage and analysis of paired net rigs in different habitats.
Capture biases were largely consistent among our three net locations: of 28 taxa, none exhibited opposing capture height biases across locations. We know of only one other study that looked for differences in capture height biases between net locations (Fitzgerald et al. 1989), which similarly found that most species showed consistent patterns between net sites within the same habitat. This consistency is to be expected of nets located in the same habitat with the same bird community if capture height bias is driven by factors such as habitat and taxon. This low net-to-net variation also suggests that researchers may be able to quantify capture height bias with a small number of nets and safely generalize from those. In our analyses, four taxa did not exhibit biases overall, but did at specific net-pairs: Bushtits, Ruby-crowned Kinglets, Orange-crowned Warblers, and Wilson’s Warblers. This may be due to very small-scale geographic variation, such as specific habitat features or differences in perceived predation risk that alter the height at which these birds fly (Cimprich et al. 2005). Alternatively, the net-to-net variation could be due to varying sample sizes at different nets resulting in some net locations having less power to detect small biases than others. Future work could place paired net rigs within multiple habitat types and test whether habitat features such as canopy height or understory height impact net-to-net variation in capture height biases.
We found little support for an effect on capture height bias of individual capture history, i.e., whether the bird had been previously caught (in a standard ground-level net, not a paired net rig): of 14 taxa, only one (7.1%) had a significant difference. This differs from Fitzgerald et al.’s (1989) observation of capture history-based height differences in more than one-quarter of their focal species. Our low rate of observed differences based on capture history suggests that there is a low risk of capture history confounding capture height bias analyses, and vice versa. We found that in Myrtle Yellow-rumped Warblers, newly banded birds were more likely to be captured in ground-level nets than recaptures. Possible explanations for this apparent upward movement of recaptured vs. newly-captured birds include (1) a net avoidance behavior (Marques et al. 2013, Roche et al. 2013) where birds previously captured at the ground level alter their activity patterns to higher habitat strata, and/or (2) a transient effect (Fitzgerald et al. 1989) where transient individuals, which are over-represented among newly banded birds (Nur et al. 2004), may move through the habitat at a lower height than resident or overwintering individuals, which are over-represented among recaptures. Net avoidance and cryptic differences between transient and (relatively) resident individuals represent important challenges in analyzing bird banding data (MacArthur and MacArthur 1974, Remsen and Good 1996, Nur et al. 2004, Roche et al. 2013), particularly when modeling survival. Although our focal taxa showed little association between capture history and capture height bias, Fitzgerald et al.’s (1989) analysis suggests that such associations can occur with moderate frequency. For taxa in which they do occur, the selection of net height could be a useful tool for researchers wishing to focus primarily on either resident or transient individuals. For example, our results suggest that a project interested in resident (over-wintering) Myrtle Yellow-rumped Warblers would be well-served by prioritizing elevated over ground-level nets.
Capture height biases at the taxon level may not be problematic for data analyses so long as the sample of the population at the ground-level remains representative of the total population available for capture. Capture height biases are more troubling if they covary with some variable of interest. Remsen and Good (1996) speculated that ground-level nets may bias samples in demographic studies if species differ in vertical habitat use by age or sex. With respect to age and sex, which we were able to analyze for 5 and 11 taxa, respectively, we found only one instance for each variable where the use of only ground-level nets would impact capture data: female Anna’s Hummingbirds and after-hatch-year Swainson’s Thrush were more likely to be captured at the ground-level than their male or hatch-year counterparts, respectively. For Anna’s Hummingbirds, this difference may be explained by the males’ aerial display, which involves the bird rising up to 35 m above the ground and then plunging in a rapid dive (Clark and Russell 2020). It is unknown why after-hatch-year Swainson’s Thrush were captured at lower heights than hatch-years, but after-hatch-year Swainson’s Thrush possess substantially higher fat stores during fall migration than hatch-years (Woodrey and Moore 1997), suggesting age differences in foraging strategy or efficiency. Broadly, our results suggest demographic differences in capture height biases are uncommon, as for both age and sex they were present in < 20% of taxa analyzed in this study. This is consistent with Bonter et al.’s (2008) migration season findings, although they too were only able to analyze age in < 10 taxa and sex in < 15 taxa. They speculated that sex might impact capture height bias more commonly during the breeding season, but seven of our eleven analyzed taxa breed at CCFS, and only one showed an effect of sex. More data on a broader variety of taxa and from more geographic regions are needed before the potential confounding impacts of demographic differences in capture height bias can be reasonably disregarded.
Seasonal changes in vertical habitat use may also bias captures and could be of particular concern at migration monitoring stations (Hutto 1985, Fitzgerald et al. 1989, Bonter et al. 2008). Of 15 taxa analyzed, 5 showed significant seasonal effects on capture height, including residents (Anna’s Hummingbirds and Song Sparrows), short-distance migrants (Ruby-crowned Kinglets and Common Yellowthroats), and a long-distance migrant (Swainson’s Thrush). Swainson’s Thrush had a greater ground-level bias in the spring than in the fall, which is consistent with capture height bias shifts between spring and fall migration observed in four other species (Bonter et al. 2008). Bonter et al. (2008) suggest that this may be due to reduced vegetative cover at higher strata in the spring compared to the fall. Song Sparrows, which are resident at our study site year-round, were captured higher in the summer than they were in the spring or the fall. Common Yellowthroats, which primarily breed at our study site (though some individuals overwinter), were also captured higher in the summer than in the spring. Ruby-crowned Kinglets, which overwinter at our study site, were captured higher in the spring than the winter, in contrast to the opposite shift observed in two New Zealand species by Fitzgerald et al. (1989). Seasonal shifts may reflect changes in food resources, usage of higher strata by dispersing birds, or differences in habitat use related to breeding (Hutto 1985, Streby et al. 2014). Seasonal changes in capture height bias are likely to be habitat-specific, because they are likely impacted by habitat structure, food resources, and interspecies interactions (Hutto 1985, Bonter et al. 2008); therefore, researchers should use caution when extrapolating results from one study area to another.
Capture height biases at our study site were common and varied among taxa, feeding guilds, and seasons. These biases were largely not affected by net location, capture history, or demography, although there were exceptions. Further research should test for capture height biases in other taxa and geographic locations, investigate the mechanisms that produce variation in capture height biases, and further consider how to improve mist-netting methods to avoid any confounding impacts of these patterns. We urge the avian monitoring community to consider issues of capture height and, where possible, deploy mist-nets in a manner that allows such effects to be detected. For example, as a result of this study Coyote Creek Field Station has constructed additional paired net rigs in different habitat types and started recording the height of the net panel in which each bird is captured. Mist-netting is one of the most important methods for avian research (Saracco et al. 2008), and therefore investment in validating and improving mist-netting protocols will improve the data underlying widespread efforts toward research and conservation.
The authors DJT and KL were sub-permitted banders for the Coyote Creek Field Station. DJT developed the idea for this study and collaborated with KL to build research questions and design analyses. DJT analyzed the data. Both authors wrote, edited, and gave final approval for the manuscript.
Josh Scullen and Dan Wenny gave important feedback throughout the development of this study. Dozens of volunteers contributed to data collection. Michelle Stantial assisted with model-building. No funders had input into the content of the manuscript. CCFS protocols follow the Guidelines to the Use of Wild Birds in Research (Fair et al. 2010) as well as best practices established by the North American Banding Council. Banding is authorized under federal permit #22109.
All bird banding data from under our Federal permit #22109 is publicly available by request from the USGS. Upon publication we will make our code and models available on the Dryad archiving service.
Bart, J., C. Kepler, P. Sykes and C. Bocetti. 1999. Evaluation of mist-net sampling as an index to productivity in Kirtland’s Warblers. Auk 116:1147-1151. https://doi.org/10.2307/4089698
Berry, D. A., and Y. Hochberg. 1999. Bayesian perspectives on multiple comparisons. Journal of Statistical Planning and Inference 82:215-227. https://doi.org/10.1016/S0378-3758(99)00044-0
Billerman, S. M., B. K. Keeney, P. G. Rodewald, and T. S. Schulenberg, editors. 2020. Birds of the world. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/bow
Biro, P. A., and N. J. Dingemanse. 2009. Sampling bias resulting from animal personality. Trends in Ecology & Evolution 24:66-67. https://doi.org/10.1016/j.tree.2008.11.001
Bonter, D. N., E. W. Brooks, and T. M. Donovan. 2008. What are we missing with only ground-level mist-nets? Using elevated nets at a migration stopover site. Journal of Field Ornithology 79:314-320. https://doi.org/10.1111/j.1557-9263.2008.00179.x
Cimprich, D. A., M. S. Woodrey, and F. R. Moore. 2005. Passerine migrants respond to variation in predation risk during stopover. Animal Behaviour 69:1173-1179. https://doi.org/10.1016/j.anbehav.2004.07.021
Clark, C. J., and S. M. Russel. 2020. Anna’s Hummingbird (Calypte anna), version 1.0. In A. F. Poole, editor. Birds of the world. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/bow.annhum.01
DeGraaf, R. M., and J. M. Wentworth. 1986. Avian guild structure and habitat associations in suburban bird communities. Urban Ecology 9:399-412. https://doi.org/10.1016/0304-4009(86)90012-4
Derlindati, E. J., and S. M. Caziani. 2005. Using canopy and understory mist-nets and point counts to study bird assemblages in Chaco forests. Wilson Bulletin 117:92-99. https://doi.org/10.1676/03-063
DeSante, D. F., K. M. Burton, P. Velez, D. Froehlich, D. Kaschube, and S. Albert. 2020. MAPS manual: 2020 protocol. Instructions for the establishment and operation of constant-effort bird-banding stations as part of the monitoring avian productivity and survivorship program. The Institute for Bird Populations, Petaluma, California, USA.
DeSante, D. F., and D. R. Kaschube. 2009. The monitoring avian productivity and survivorship (MAPS) program 2004, 2005, and 2016 report. Bird Populations 9:86-269.
DeSante, D. F., D. R. O’Grady, and P. Pyle. 1999. Measures of productivity and survival derived from standardized mist-netting are consistent with observed population changes. Bird Study 46:178-188. https://doi.org/10.1080/00063659909477244
DomèNech, J., and J. C. Senar. 1997. Trapping methods can bias age ratio in samples of passerine populations. Bird Study 44:348-354. https://doi.org/10.1080/00063659709461070
Dufour, K. W., and P. J. Weatherhead. 1991. A test of the condition-bias hypothesis using Brown-headed Cowbirds trapped during the breeding season. Canadian Journal of Zoology 69:2686-2692. https://doi.org/10.1139/z91-377
Dunn, E. H., and C. J. Ralph. 2004. The use of mist-nets as a tool for bird population monitoring. Studies in Avian Biology 29:1-6.
Fair, J., E. Paul, and J. Jones, editors. 2010. Guidelines to the use of wild birds in research. Ornithological Council, Washington, D.C., USA.
Fitzgerald, B. M., H. A. Robertson, and A. H. Whitaker. 1989. Vertial distribution of birds mist-netted in a mixed lowland forest in New Zealand. Notornis 36:311-321.
Gardali, T. and G. Ballard. 2020. Warbling Vireo (Vireo gilvus), version 1.0. In A. F. Poole and F. B. Gill, editors. Birds of the world. Cornell Lab of Ornithology, Ithaca, New York, USA. https://doi.org/10.2173/bow.warvir.01
Gelman, A., J. B. Carlin, H. S. Stern, and D. B. Rubin. 2004. Bayesian data analysis. Second edition. Chapman and Hall/CRC, Boca Raton, Florida, USA.
Gelman, A., J. Hill, and M. Yajima. 2012. Why we (usually) don’t have to worry about multiple comparisons. Journal of Research on Educational Effectiveness 5:189-211. https://doi.org/10.1080/19345747.2011.618213
Gilbert, A. D., C. N. Jacques, J. D. Lancaster, A. P. Yetter, and H. M. Hagy. 2021. Visibility bias of waterbirds during aerial surveys in the nonbreeding season. Wildlife Society Bulletin 45:6-15. https://doi.org/10.1002/wsb.1150
Hox, J., R. van de Schoot, and S. Matthijsse. 2012. How few countries will do? Comparative survey analysis from a Bayesian perspective. Survey Research Methods 6:87-93. https://doi.org/10.18148/srm/2012.v6i2.5033
Hutto, R. L. 1985. Seasonal changes in the habitat distribution of transient insectivorous birds in southeastern Arizona: competition mediated? Auk 102:12-132. https://doi.org/10.2307/4086827
Insley, H., and B. Etheridge. 1997. Catching bias in cannon and mist netted samples of Redshanks Tringa totanus on the Inner Moray Firth. Ringing & Migration 18:70-77. https://doi.org/10.1080/03078698.1997.9674144
Joseph, L., and P. Bélisle. 2019. Bayesian consensus-based sample size criteria for binomial proportions. Statistics in Medicine 38:4566-4573. https://doi.org/10.1002/sim.8316
Joseph, L., D. B. Wolfson, and R. Du Berger. 1995. Sample size calculations for binomial proportions via highest posterior density intervals. Statistician 44:143-154. https://doi.org/10.2307/2348439
Karr, J. R. 1981. Surveying birds with mist-nets. Studies in Avian Biology 6:62-67.
Kellner, K. 2019. jagsUI: A wrapper around “rjags” to streamline “JAGS” analyses.
LaBarbera, K., and J. C. Scullen. 2021. Using individual capture data to reveal large-scale patterns of social association in birds. Journal of Ornithology 162:795-811. https://doi.org/10.1007/s10336-021-01863-3
Larsen, R. J., K. A. Boegler, H. H. Genoways, W. P. Masefield, R. A. Kirsch, and S. C. Pedersen. 2007. Mist netting bats, species accumulation curves, and the rediscovery of two bats on Montserrat (Lesser Antilles). Acta Chiropterologica 9:423-435. https://doi.org/10.3161/1733-5329(2007)9[423:MNBSAC]2.0.CO;2
Lunn, D. J., A. Thomas, N. Best, and D. Spiegelhalter. 2000. WinBUGS - A Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing 10:325-337. https://doi.org/10.1023/A:1008929526011
MacArthur, R. H., and A. T. MacArthur. 1974. On the use of mist nets for population studies of birds. Proceedings of the National Academy of Sciences 71:3230-3233. https://doi.org/10.1073/pnas.71.8.3230
Marques, J. T., M. J. Ramos Pereira, T. A. Marques, C. D. Santos, J. Santana, P. Beja, and J. M. Palmeirim. 2013. Optimizing sampling design to deal with mist-net avoidance in Amazonian birds and bats. PLoS ONE 8(9):e74505. https://doi.org/10.1371/journal.pone.0074505
Nur, N., G. R. Geupel and G. Ballard. 2004. Estimates of adult survival, capture probability, and recapture probability: evaluating and validating constant-effort mist-netting. Studies in Avian Biology 29:63-70.
Pagen, R. W., F. R. Thompson III, and D. E. Burhans. 2002. A comparison of point-count and mist-net detections of songbirds by habitat and time-of-season. Journal of Field Ornithology 73:53–59. https://doi.org/10.1648/0273-8570-73.1.53
Plummer, M. 2017. JAGS: Just Another Gibbs Sampler. [online] URL: https://sourceforge.net/projects/mcmc-jags/
Pyle, P. 1997. Identification guide to North American birds Part I. Slate Creek, Bolinas, California, USA.
R Core Team. 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing Vienna, Austria.
Remsen Jr., J. V., and D. A. Good. 1996. Misuse of data from mist-net captures to assess relative abundance in bird populations. Auk 113:381-398. https://doi.org/10.2307/4088905
Roche, E. A., C. R. Brown, M. B. Brown, and K. M. Lear. 2013. Recapture heterogeneity in Cliff Swallows: increased exposure to mist nets leads to net avoidance. PLoS ONE 8:e58092. https://doi.org/10.1371/journal.pone.0058092
Saracco, J. F., D. F. Desante, and D. R. Kaschube. 2008. Assessing landbird monitoring programs and demographic causes of population trends. Journal of Wildlife Management 72:1665-1673. https://doi.org/10.2193/2008-129
Silkey, M., N. Nur, and G. R. Geupel. 1999. The use of mist-net capture rates to monitor annual variation in abundance: a validation study. Condor 101:288-298. https://doi.org/10.2307/1369992
Sjölander, A., and S. Vansteelandt. 2019. Frequentist versus Bayesian approaches to multiple testing. European Journal of Epidemiology 34:809-821. https://doi.org/10.1007/s10654-019-00517-2
Spotswood, E. N., K. R. Goodman, J. Carlisle, R. L. Cormier, D. L. Humple, J. Rousseau, S. L. Guers, and G. G. Barton. 2012. How safe is mist netting? Evaluating the risk of injury and mortality to birds. Methods in Ecology and Evolution 3:29-38. https://doi.org/10.1111/j.2041-210X.2011.00123.x
Stokes, A. E., B. B. Schultz, R. M. Degraaf, and C. R. Griffin. 2000. Setting mist nets from platforms in the forest canopy. Journal of Field Ornithology 71:57-65. https://doi.org/10.1648/0273-8570-71.1.57
Streby, H. M., J. M. Refsnider, S. M. Peterson, and D. E. Anderson. 2014. Retirement investment theory explains patterns in songbird nest-site choice. Proceedings of the Royal Society B: Biological Sciences 281:20131834. https://doi.org/10.1098/rspb.2013.1834
Vanderkist, B. A., X.-H. Xue, R. Griffiths, K. Martin, W. Beauchamp, and T. D. Williams. 1999. Evidence of male-bias in capture samples of Marbled Murrelets from genetic studies in British Columbia. Condor 101:398-402. https://doi.org/10.2307/1370004
Vecchi, M. B., and M. A. S. Alves. 2015. Bird assemblage mist-netted in an Atlantic Forest area: a comparison between vertically-mobile and ground-level nets. Brazilian Journal of Biology 75:742-751. https://doi.org/10.1590/1519-6984.00914
Vold, S. K., C. M. Handel, and L. B. McNew. 2017. Comparison of acoustic recorders and field observers for monitoring tundra bird communities. Wildlife Society Bulletin 41:566-576. https://doi.org/10.1002/wsb.785
Wang, Y., and D. M. Finch. 2002. Consistency of mist-netting and point counts in assessing landbird species richness and relative abundance during migration. Condor 104:59-72. https://doi.org/10.1093/condor/104.1.59
Westfall, P. H., W. O. Johnson, and J. M. Utts. 1997. A Bayesian perspective on the Bonferroni adjustment. Biometrika 84:419-427. https://doi.org/10.1093/biomet/84.2.419
Woodrey, M. S., and F. R. Moore. 1997. Age-related differences in the stopover of fall landbird migrants on the coast of Alabama. Auk 114:695-707. https://doi.org/10.2307/4089289