Using doubleobserver aerial surveys to monitor nesting bald eagles in ...

3 downloads 2322 Views 269KB Size Report
Jul 17, 2014 - Monitor Nesting Bald Eagles in Alaska: Are All Nests Available for Detection? TAMMY L. WILSON,1 Southwest Alaska Network, National Park ...
The Journal of Wildlife Management 78(6):1096–1103; 2014; DOI: 10.1002/jwmg.753

Research Article

Using Double-Observer Aerial Surveys to Monitor Nesting Bald Eagles in Alaska: Are All Nests Available for Detection? TAMMY L. WILSON,1 Southwest Alaska Network, National Park Service, 240W. 5th Avenue, Anchorage, AK 99501, USA JOSHUA H. SCHMIDT, Central Alaska Network, National Park Service, 4175 Geist Road, Fairbanks, AK 99709, USA WILLIAM L. THOMPSON,2 Southwest Alaska Network, National Park Service, 240W. 5th Avenue, Anchorage, AK 99501, USA LAURA M. PHILLIPS, Kenai Fjords National Park, National Park Service, P.O. Box 1727, Seward, AK 99664, USA

ABSTRACT The abundance of nesting eagles is often identified as the parameter of primary interest for

monitoring their populations. We compared the standard dual-frame estimator, which is recommended in the bald eagle post-delisting monitoring plan, with a Bayesian multistate capture-recapture approach to estimate the total number and number of active nests (nests with incubating adults) along the remote Kenai Fjords National Park coastline from 2009 to 2012. Two independent observers conducted aerial surveys of random transects during peak nest initiation in May. Both methods produced similar estimates of nest abundance, but the Bayesian multistate model allowed more flexibility to accommodate shifting management priorities. Estimates of the total number of nests and the number of active nests increased by approximately 49% between 2009 and 2012. This increase was much greater than expected based on feasible rates of nest loss and creation for our study area, indicating apparent estimator bias. Survey-specific conditions (e.g., aircraft height) that made some nests unavailable to both observers were the most likely cause of the bias. We recommend that bald eagle nest monitoring include 2 surveys during the early breeding season to reduce bias of annual capture-recapture estimators. Our results demonstrate that incomplete availability may be an important source of bias for many double-observer aerial wildlife surveys. Published 2014. This article is a U.S. Government work and is in the public domain in the USA. KEY WORDS Alaska, availability bias, bald eagle, Bayesian multistate model, detection probability, double-observer, dual-frame design, Haliaeetus leucocephalus, long-term monitoring, nest activity, perception bias.

Assessment of animal abundance is a fundamental goal of many wildlife-monitoring programs, especially as part of species recovery. Abundance usually must be modeled from observations because of difficulties with obtaining a complete count (Williams et al. 2002, Nichols et al. 2009). Over the years, there has been considerable theoretical development in reducing detection bias related to spatial coverage, cue availability, and perception ability, all of which are addressed differently by each sampling and analytical method (Nichols et al. 2009, Schmidt et al. 2013). Each sampling approach makes implicit assumptions about detection components that must be met for inferences to be valid and unbiased. Logistical difficulties, such as isolation and weather, also can affect the feasibility of many potential monitoring designs. For example, wildlife monitoring in Alaska often relies on observation of unmarked animals from aircraft (e.g., Reynolds et al. 2011, McIntyre and Schmidt 2012, Schmidt Received: 29 November 2013; Accepted: 23 May 2014 Published: 17 July 2014 1

E-mail: [email protected] Present address: U.S. Fish and Wildlife Service, 300 Westgate Center Drive, Hadley, MA 01035, USA 2

1096

et al. 2012), which imposes severe budgetary, logistical, and methodological constraints on sampling designs. As a result, practitioners must consider both budgets and objectives, when choosing among available monitoring approaches to obtain robust estimates of the population of interest (Field et al. 2007, Lindenmayer and Likens 2009, Reynolds et al. 2011). Both public and private organizations have made concerted efforts to develop appropriate protocols for monitoring populations of nesting bald eagles (Haliaeetus leucocephalus) as they recover from past population declines (Grier 1982). At present, the bald eagle post-delisting monitoring plan employs a dual-frame sampling scheme that is combined with double-observer aerial survey methods to account for incomplete detection (U.S. Fish and Wildlife Service 2009). Dual frame refers to a list frame and an area frame; the list frame contains locations and other attributes of nests detected during previous surveys, whereas the area frame contains all nests (detected and undetected) within the entire geographical area of interest (Haines and Pollock 1998). The surveyor divides the area frame into survey units and then randomly selects a subset of these to sample. Estimator precision is expected to improve through subsequent surveys, because newly detected nests are added to the list frame, The Journal of Wildlife Management



78(6)

continually increasing the number of known nest locations. Absent actual changes in the overall population, the magnitude of the estimates should not change much if detection is adequately modeled by the double-observer method. The dual-frame estimator modified by doubleobserver sampling is the standard approach used throughout the contiguous United States, but it has not been previously implemented in remote areas such as coastal bald eagle nesting habitats in Alaska. The breeding population of bald eagles in Kenai Fjords National Park (KEFJ) was identified for long-term monitoring as part of the United States National Park Service’s Inventory and Monitoring Program (Bennett et al. 2006). When designing and implementing the bald eagle monitoring program in this area, we used accepted protocols developed for post-delisting monitoring in the contiguous United States. The challenges presented by surveying the remote coastlines of Alaska are not typically encountered by researchers working in the contiguous 48 states (e.g., extended travel times, lack of roads). For example, in the contiguous United States, sampling the list frame is typically less expensive than sampling the area frame (Alpizar-Jara et al. 2005) because list frame nests are often located near roads and populated areas, making them accessible for regular monitoring. However, the opposite may be true for remote sites in Alaska, where a general lack of accessibility affects list and area frames similarly. With these factors in mind, practitioners need to implement a robust and defensible survey approach that is feasible in remote areas while meeting local management needs. If monitoring efforts are to be useful for change detection and decision-making, annual surveys must be repeatable and the resulting estimates directly comparable. Our main objective was to develop a protocol for long-term monitoring of the breeding bald eagle population using nest structures on the coastline of Kenai Fjords National Park (Thompson et al. 2009) that was straightforward, efficient to implement, unbiased, and relevant for park managers (Field et al. 2007, Fancy et al. 2009, Lindenmayer and Likens 2009). Bald eagle nests persist on the landscape for many years and could be used by eagles for reproduction or not in any given year. We therefore were interested in estimating both the number of nests available for nesting eagles and the number of nests used by eagles for reproduction (active nests) in any given year. We compared 2 analytical approaches, the standard dual-frame estimator (sensu U.S. Fish and Wildlife Service 2009) and a Bayesian multistate approach, for estimating the total number of nests and the number of active nests initiated each year in our area of interest. We also evaluated the robustness of assumptions about the observation process and provide recommendations for changes to field and analytical protocols that would improve monitoring of bald eagle nests and other species commonly monitored by aerial surveys.

STUDY AREA Kenai Fjords National Park was a 2,712-km2 park located on the southeastern coast of the Kenai Peninsula in southcentral Wilson et al.



Monitoring Bald Eagle Nests in Alaska

Figure 1. Locations of survey transects and bald eagle nests in Kenai Fjords National Park, Alaska. The divisions between the management units are denoted by black lines.

Alaska (Fig. 1). The park contained approximately 800 km of coastline characterized by steep mountains reaching over 1,500 m above sea level, deep-water fjords, and tidewater glaciers. The coastline woody vegetation was dominated by temperate coniferous forest, consisting of Sitka spruce (Picea sitchensis) and mountain hemlock (Tsuga mertensiana), which provided nesting habitat for bald eagles. Rocky cliffs, sea stacks, and rocky islands were also used by nesting eagles. At the time of the study, the local bald eagle population was thought to be stable, with all suitable habitat occupied (Hodges 2011). We divided the Kenai Fjords coastline into 5 regions based on the geography of the fjords: Resurrection Bay, Aialik Bay, Northwestern Fjord, Outer Coast, and Nuka Bay (Fig. 1). The northeastern boundary of the park began near the entrance to Resurrection Bay, the largest fjord on the Kenai Fjords coastline, which included the town of Seward, a base for tour and commercial operations in the area (Cook and Norris 1998). Aialik Bay was a deep inlet over 30 km long, which was created by the retreat of Aialik, Pedersen, and Holgate Glaciers. Aialik Bay was second only to Resurrection Bay in the numbers of visitors, with dozens of tour boats, kayakers and recreational boaters visiting each day. Eight retreating glaciers dominated the geography of Northwestern Fjord. The Outer Coast between Harris Point and McArthur Pass consisted of a series of exposed rocky headlands and cliffs divided by a series of relatively short bays. Nuka Bay was composed of 2 arms with woody, mountainous shorelines and a number of productive salmon streams. Both the Outer Coast and Nuka Bay received little visitation, most of which was related to sport or commercial fishing. 1097

METHODS Study Design and Field Methods Eagle nests included all nest structures large enough to support eagle reproduction, regardless of eagle activity observed at the time of survey. We further classified nests into 2 activity states (active or empty) based on observed eagle activity during the field survey. We classified a nest as active if we observed an eagle sitting on the nest in a manner consistent with incubation or if we observed eggs or chicks; we classified nests as empty otherwise. Our definition is consistent with the bald eagle monitoring literature (U.S. Fish and Wildlife Service 2009, Watts and Duerr 2010, Sauer et al. 2011). We did not identify territories or attempt to reduce the list of known nests by identifying satellite or alternative nests, because of the difficulty in doing so in high-density coastal populations (Hodges 1982), particularly when nest state cannot be determined with certainty in a single visit. Although Kenai Fjords park staff had conducted boatbased eagle monitoring surveys from 1990 to 2002 (Thompson et al. 2009), the data were inadequate to compose a list frame for our aerial survey. Therefore, to generate a list frame, we conducted an initial bald eagle nest survey throughout all suitable habitat (approx. 650 km) along the 800-km Kenai Fjords coastline from Nuka Bay in the west, to the east end of Resurrection Bay between 14 and 19 May 2009 (Fig. 1). We chose a survey date during nest initiation based on data from Prince William Sound, Alaska (U.S. Fish and Wildlife Service, unpublished data), which suggested most breeding eagles would be incubating nests at that time. We used flight lines generated by a GPSMap1 76CSx (Garmin International, Olathe, KS) during the 2009 survey and the adjacent coastal habitat to define our area frame. Using the 2009 flight line, we generated 51, 12.5-km possible sample units. We then used a generalized randomtessellation stratification (GRTS; Stevens and Olsen 2004) design to select a spatially balanced sample of 26 permanent survey transects to be used for all future surveys. We used simulations to obtain both the lengths and numbers of transects based on bald eagle nest locations from the 2009 survey and historical surveys conducted in Kenai Fjords National Park (Thompson and Phillips 2011). We conducted eagle nest surveys on these transects from 9 to 12 May 2010, 9 to 12 May 2011, and 7 to 9 May 2012, covering each transect once per year (Fig. 1). We conducted bald eagle nest surveys using a Robinson R44 Clipper II helicopter (Robinson Helicopter Co., Torrance, CA) equipped with fixed floats flown 90–250 m above mean sea level at 40–75 km/hr just offshore. Because of the steepness of the terrain, we were frequently able to visually survey all suitable nesting habitats along the coastline. Two experienced observers sat in the front and rear seats on the left side of the helicopter, always facing the coastline. All observers had extensive experience conducting wildlife surveys from small aircraft in coastal Alaska, and most had previous experience conducting bald eagle surveys 1098

in Kenai Fjords National Park. Each year, observers calibrated their search image by searching for nests during the 2-hour flight from Seward to transect locations in Nuka Bay. We carried printed photos of all known nests to aid relocations of previously detected nests. We used an independent, double-observer field protocol (Nichols et al. 2000, Koneff et al. 2008) to record eagle nests found during transect surveys. Front and rear seat observers recorded observations independently of each other and observers were not rotated during sampling. The front left window of the helicopter was larger than the back left window, otherwise observers were seated very close to one another and had similar viewing opportunities; most of the nests visible to one should have been visible to the other. When either observer spotted a nest, they waited until the nest was out of view before announcing it had been detected. The pilot then returned to the nest and we recorded, which observer(s) saw the nest during the initial pass, the global positioning system (GPS) location of the nest, nest activity state (i.e., active or empty), and nest attributes (e.g., tree species). On rare occasions, the pilot detected nests that both observers missed; we recorded these nests so they could be added to the list frame. We recorded detection data for each nest only upon first observation, because locations of known nests were stored in a GPS, allowing the pilot to navigate directly to them. We recorded the activity state of each nest known from previous surveys (i.e., list frame nests) that were also located on the area frame transect sampled, while we also searched for previously undetected nests. We did not revisit list frame nests located outside of the area frame sample. We added newly discovered nests to the list of known nests so that they could be revisited the following year. On rare occasions, we failed to revisit a known nest for some reason (e.g., because of a poor GPS location); we marked the activity state of these nests as unknown for the survey year in which we did not visit the nest although we still assumed the nest to be present in the population. For observed nests, we assumed that observed eagle activity was unambiguous (i.e., observed empty nests were really empty). Standard Dual-Frame Analysis We first used the dual-frame estimator described in the bald eagle post-delisting plan (U.S. Fish and Wildlife Service 2009) to estimate the total number of eagle nests and the number of active eagle nests along the coast of Kenai Fjords National Park each year from 2009 to 2012. Dualframe estimation combines nests from a list of known nests with additional random sampling efforts to estimate the total number of nests in an area of interest. The list frame contains the location information for all known nests. The area frame sampling is used to estimate the number of nests that do not occur on the list frame through a process called unduplication. This estimate can be improved by accounting for imperfect detection using any one of the available detection estimation methods (e.g., double observer). The estimate from the area frame sampling is then added to the list frame to obtain the total number of nests. The Journal of Wildlife Management



78(6)

The list frame for surveys after 2009 consisted of all nests observed in 2009, plus any other new nests found during any previous year’s survey. The area frame included all nests located along the Kenai Fjords coastline, and was sampled using the area frame transects described above. We detected 80 nests during list frame sampling in 2009, 31 of which were located on the subsequent years’ 26 area frame transects. Using the algorithm described in Appendix 1 of the postdelisting monitoring plan (U.S. Fish and Wildlife Service 2009), we used the new nests observed during annual area frame surveys to augment the list frame, thereby obtaining an estimate of the total number of nests present (Haines and Pollock 1998, U.S. Fish and Wildlife Service 2009). We corrected the annual counts for imperfect detection using the double-observer method (Nichols et al. 2000, U.S. Fish and Wildlife Service 2009) described below. We produced estimates for 2009 using only the area frame estimation method. Although the method described above estimated the total number of nests in the population, we were also interested in estimating the number of those nests used for eagle reproduction in any given year (i.e., were active). To estimate the number of active nests in the study area, we classified each nest into 1 of 2 activity states (active or empty). Interannual changes in observed nest activity made it difficult to determine which known nests should appear on the list frame for estimating the number of active nests. The postdelisting monitoring plan used list nests submitted by partner agencies and was not designed to address annual turnover in activity status (Watts and Duerr 2010, Sauer et al. 2011). To use the delisting plan’s dual-frame estimator, we found we had to assume that eagle activity at nests was constant through time. We included nests on the list of active nests if we observed they were active at the time of discovery and we kept them on the active list for the duration of the study. Although this is unsatisfactory for many reasons, the estimator will not be affected if the proportion of active nests does not change much through time. We estimated detection (p) using the independent doubleobserver method-using program DOBSERV (Nichols et al. 2000). We used the initial front- and rear-observer detection data for each nest to fit 4 detection models: 1) detection was the same for all years and both seating position; 2) detection was different for each year but was equal for both seating position; 3) detection was equal for all years but different for each seating position; and 4) detection was different for all years and each seating position. We chose the model with the lowest Akaike’s Information Criterion (AICc; Burnham and Anderson 2002) to correct area frame sample estimates for imperfect detection. We did not use nests detected only by the pilot to estimate detectability. Bayesian Multistate Analysis Based on our preliminary use of the dual-frame estimator, we suspected that the application of multistate capturerecapture methods implemented in a Bayesian framework would provide improved inference and interpretability of estimates. Because we detected both empty and active nests, Wilson et al.



Monitoring Bald Eagle Nests in Alaska

we wished to build a model where we could model the activity state at unobserved nests, rather than assume it to be constant as required by the model in the post-delisting monitoring plan. We used a modified form of multistate capture-recapture models (Royle and Link 2005, Nichols et al. 2007), combined with data augmentation (Royle et al. 2007, Royle and Dorazio 2009), to estimate the number of nests in each state in each year. We used the basic multistate model formulation described by Ke´ry and Schaub (2012) but included an arbitrarily large number of potential undetected nests in the first state, representing a population of potential nest structures that were missed by observers. We then estimated the true state (i.e., not a real nest, a real nest), and the conditional activity state (active or empty) of each real nest during Markov chain Monte Carlo simulations (MCMC). An active nest is conditional on the nest existing, and the multistate framework incorporates this relationship in a much more straightforward fashion. The Bayesian multistate model was similar to the standard approach with 2 important modifications. First, we included a transect-level random effect to account for heterogeneity in nest density among transects. Second, we included nests that were known to exist but were detected outside of the doubleobserver survey protocol (e.g., by the pilot) by assigning the known state value to the otherwise undetected nest (J. A. Royle, Patuxent Wildlife Research Center, personal communication). This provided a formal mechanism for including these auxiliary data, eliminating the need to discard useful information that violated the survey protocol. In addition to estimating the numbers of nests, we derived estimates of the difference in the number of nests and proportional increase in the number of nests between 2009 and 2012 during the updating process to help quantify bias due to incomplete availability. For the years 2010–2012, when we surveyed only a sample of transects, we assumed each known nest occurring on an unsurveyed transect remained in the population, although the activity state (empty vs. active) was unknown and was estimated during the updating process. For the multistate analyses, we used R Version 2.14.2 (R Development Core Team, www.r-project. org) for all data manipulation and WinBUGS Version 1.4.1 (MRC Biostatistics Unit, www.mrc-bsu.cam.ac.uk/software/ bugs/) for model fitting. Estimates are presented as means and 95% credible intervals (CrI). Key Model Assumptions When using either estimation approach, several important assumptions must be met or addressed to prevent bias. First, the population must be closed to additions of new nests and losses of existing nests between 2009 and 2012. Therefore, if nests were found for the first time after 2009, the nest was assumed to have been present but undetected in prior years’ surveys. Although this was unlikely to be strictly true, we expected minimal loss of nesting structures over the short duration of our study (Stalmaster 1987). Violations of the closure assumption would lead to positive bias in the estimated number of nests. The second assumption was that all nests were equally available for sampling during each 1099

survey and the sole source of incomplete detection of nests was caused by observer ability (Koneff et al. 2008, Nichols et al. 2009). Because observers were surveying from the same platform, if any of the nests were unavailable (p ¼ 0) to an observer, then it would likely also be unavailable to the other, thereby causing substantial negative bias in the estimated number of nests. The last assumption was that the activity state of each observed nest was determined with certainty. Because observations were simultaneous and we did not conduct revisits, we were precluded from assessing activity state uncertainty. Violations of this assumption would cause negative bias in the number of active nests.

within management subunits of the Kenai Fjords coastline (Fig. 2). The large number of nests that we added to the list frame during area frame sampling (e.g., 27 in 2010) resulted in estimated increases in the numbers of bald eagle nests (Table 1) and the conditionally estimated number of active nests (Table 2) throughout the sampling period. The difference between the total number of nests from 2009 to 2012 for the multistate model was 47 nests (95% CrI ¼ 30–63), which was a proportional increase of 0.49 (95% CrI ¼ 0.28–0.72). This result was unexpected given our relatively high combined estimated probability of detection and suggested that the assumption of equal availability of nests for sampling was violated.

RESULTS We detected 80, 27, 7, and 12 previously undetected bald eagle nests in 2009–2012, respectively. The overall estimated detection probability for nests was 0.87 (95% CI: 0.80–0.94, n ¼ 118) using DOBSERV. The pilot alone detected 3 nests in 2009 and 3 in 2010. The pilot-detected nests were located on the left side of the aircraft and therefore available to the observers but not detected. We found evidence that detection probability differed between observer seating positions. The estimated probability of detecting nests was 0.72 (95% CI: 0.64–0.84. n ¼ 118) for the front observer and 0.53 (95% CI: 0.44–0.62, n ¼ 118) for the rear observer. The model allowing detection probability to vary by year was not strongly supported by the data (DAICc > 5). Although other survey factors may have affected nest detection (e.g., observer identity and survey conditions), we found too few newly detected nests during later surveys (2011 and 2012) to fit more complicated detection models. The detection probabilities based on the Bayesian multistate model were similar to those generated by the double-observer method. The estimates of detection probabilities were 0.69 (95% CrI: 0.57–0.79, n ¼ 118) for the front observer and 0.42 (95% CrI: 0.32–0.51, n ¼ 118) for the rear observer using the Bayesian multistate model fit with the pilot-detected nests. The estimated numbers of both nests (Table 1) and active nests (Table 2) were similar between analytical methods within years. Relaxing our assumption that nest state remained constant through time by allowing the status of unobserved nests to be estimated during Bayesian updating did not substantially change estimates. Through inclusion of the transect-level random effect, the Bayesian multistate model also allowed us to estimate the number of active nests

DISCUSSION We found that the Bayesian multistate approach offered several advantages over the standard dual-frame method, but we also discovered that further survey design refinements are required if nests are to be used for bald eagle monitoring. Although estimates were comparable between the 2 analytical approaches, the Bayesian method treated data more intuitively and allowed greater flexibility to respond to shifting management priorities. The ability to produce estimates for individual bays was a particularly appealing improvement over the standard dual-frame approach because fine-grained estimates could be used to assess the effects of a disturbance (e.g., localized oil spill) on a specific subset of the population. Although the precision of local estimators is lower for smaller areas, local estimators will become more precise over time as additional data are collected. Despite these improvements, we found that not all nests were available for detection, resulting in estimator bias. These results provided a rare example from a field study where unexpected assumption violations and their effects are clearly demonstrated. We do not think the estimated 49% increase in the total number of bald eagle nests from 2009 to 2012 resulted from an increase in the eagle population, because bald eagle populations in Alaska have been stable since 1980 (Hodges 2011). The modeled increase was too large to have been caused by nest construction or loss, given that nest structures last for many years and eagles tend to reuse nests (Stalmaster 1987). Rather, the observed increase was likely due to a failure of the double-observer method to fully address the issue of incomplete detection, resulting in

Table 1. Estimated numbers of bald eagle nests in Kenai Fjords National Park, Alaska, USA from 2009 to 2012 using 2 different estimation methods. The Chao 95% confidence intervals (standard) or 95% credible intervals (Bayesian multistate) are presented in parentheses. Year

Method

2009

Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate

2010 2011 2012

1100

Estimate 92 97 134 139 123 124 141 143

(85–108) (86–112) (119–153) (123–159) (111–138) (117–137) (128–158) (132–157)

SE

CV

5.31 6.93 8.56 8.96 6.93 5.18 7.81 6.32

0.06 0.08 0.06 0.06 0.06 0.04 0.06 0.04

The Journal of Wildlife Management



78(6)

Table 2. Estimated numbers of active bald eagle nests in Kenai Fjords National Park, Alaska, USA from 2009 to 2012 using 2 different estimation methods. The Chao 95% confidence intervals (standard) or 95% credible intervals (Bayesian multistate) are presented in parentheses. Year

Method

2009

Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate Standard dual frame Bayesian multistate

2010 2011 2012

Estimate 51 54 78 74 68 69 88 88

negative bias. This bias occurred despite survey standardization, adherence to strict protocols, and use of experienced observers. That the estimated numbers of nests increased annually at a rate much higher than was biologically reasonable indicated unmodeled heterogeneity in detection probability and/or incomplete availability of nests during annual surveys (i.e., p ¼ 0). Simply applying a sampling approach that addresses incomplete detection may be insufficient and misleading unless the ability to meet the implicit underlying assumptions is carefully evaluated. Unmodeled heterogeneity in detection functions is known to cause bias in resulting population estimators (Link 2003). Heterogeneity can come from sources that are difficult to model, such as those unique to individuals or to variable survey conditions. Availability bias is a special form of heterogeneity where the detection of some individuals (i.e., nests) in the population of interest during a sampling event is 0 (Nichols et al. 2009). In the context of general capturerecapture terminology, the inferential problems we discovered in our study were analogous to temporary emigration (Kendall et al. 1997) and the concept of a superpopulation (Schwarz and Aranason 1996). However, unlike marked animals, nests do not move. In our case, variation in our sampling process (e.g., survey elevation or view angle)

40 35

Number of occupied nests

30 25 20 15 10 5 0 2009

2010

Ailik Bay

Resurrecon Bay

2011 Year Nuka Bay

2012

Northwestern Fjord

Outer Coast

Figure 2. Estimated number of active bald eagle nests for each year from 2009 to 2012 along the coastline of Kenai Fjords National Park, Alaska. We derived estimates from the Bayesian multistate method. Error bars represent 95% credible intervals for each estimate. Wilson et al.



Monitoring Bald Eagle Nests in Alaska

(47–61) (47–64) (67–92) (58–93) (60–79) (58–82) (77–103) (75–104)

SE

CV

3.46 4.37 6.30 9.26 4.95 6.17 6.43 7.33

0.07 0.08 0.08 0.13 0.07 0.09 0.07 0.08

exposed additional nests to sampling each year. For example, a small elevation change on a steep coastline could change the angle of view of a nest from available for detection to unavailable for detection (e.g., blocked by part of the aircraft) simultaneously for both observers (Marsh and Sinclair 1989). Each year a portion of the nests were temporarily unavailable, but adding these to the list frame through time resulted in estimates of the superpopulation of nests exposed to sampling over multiple years, and reducing bias in later years. Unfortunately, we had expected to sample the entire population annually and our inability to do so caused bias in our estimators relative to our stated objectives. We thought the sampled population for each survey was the entire coastal population of eagle nests and the subset of those nests that were active, but we were likely estimating the abundance of an unknown proportion of nests in any given year. The apparent 49% increase in abundance through time (Tables 1 and 2) caused by incomplete availability would mask declines through time or incorrectly imply population growth, with obvious consequences for a monitoring program aimed at the conservation and management of bald eagles. Simple variation in availability between years would decrease our ability to detect actual changes in the population over time because interpretation of the proportion of active nests is not possible in the absence of a good estimate of the total population. Although logistical factors are different in our study area than in typical survey areas in the contiguous 48 states, similar variation in availability of nests for sampling could occur in other areas with obvious implications for accurate population assessment. In addition to addressing availability bias, 2 important avenues of future research would extend the utility of these models as a monitoring tool: relaxing the closure assumption and improving observation of the nest activity state. We assumed known nests did not leave the population and, although nest structures may be serviceable for many years, they can be abandoned or lost for any number of reasons (Stalmaster 1987). A long-term monitoring program should allow the nest lists to be dynamic through time; future model developments should relax the closure assumption by estimating survival of nest structures through time (e.g., Sauer et al. 2011). Second, the current analysis assumes that eagle activity status at nests was observed without error during our single visit in May. We timed nest surveys such that visits occurred during a period when most nests were likely to be attended by an incubating adult (Thompson 1101

et al. 2009). However, nest activity is unlikely to be observed without error during a single visit, because the timing of peak activity is difficult to know in advance of the survey (Fraser et al. 1983). Yearly variation in eagle nesting phenology could negatively bias estimators of the proportion of active nests by an unknown amount depending on the timing of the survey relative to peak nest attendance. This unmodeled heterogeneity would mask population trends over time. Survey modifications might help to mitigate problems we encountered with violating analytical assumptions. Conducting a repeated survey over at least a portion of the survey area would allow us to address the availability bias caused by aircraft position and relax the assumption that the nest activity state was observed perfectly. This would make trend estimation more robust and less reliant on assumptions about the population of nests over multiple years. With this slight modification of the bald eagle sampling protocol, the multistate approach shows promise for addressing the biases found with the present analysis. Accounting for incomplete detection has received much attention in recent years (Nichols et al. 2009, Schmidt et al. 2013) and we demonstrated that a careful assessment of the assumptions of the survey methodology is critical. Although double-observer methods are the recommended approach for addressing incomplete detection of eagle nests, our results suggest that detection heterogeneity resulting in incomplete availability of nests for sampling is also a concern. This finding has important implications for any single-pass, aircraft-based surveys that use double-observer sampling to correct for observation bias. Therefore, survey protocols for species such as brown bears (Ursus arctos; Quang and Becker 1997), dugongs (Dugong dugon; Marsh and Sinclair 1989) and waterfowl (Conroy et al. 2008), may be similarly affected if a proportion of the objects or individuals of interest are unavailable for detection because of aircraft position, or other factors (Marsh and Sinclair 1989). If availability cannot be addressed directly, inference must be restricted to the area actually sampled (Nichols et al. 2009, Schmidt et al. 2013). This will be true for any survey type where incomplete detection is an issue and all components of the detection process must be addressed through design (Schmidt et al. 2013).

MANAGEMENT IMPLICATIONS Well-designed long-term monitoring programs approach objectives and learning as an iterative process requiring periodic reviews of study design, data collection, and analysis protocols (Lindenmayer and Likens 2009). Programs that fail to correct deficiencies or are unresponsive to shifting management priorities risk becoming irrelevant through time (Field et al. 2007, Lindenmayer and Likens 2009). Although the Bayesian estimator was more useful and intuitive than the standard approach for monitoring bald eagle activity at nests, we uncovered an important deficiency in the way we estimated incomplete detection. We recommend that monitoring programs using aerial bald eagle surveys adopt a robust-design sampling protocol (Pollock 1982) for area frame sampling. This design change 1102

requires that future bald eagle monitoring incorporate a second survey flight during the incubation period to reduce availability bias in detection estimates and improve estimation of nest activity. Future data analysis should use an open multistate capture-recapture model allowing new nests to enter and fallen nests to leave the population. Although it is difficult to increase field-sampling efforts during times of budget reduction, failing to produce rigorous estimates reduces the relevancy of monitoring programs (Field et al. 2007). The expected increase in estimator precision may allow for a small reduction in the number of transects required to be sampled, thereby mitigating the cost of additional sampling. Based on our results with the doubleobserver method, other researchers using aerial surveys to conduct animal counts should assess the extent to which the assumption of equal availability is violated.

ACKNOWLEDGMENTS We thank S. Hall, F. Klasner, D. MacLean, T. B. Murphy, and M. Shephard for their logistical support during surveys. We also appreciate our observers L. Adams, S. Hall, C Lindsay, and L. Whitter. Thanks to our pilots R. Hodges, C. Jordon, T. Levanger, C. Redd, M. Roulet, and T. Townsend for allowing us to complete this work safely. C. Lindsay served as a helicopter manager and assisted with GIS needs. We thank M. C. Otto and M. N. O’Ferror for assistance with the dual-frame estimator. Reviews and editorial comments by J. Nichols, W. Thogmartin, E. Merrill, and an anonymous reviewer improved the quality of this manuscript. Funding was provided by the Southwest Alaska Inventory and Monitoring network, Kenai Fjords National Park, and Oceans Alaska Science and Learning Center. Mention of trade or firm names is for reader information, and does not imply endorsement by the National Park Service for any product or service.

LITERATURE CITED Alpizar-Jara, R., K. H. Pollock, and D. E. Haines. 2005. Mark-recapture estimators for dual frame population size of prominent nesting structures: the effect of uncertain detection probability. Environmental and Ecological Statistics 12:155–168. Bennett, A. J., W. L. Thompson, and D. C. Mortenson. 2006. Vital signs monitoring plan Southwest Alaska Network inventory and monitoring program. Southwest Alaska Network, National Park Service, Anchorage, Alaska, USA: http://science.nature.nps.gov/im/units/swan/publications. cfm?tab¼2. Accessed 29 May 2014. Burnham, K. P., and D. R. Anderson 2002. Model selection and multimodel inference: a practical information-theoretic approach. Second edition. Springer, New York, New York, USA. Conroy, M. J., J. T. Peterson, O. L. Bass, C. J. Fonnesbeck, J. E. Howell, C. T. Moore, and J. P. Runge. 2008. Sources of variation in detection of wading birds from aerial surveys in the Florida Everglades. Auk 125:731– 743. Cook, L. A., and F. B. Norris 1998. A stern and rock-bound coast: Kenai Fjords National Park historic resource study. National Park Service, Alaska Support Office, Anchorage, Alaska, USA: http://www.nps.gov/ history/history/online_books/kefj/hrs/hrs.htm. Accessed 29 May 2014. Fancy, S. G., J. E. Gross, and S. L. Carter. 2009. Monitoring the condition of natural resources in US national parks. Environmental Monitoring and Assessment 151:161–174. Field, S. A., P. J. O’Connor, A. J. Tyre, and H. P. Possingham. 2007. Making monitoring meaningful. Austral Ecology 32:485–491. The Journal of Wildlife Management



78(6)

Fraser, J. D., L. D. Frenzel, J. E. Mathisen, F. Martin, and M. E. Shough. 1983. Scheduling bald eagle reproduction surveys. Wildlife Society Bulletin 11:13–16. Grier, J. W. 1982. Ban of DDT and subsequent recovery of reproduction in bald eagles. Science 218:1232–1235. Haines, D. E., and K. H. Pollock 1998. Estimating the number of active and successful bald eagle nests: an application of the dual frame method. Environmental and Ecological Statistics 5:245–256. Hodges, J. I. 1982. Bald eagle nesting studies in Seymour Canal, southeast Alaska. Condor 84:125–127. Hodges, J. I. 2011. Bald eagle population surveys of the North Pacific Ocean 1967–2010. Northwestern Naturalist 92:7–12. Kendall, W. L., J. D. Nichols, and J. E. Hines. 1997. Estimating temporary emigration using capture–recapture data with Pollock’s robust design. Ecology 78:563–578. Ke´ry, M., and M. Schaub 2012. Bayesian population analysis using WinBUGS: a hierarchical perspective. Academic Press, Waltham, Massachusetts, USA. Koneff, M. D., J. A. Royle, M. C. Otto, J. S. Wortham, and J. K. Bidwell. 2008. A double-observer method to estimate detection rate during aerial waterfowl surveys. Journal of Wildlife Management 72:1641–1649. Lindenmayer, D. B., and G. E. Likens 2009. Adaptive monitoring: a new paradigm for long-term research and monitoring. Trends In Ecology & Evolution 24:482–486. Link, W. A. 2003. Nonidentifiability of population size from capturerecapture data with heterogeneous detection probabilities. Biometrics 59:1123–1130. Marsh, H., and D. F. Sinclair 1989. Correcting for visibility bias in strip transect aerial surveys of aquatic fauna. Journal of Wildlife Management 53:1017–1024. McIntyre, C. L., and J. H. Schmidt 2012. Ecological and environmental correlates of territory occupancy and breeding performance of migratory golden eagles Aquila chrysaetos in interior Alaska. Ibis 154:124–135. Nichols, J. D., J. E. Hines, D. I. MacKenzie, M. E. Seamans, and R. J. Gutierrez. 2007. Occupancy estimation and modeling with multiple states and state uncertainty. Ecology 88:1395–1400. Nichols, J. D., J. E. Hines, J. R. Sauer, F. W. Fallon, J. E. Fallon, and P. J. Heglund. 2000. A double-observer approach for estimating detection probability and abundance from point counts. Auk 117:393–408. Nichols, J. D., L. Thomas, and P. B. Conn. 2009. Inferences about landbird abundance from count data: recent advances and future directions. Pages 201–236. in D. L., Thomson, E. G., Cooch, and M. J. Conroy, editors. Modeling demographic processes in marked populations. Springer, Boston, Massachusetts, USA. Pollock, K. H. 1982. A capture-recapture design robust to unequal probability of capture. Journal of Wildlife Management 46:757–760. Quang, P. X., and E. F. Becker 1997. Combining line transect and double count sampling techniques for aerial surveys. Journal of Agricultural, Biological, and Environmental Statistics 2:230–242. Reynolds, J. H., W. L. Thompson, and B. Russell. 2011. Planning for success: identifying effective and efficient survey designs for monitoring. Biological Conservation 144:1278–1284.

Wilson et al.



Monitoring Bald Eagle Nests in Alaska

Royle, J. A., and R. M. Dorazio 2009. Hierarchical modeling and inference in ecology: the analysis of data from populations, metapopulations and communities. Academic Press, New York, New York, USA. Royle, J. A., R. M. Dorazio, and W. A. Link. 2007. Analysis of multinomial models with unknown index using data augmentation. Journal of Computational and Graphical Statistics 16:67–85. Royle, J. A., and W. A. Link 2005. A general class of multinomial mixture models for anuran calling survey data. Ecology 86:2505–2512. Sauer, J. R., M. C. Otto, W. L. Kendall, and G. S. Zimmerman. 2011. Monitoring bald eagles using lists of nests: response to Watts and Duerr. Journal of Wildlife Management 75:509–512. Schmidt, J. H., C. L. McIntyre, and M. C. MacCluskie. 2013. Accounting for incomplete detection: What are we estimating and how might it affect long-term passerine monitoring programs? Biological Conservation 160:130–139. Schmidt, J. H., K. L. Rattenbury, J. P. Lawler, and M. C. MacCluskie. 2012. Using distance sampling and hierarchical models to improve estimates of Dall’s sheep abundance. Journal of Wildlife Management 76:317–327. Schwarz, C. J., and A. N. Aranason 1996. A general methodology for the analysis of capture-recapture experiments in open populations. Biometrics 52:860–873. Stalmaster, M. V. 1987. The bald eagle. Universe Books, New York, New York, USA. Stevens, D. L., and A. R. Olsen 2004. Spatially balanced sampling of natural resources. Journal of the American Statistical Association 99:262–278. Thompson, W. L., S. Hall, and C. R. Lindsay. 2009. Evaluation of a survey method for estimating number and monitoring occupancy of bald eagle nests in Kenai Fjords National Park. Natural Resource Technical Report NPS/SWAN/NRTR–2009/271. National Park Service, Natural Resource Program Center, Fort Collins, Colorado, USA: http://www.nature.nps. gov/publications/nrpm/nrtr.cfm. Accessed 29 May 2014. Thompson, W. L., and L. M. Phillips 2011. Evaluation of a dual-frame design to estimate occupancy and productivity of bald eagle nests in Kenai Fjords National Park. Natural Resource Technical report NPS/SWAN/ NRTR–2011/413. National Park Service, Natural Resource Program Center, Fort Collins, Colorado, USA: http://www.nature.nps.gov/ publications/nrpm/nrr.cfm. Accessed 29 May 2014. U.S. Fish and Wildlife Service. 2009. Post-delisting monitoring plan for the bald eagle (Haliaeetus leucocephalus) in the contiguous 48 states. U.S. Fish and Wildlife Service, Divisions of Endangered Species and Migratory Birds and State Programs, Midwest Regional Office, Twin Cities, Minnesota, USA. http://www.fws.gov/midwest/eagle/protect/FinalBAEA_ PDMPlan.html. Accessed 29 May 2014. Watts, B. D., and A. E. Duerr 2010. Nest turnover rates and list-frame decay in bald eagles: implications for the national monitoring plan. Journal of Wildlife Management 74:944–970. Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of animal populations. Academic Press, San Diego, California, USA.

Associate Editor: Wayne Thogmartin.

1103