Comparing waterfowl densities detected through helicopter and airplane sea duck surveys in Labrador, Canada

,


INTRODUCTION
Most sea duck species winter in coastal waters and breed in the northern boreal forest and tundra (Johnsgard 2010).They are the most data deficient group of waterfowl in North America (Caithamer et al. 2000, Sea Duck Joint Venture (SDJV) 2003) because they breed in remote locations and spend much of their annual lifecycle at sea where they are difficult to study (Zipkin et al. 2010, Flint 2013).Of the fifteen North American species, the population trends of seven species are unknown, and managers have high confidence in the trends for only four species (Bowman et al. 2015).There is, however, evidence that ten of the fifteen species have declined from historic levels and scoter species are of particular concern (SDJV 2003, 2007, Bowman et al. 2015).Determining the current population sizes, trends, and trajectories of sea duck populations is, therefore, imperative to developing effective management strategies (Roy et al. 2019).
For many species of waterfowl in North America, the Waterfowl Breeding Population and Habitat Survey (WBPHS; Smith 1995) and Eastern Waterfowl Survey (EWS; Zimmerman et al. 2012) provide the data to estimate population abundances and trends.However, these surveys do not provide adequate information to monitor scoters and scaups because they only cover part of the breeding range and are conducted prior to the breeding season for these species (Bowman et al. 2015, Roy et al. 2019).Additionally, the fixed-wing components of these surveys suffer from low and variable detection rates and observers are unable to discriminate between species for some groups, resulting in generic abundance and trend estimates for the three scoter, two scaup, and two merganser species (SDJV 2003, 2007, Bowman et al. 2015, Schummer et al. 2018).Helicopter or ground surveys have been used to calculate Visibility Correction Factors (VCF) to adjust fixed-wing estimates upward in an attempt to correct for low detection (Koneff et al. 2008, Zimmerman et al. 2012).However, these methods do not directly assess detection of the fixed-wing aircraft but rather calculate VCFs based on differences in counts from the fixed-wing and ground or helicopter surveys.These methods assume perfect detection of helicopters and ground counts and, though flawed, are the basis of all breeding population estimates for the WBPHS and EWS and drive international waterfowl management decision-making.
Improving population estimates for the Surf Scoter (Melanitta perspicillata), Black Scoter (M.americana), and White-winged Scoter (M.deglandi) is of particular interest because these species are harvested and likely declining.Data from the WBPHS suggests that scoter populations in North America decreased from the 1980s to 2000s, although they may have begun to recover in 2004 (Bowman et al. 2015).In the Traditional Survey Area of the WBPHS, scoters declined most quickly from 1977 to 1989, possibly due to ocean conditions in the North Pacific (Flint 2013).Unfortunately, the limited ability to distinguish between scoter species means the WBPHS provides only limited insight into the population dynamics of individual scoter species.Other data sources suggest that, although it is difficult to assess their true population size, White-winged Scoter and Surf Scoter populations are both below historic levels (Nisbet et al. 2013, Bowman et al. 2015).Black Scoters have recently been divided into two population units (Bowman et al. 2021, Lamb et al. 2021).The Pacific population is thought to be increasing but is still below historic levels (Stehn, unpublished manuscript), while the Atlantic population's status is unknown (Bowman et al. 2015).
Scoters and other sea ducks are important components of the subsistence harvest of northern Indigenous people and remain popular with Atlantic and Pacific coast waterfowl hunters (Rothe et al. 2015), but setting sustainable harvest levels for licensed hunting is difficult without reliable indices of abundance and trend (Caithamer et al. 2000, Koneff et al. 2017).There is currently considerable uncertainty in establishing allowable harvest for scoters and other sea ducks for licensed hunters.Current harvest levels might leave some populations, such as the Eastern population of Surf Scoters, at risk of overharvest (Koneff et al. 2017).While it entails some risks, calculating the maximum sustainable harvest without knowing the exact population size may be possible if the demography and life history of the species is well known (Niel andLebreton 2005, Koneff et al. 2017).However, without reliable population estimates it is impossible to determine the trajectory of the populations and assess whether the current harvest management strategy is effective; therefore, improving population estimates should be a top priority for wildlife managers.
In 2009, the Canadian Wildlife Service ran a trial survey in central Labrador to evaluate the feasibility of conducting specialized surveys for breeding sea ducks.There were two concurrent components of this survey: a helicopter plot survey and a fixedwing line-transect survey.The goals of this trial survey were to determine whether a mid-June survey would adequately capture the breeding season of sea ducks, assess detection rates, and develop survey methodology that integrates fixed-wing and helicopter platforms.Because waterfowl can be difficult to detect from the air, we must assess and then account for waterfowl detection from both survey platforms to accurately estimate abundance.To take advantage of differences between the survey platforms, the helicopter plot survey used a double-observer method to evaluate detection (Cook and Jacobson 1979, Nichols et al. 2000, Roy et al. 2021) and the fixed-wing line-transect survey used a distance method (Buckland et al. 2015).
Our primary goal was to develop methodologies for monitoring waterfowl in remote regions of the northern boreal forest.Through time, there will be changes in the type, make and model of aircraft, sampling design, pilots and observers, and the possible inclusion of remotely sensed imagery in the future, all of which can introduce biases in abundance estimates.Effective long-term monitoring programs must be able to accommodate biases introduced by changes to the survey's design.One of the major sources of error that is expected to vary through time is the probability of detection of groups of birds.Here, we present results from the fixed-wing component of the trial scoter surveys.We used a simple distance-sampling approach within a hierarchical-model framework to correct for imperfect detection.Our hierarchical-model framework draws species-specific parameters from community distributions using random effects, which allows the overall community data to improve estimates for uncommon species (Sollmann et al. 2016).We were specifically interested in whether correcting for detection allowed density estimates from the fixed-wing line-transect surveys to be directly comparable to those from the helicopter plot survey.We thus compared the density and detection from this study to those from the concurrent helicopter plot survey (Roy et al. 2021).

Survey Area
The survey covered approximately 24,000 km² in central Labrador and the upper regions of the George River watershed in Quebec (Fig. 1).This area is primarily subarctic boreal forest with flat to rolling plateaus, open lichen woodlands and extensive wetlands at low elevations, and subarctic tundra with extensive areas of alpine barrens dominated by rock and open ground with patches of mosses and lichens at higher elevations (Roberts et al. 2006).Waterfowl are typically found in the many wetlands, lakes, rivers, ponds, and peatlands in the area (Bateman et al. 2017).

Survey Methods
We adapted standard USFWS fixed-wing waterfowl survey protocols (Smith 1995) to incorporate distance-sampling methodology.The survey consisted of 12 line transects (mean 140.8 km, range 129.7-150.5 km) oriented latitudinally (Fig. 1).Surveys were conducted on 17-19 June 2009 from a USFWS Partenavia (P.68C) fixed-wing aircraft equipped with a radar altimeter and GPS navigation system.The fixed-wing flew at 170-200 km/hr and followed the transect lines at a constant altitude of 45 m.Pilots were instructed to not fly when there was moderate to high turbulence or the crabbing angle was greater than 30°b ecause this would affect distance estimates.
The fixed-wing crew consisted of a pilot and two observers positioned in the front left and rear right of the aircraft.The observers remained in one position for the duration of the survey, which confounds the effect of observer position and observer identity.Each observer was responsible for recording the species, sex, location, and number of individuals for each waterfowl sighting using USFWS-GPS-Voice recording software (J.Hodges, USFWS).They recorded sightings independently for different sides of the aircraft.
Observers estimated each group's perpendicular distance from the line transect.To minimize the error and time required to measure the distances, observers binned distance measurements into four distance intervals: 14m-98m, >98m-171m, >171m-259m, and >259-523m from the transect line.We excluded the area 0-14m from the transect line because this area was directly under the aircraft and, therefore, unobservable.We calculated the cut points of the distance intervals assuming a half-normal detection function (Buckland et al. 2001) and then adapted these cut points slightly to fall on 5° increments of the inclinometer.
The observers recorded into which distance interval each sighting of birds fell (Appendix 1, for more details on distance binning procedure).They also recorded if the sighting fell on the boundary between distance intervals and we later randomly assigned half of the sightings on the boundary to one distance interval and half to the other.When possible, observers visually determined the sex (male, female, unknown) and species of all birds, excluding the scaups (Aytha spp) which were recorded to genus.

Fixed-wing Distance Detection Analysis
We adapted a Bayesian hierarchical distance model from Sollmann et al. (2016) to analyze the data collected during the fixed-wing aircraft survey.The model breaks the survey data into four interrelated components: 1) detection rates, 2) the true number of groups, 3) the size of each group and, 4) the sex composition of each group.Due to the complexity of the model, we used a multi-stage strategy and first identified the variables that influenced the detection of the groups before running the fully parametrized model (Morin et al. 2020).

Group Detection
For each group of birds observed, we modeled probability of detection as a function of the perpendicular distance of the transect line (x) using a half-normal function : Where σ is the scale parameter of the half-normal function.Given that our data were binned we evaluated the probability of detecting a group (p) in distance class k using the integral: Where v k is the width of the distance class k.We evaluated the probability that each observation fell in a given distance class via a categorical distribution where the vector of probability for each distance class is: Where ψ k is the proportion of the sampled area covered in the distance class k.
For the group detection sub-component, we initially considered six different possible detection models (Table 1).We modeled the effect of explanatory variables on the detection function on the log scale: In our most basic model, we held detection constant across all species.We considered a model that allowed the detection function to vary only as a function of the species, treating species as a random intercept drawn from a normal distribution with a mean across species of μ and standard deviation σ.We also considered a model where, in addition to the species, lone birds would have a different detection probability than groups of any other size (Royle 2008).The next model still included the effect of species but considered that detection would change linearly as a function of the size of the group, as other surveys have found that large groups can be easier to detect (e.g., Clement et al. 2017, Pearse et al. 2008).Additionally, we considered a detection model that included the effect of the observer because the detection function can vary between observers and as a function of observer experience (Fleming and Tracy 2008) and/or position in the aircraft.Finally, we considered detection models that included the additive effects of the species, the size of the group, and the observer (Table 1).We compared the six detection functions using the Watanabe-Akaike Information Criterion (WAIC) and used the detection function with the lowest WAIC in the fully parameterized model.

2) Group Density
We derived the total detection probability for a given species i from the detection function and estimated the true abundance of bird groups (flocks) from the total number of groups seen for a given species in a transect j to via a binomial distribution.
Flocks i , j ~Poisson( λ i , j × ρ i , j ) ρ i , j ~Gamma (r i , r i ) (1) ρ i , j ~Gamma (r i , r i ) The USFWS typically fly 200 m wide strip transects for their aerial surveys.To provide a useful metric of detection for the USFWS methodology, we also derived the species-specific detection rates for the inner 200 m transect by integrating the species-specific detection functions across the 0-200m strip transect.
The true number of groups of species i in transect j is a latent variable that is drawn from via a compound Gamma-Poisson distribution (i.e., a negative binomial distribution) to account for over-dispersion: Flocks i , j ~Poisson( λ i , j × ρ i , j ) ρ i , j ~Gamma (r i , r i ) Flocks seen i , j ~Binomial (Flocks i , j , p T i ) Flocks i , j ~Poisson( λ i , j × ρ i , j ) ρ i , j ~Gamma (r i , r i ) ρ i , j ~Gamma (r i , r i ) where λ i,j is the expected numbers of groups for species i on transect j and ρ i,j is the over-dispersion term for species i on transect j. λ i,j was modeled as depending on the length of transect j with a species-specific random-effect intercept (μ i drawn from a normal distribution with a shared mean and standard deviation. Following Sollmann et al. (2016), we used a data-augmentation approach to calculate the number of sightings that would have been made had the strip transect directly below the aircraft been observable.

3) Group Size
We also used a Gamma-Poisson distribution to model group size, with random effects for species modified from Sollman et al. (2016).Because our initial analysis of the detection subcomponent showed that group size did not affect detection, we assumed that undetected and detected groups were similar in size in our estimation of the size of undetected groups.We used a data-augmentation scheme to estimate the size of the undetected groups (Sollman et al. 2016, Roy et al. 2021).

4) Group Sex Composition
To determine the sex ratio in groups of dimorphic species, we assumed that the observed sex of individual o in group k was drawn from a categorical distribution depicting the probability an individual was male or female (π i, 1:2 ): ρ i , j ~Gamma (r i , r i ) log( λ i , j ) ~μ i +log(length j ) We followed procedures developed by Roy et al. (2022), which use the sex ratio in observed groups to predict the sex composition of unobserved groups and individuals of unknown sex.From the predicted sex compositions of all groups, we derived the total number of indicated breeding pairs (IBP) in the survey area using the Black Duck Joint Venture's standardized protocols (Bordage et al. 2017).

Model Estimation
All data processing was conducted in R version 3.6.1 (R Core Team 2019).All analyses were conducted in JAGS version 4.3.0(Plummer 2003), called via the R package jagsUI (Kellner 2019).We used non-informative priors for all parameters and ran three chains with randomized initial start values for 65,000 iterations, using the first 10,000 iterations as burn in and thinning the chains to save every 10 th iteration, with 2000 iterations of adaption.The monitored parameters had an average of 11,717 [90% CI 4066-16500] effective samples.We assessed model convergence visually and by making sure that the Gelman-Rubin statistic was less or equal to 1.1 (Gelman et al. 2004).We also used a posterior predictive check to evaluate the performance of each component of the model.(Appendix 2, for model code and data details.)All results are reported in terms of posterior means with 90% Bayesian Credible intervals (BCI).We considered covariate effects clear if their 90% BCI did not overlap zero.In tables and figures, species are referenced by their standard alpha species codes (Table 2).

Comparison between Fixed-wing and Helicopter Surveys
The fixed-wing transect survey was conducted concurrently with a helicopter plot survey (June 10-19, 2009) that covered the same study area but did not cover the exact same areas within the study area.The helicopter survey employed a double-observer method to quantify detection rates (Roy et al. 2021).Briefly, a helicopter flew plot surveys carrying a primary and secondary pair of observers.The primary pair carried out a standard waterfowl survey, while the secondary pair made note of any groups that the primary pair failed to detect.These undetected groups formed the basis of the detection rate calculations.Roy et al. (2022) provide more details about the survey methodology and doubledetection modeling protocol.
To assess differences between survey platforms, we compared the posterior distributions for group density, individual bird density, and detection for all species.

RESULTS
Observers recorded 774 birds in 340 groups from the fixed-wing along the 12 line transects (1690 km in total).Three species (Common Goldeneye [Bucephala clangula], Ring-necked Duck [Aythya collaris], and Common Merganser [Mergus Merganser]) had less than 10 sightings each and were not included in further analyses, leaving 738 birds from 327 sightings for the analysis.There were either one (N = 107) or two (N = 172) individuals in 85% of the sightings, and the largest sighting was of a group of 35 Canada Geese (Branta canadensis).Observers recorded the most sightings within the first two distance intervals (14-171m), with fewer sightings in farther intervals (Fig. 1).
Only 50% of scoters were identified to species.For the distance detection analysis, we, therefore, pooled all scoter species and both scaup species together into unidentified scoters (combined scoters, regardless of whether species identification was possible) and unidentified scaups, respectively.
The front-left observer recorded more groups in the second distance interval (>98-191m ) than in the first distance interval (14-98m), violating the assumption of decreasing detection with distance.This discrepancy was likely caused by the observer incorrectly assigning distances in these intervals because the configuration of the front window in the Partenavia is such that the reference-distance marks were difficult for the observer to use.To address this issue, we combined the data from the first two intervals for all analyses for both observers (intervals considered were 14m-191m, >191m-259m, and >259-523m).

Group Detection
Detection probability as a function of distance differed between species (Table 1; Fig. 2).Mean detection across all distance categories was highest for unidentified scoters and scaups and lowest for Green-winged Teal (Anas carolinensis, Table 2).Effects of observer identity/position and/or group size or lone birds on detection were not supported (Table 1).Accordingly, for the fully parametrized model, we calculated group density, size, and composition assuming that detection declined with distance and differed only between species.

Group Density
Predicted group density was substantially higher than observed group density and was highest for unidentified scoters, Canada Geese, and Red-breasted Mergansers (Table 2).

Group Size and Composition
Observations ranged from single birds up to 35 birds, with groups of two (likely breeding pairs) being most common for the majority of species.All six groups larger than 10 were Canada Geese.The model predicted that most species were more commonly found in pairs (Fig. 2), with the exception of Canada Geese which were more regularly observed in small groups (3.53 [3.07-4.04]).

Comparisons of fixed-wing and helicopters
Detection rates were lower in the fixed-wing survey than in the helicopter survey.However, the degree to which the survey platform affected detection differed among species (Fig. 3a).For most species, the fixed-wing survey estimated lower densities of both groups of birds and individual birds than the helicopter survey after correcting both surveys for detection probability (Fig. 3b,3c).Corrected IBP densities were more comparable between the survey platforms, but the estimates of breeding pair densities from the fixed-wing component were lower than from the helicopter for Green-Wing Teal, Canada Goose, and unidentified scoters (Fig. 3d).

DISCUSSION
We effectively used distance-sampling methods to control for imperfect detection during a trial breeding sea duck fixed-wing survey.After accounting for detection, the predicted densities were approximately double those initially observed for all species.These results were more comparable to estimates obtained through a helicopter survey using a double-observer protocol to assess detection rates conducted over the same study area at the same time where >95% of groups were detected (Roy et al. 2021).This improvement is notable and did not require any additional flight time or observers to implement the distance-sampling protocol.However, when compared to the helicopter survey, we found that the fixed-wing survey still underestimated waterfowl densities for many species.
Scoters and Red-breasted Mergansers had the highest indicated pair densities of the species surveyed.Scoter density in the study area (0.15 [0.12-0.18])indicated pair per km²) was lower than in the Hudson Bay lowlands, Ontario, where the combined scoter density was 0.33 indicated pairs per km² (Brook et al. 2012) but comparable to densities near Lake Bienville, Quebec (0.06-0.20 pairs per km²; Savard and Lamonthe 1991).Although direct comparisons among studies are difficult to make due to differences in methodology, survey platforms, and unmeasured bias related to detection probabilities in the latter two studies, these results indicate that waterfowl densities were generally low in our study area.
Our models indicated that detection varied by species and by distance from the aircraft.We did not find any differences in detection related to group size, which is consistent with other breeding waterfowl survey analyses (Koneff et al. 2008, Roy et al. 2021, Royle 2008;Wilson, Stehn, and Fischer, unpublished manuscript).The lack of effect of group size on detection may be common in surveys of breeding birds because most observations during the breeding season are composed of either single males or pairs and exhibit little variation in size (Pearse et al. 2008).We also found no differences in detection rate between observers, suggesting that both observers were equally skilled and not hampered by their different positions in the aircraft.However, observer identity was confounded with position in the aircraft.This means we are unable to distinguish between observer and position effects.Previous studies found that the rear observer in fixed-wing had lower detection (Koneff et al. 2008;Stehn, unpublished manuscript;Wilson, Stehn, and Fischer, unpublished manuscript), suggesting that our rear observer may have had high detection regardless of seat position.Nevertheless, our results reinforce the conclusion that patterns of detection vary across crews and may introduce survey bias unless measured and corrected.
We found that detection of sea ducks across the entire 0-523m strip transect was low in the fixed-wing platform (~70% of groups in our survey area were undetected) and addressing imperfect detection in fixed-wing surveys nearly doubled the density estimates for the six species of waterfowl included in this study.To facilitate comparisons with detection estimates calculated by the USFWS for their breeding population surveys (Stehn, unpublished manuscript; Wilson, Stehn, and Fischer, unpublished manuscript), we derived the species-specific detection rates for the inner 200 m strip transect using the speciesspecific detection functions.Across a 200 m strip transect, our derived detection rates for scaups (0.78 [0.72-0.84])were higher than both double-observer based detection estimates for Greater Scaups (Aythya marila) (~0.6;Wilson, Stehn, and Fischer, unpublished manuscript) and unidentified scaup species (0.55 [0.50-0.59];Stehn, unpublished manuscript).Both of these surveys were conducted by the USFWS using 200m fixed-wing line transects in Alaska.Unidentified scoters had higher detection in our survey than in the Alaska survey: (0.79 [0.74-0.83]and 0.59 [0.55-0.63],respectively; Stehn, unpublished manuscript).Canada Goose detection estimates were very similar in this survey and in Alaska (Wilson, Stehn, and Fischer, unpublished manuscript).Though there were effects of observer position, average detection rates of Black Ducks in Eastern Canada, as measured through double-observer methods, were approximately 0.60, also matching our detection estimate of 0.61 [0.49-0.72](Koneff et al. 2008).Ultimately, the few published estimates of detection rates for similar species suggest that estimating detection using distance sampling provides comparable estimates to double-counting procedures, without requiring additional flight time or observers.
We found that, even after correcting for detection, densities measured using fixed-wing line transects were low when compared to densities measured using helicopter plots.Detection consists of two components: perception and availability bias.Perception bias is the proportion of birds that are visible but not seen by observers and availability bias to birds that are not available to observers because they are concealed (e.g., females on nests, diving ducks under water, etc.; Marsh and Sinclair 1989).Distance-sampling and double-observer methods for estimating detection only allow estimation of the perception component of detection (Marsh andSinclair 1989, Winiarski et al. 2014).Our results suggest that availability bias in the fixed-wing is probably substantial for some species and greater from a fixed-wing than from a helicopter.
The difference in availability bias between studies could be explained by both the survey design (transect vs. plot) and survey platform (fixed-wing vs helicopter).Transect surveys such as this fixed-wing survey require the aircraft to follow a line at a fixed altitude and speed, which only allows observers a few seconds to scan any particular area for birds.With their low maneuverability, fixed-wing aircraft are best suited to flying line or strip transects at constant course, speed, and altitude.They are not suitable for surveying mountainous or hilly areas because the height of the ground changes faster than the pilot can compensate.In contrast, plot survey design allows observers to actively search available habitat and investigate any disturbances on the water's surface (Ross 1985).Helicopters' slower speed, maneuverability, and ability to hover allows observers more time to search for birds and detect diving birds on the surface of the water during a plot survey that would be possible during a transect survey.Additionally, because helicopters can fly at lower altitudes than fixed-wing, more birds move in response to the disturbance from the aircraft, which may increase both their availability and probability of detection.Helicopters' maneuverability also allows navigators to choose an approach that minimizes effects of glare and wind on visibility and allows pilots to navigate topography.When fitted with rear bubble windows, helicopters can provide a 270° field of view, much greater than the rear fixed-wing window used in this fixed-wing study, which increases the likelihood that the observers detect available waterfowl.All of these factors likely contribute to lowering the availability of groups for detection in fixed-wing surveys, although some level of availability bias still occurs in helicopter surveys.
Another limitation of the fixed-wing is that observers could not reliably distinguish between scoter and scaup species.Observers were only able to identify 50% of scoters to species level, and scaups are never identified to species level.Additionally, experienced waterfowl observers express concerns that these species are likely to be misidentified from the air as they look very similar.Mis-identification or biases in which species are more easily identified may explain discrepancies in the scoter species ratios seen from the helicopter compared to the fixed-wing (65 [63-66]% Surf Scoters, 28 [27-30]% Black Scoters, and 6.5 [6.1-7.5]%White-winged Scoters compared to 39% Surf Scoters and 61% Black Scoters respectively).This limitation to genera-level identification is severe but shared with all traditional fixed-wing surveys like the WBPHS.As such, there is extremely limited information available for species level management of scoters in the western boreal.In contrast, helicopter surveys, such as the EWS, are able to distinguish between scoter species (Ross 1985, Bateman et al. 2017, Roy et al. 2021) as observers can use binoculars and cameras with image stabilization to improve identification.The helicopter scoter species information is not fully integrated with the fixed-wing data for the EWS and the data are analyzed at the genus level (Zimmerman et al. 2012).Individual species estimates can be derived from the helicopter survey but have limited values due to the limited coverage of the helicopter surveys in the core breeding area of these species.Scaups can still be difficult to distinguish from a helicopter, and abundances and trends are still presented for the two species combined.

Recommendations
Despite some of the shortcomings of the fixed-wing aircraft for waterfowl surveys (e.g., lower detection), they have much greater range and are less expensive to operate than helicopters, making fixed-wing better suited for surveys in remote regions of the Canadian north.Thus, any large-scale survey of breeding sea ducks will probably rely heavily on fixed-wing aircraft.
Implementing distance sampling within the fixed-wing survey protocol can help overcome many of the shortcomings of fixedwing surveys by correcting for imperfect detection without imposing additional costs.For a long-term monitoring program, the methods used to estimate abundance should be resilient to changes in methodology.Because the makes or models of aircraft, and the observers and pilots that crew the planes, change over time, including methods to estimate detection allows us to account for the effects of these biases on estimates of abundance.
The comparisons of the results from the fixed-wing and helicopters suggest that waterfowl are less available to be surveyed during fixed-wing transect surveys than helicopter plot surveys and that this availability bias is substantial.As there are currently no practical means to estimate availability bias, it would be advantageous to combine the strength of both fixed-wing aircraft and helicopters by using helicopter plot surveys to provide the information on species composition and address the high rate of availability bias in the fixed-wing.Such an integrated, detectionbased helicopter and fixed-wing model could be applied more broadly to other waterfowl surveys, including the EWS which has been collecting detection data since 2010.Ongoing efforts to quantify fixed-wing detection and availability biases for waterfowl may also inform VCFs for existing fixed-wing surveys (e.g., WBPHS).Finally, our next step in developing a sea duck monitoring strategy is to develop a methodology to reconcile and combine the helicopter and fixed-wing components by accounting for availability bias.
Responses to this article can be read online at: https://www.ace-eco.org/issues/responses.php/2260

Fig. 1 .
Fig. 1.Map of survey areas on the border of Labrador and Quebec, Canada.Parallel lines represent line transects.

Fig. 2 .
Fig. 2. Flock detection probabilities of six waterfowl species by distance.Lines and dark gray shaded areas represent mean detection probability and 90% BCIs.

Fig. 3 .
Fig. 3. A) Mean detection of flocks (calculated across distance intervals for the fixed-wing aircraft survey), B) flock density, C) density of birds, and D) density of indicated pairs for this fixed-wing aircraft transect survey (black) compared with a concurrent helicopter plot survey (gray; Roy et al. 2021).Points represent means; bars represent 90% BCIs.

Figure A1 :
Figure A1: Observations of flocks of birds in each distance interval, per species.

Figure A2 :
Figure A2: Mean flock size.Points represent means; bars represent 90% BCIs.The dashed horizontal line represents a flock size of two, which would likely be a breeding pair.

Figure A3 :
Figure A3: Mean proportion of male birds for each dimorphic species.Points represent means; bars represent 90% BCIs.

Table 1 .
Mean widely applicable information criterion (WAIC)for each detection model.Delta WAIC is calculated as the difference from the minimum WAIC.

Table 2 .
Detection and observed and predicted density of six species of waterfowl calculated as flocks, individuals, or total indicated pairs (TIP) per km2.Observed density estimates are not corrected for detection; predicted density corrects for the decline in detection rates with increasing distance between fixed-wing and flock.The estimated detection out to 200 m is comparable to USFWS standard fixed-wing transect surveys.Model parameter estimates are presented as mean[90%CI].Scoters (USCO) includes Black Scoters (BLSC), White-winged Scoters (WWSC), Surf Scoters (SUSC), and scoters that were not identifiable to species level.Scaup (USCA) are a combination of Greater (GRSC) and Lesser Scaup (LESC) both of which were unidentifiable.