Long-billed curley (numenius americanus) rangewide survey and monitoring guidelines

              Final: October 2, 2003
1
Long-Billed Curlew (Numenius americanus)
Rangewide Survey and Monitoring Guidelines
Stephanie L. Jones
Division of Nongame Migratory Birds, USFWS, Region 6,
PO Box 25486, Denver, CO 80225
303-236-4409, E-mail: stephanie_jones@fws.gov
Dr. Thomas R. Stanley
Research Wildlife Biologist, USGS, Fort Collins Science Center
2150 Centre Avenue, Bldg. C
Fort Collins, CO 80526
970-226-9360; E-mail: tom_stanley@usgs.gov
Dr. Susan K. Skagen
Research Wildlife Biologist, USGS, Fort Collins Science Center
2150 Centre Avenue, Bldg. C
Fort Collins, CO 80526
970-226-9461, E-mail: susan_skagen@usgs.gov
Roland L. Redmond
Montana Cooperative Wildlife Research Unit, University of Montana
Missoula, MT 59812
Photo by Bob Gress
Final: October 2, 2003
2
Focal Taxon
Long-billed Curlew
(Numenius americanus)
Breeding, migration and wintering
ranges
Breeding: Long-billed Curlews
(LBCUs) are found in low densities
in the Great Basin, Columbia River
Basin, and the shortgrass prairie
region of the Great Plains.
Migration: Pacific coast and central
U.S.
Wintering: California coast south of Humboldt Bay, southeast Atlantic coast from South
Carolina to central Florida, occasionally North Carolina, northern and central Mexico, Baja
California, and the Gulf of Mexico.
Background
LBCUs are a species of special concern throughout much of their breeding range in
North America, with both the U.S. and Canadian Shorebird Plans listing them as “Highly
Imperiled” (Brown et al. 2001). LBCUs are also listed in the U.S. as a Bird of Conservation
Concern, at the National level, within FWS Regions 1, 2, 4 and 6, and for many Bird
Conservation Regions (U.S. Fish and Wildlife Service 2002). This level of concern is due to
apparent population declines, particularly in the shortgrass and mixed-grass prairie of the
western Great Plains (Brown et al. 2001). Threats include habitat loss and fragmentation due to
Final: October 2, 2003
3
agricultural conversion (cropland and tame pasture), encroachment of woody vegetation, and
urban development. For details on LBCU ecology, management, and conservation, refer to
Dugger and Dugger (2002).
Breeding Bird Survey (BBS) data for LBCUs presently includes 227 survey routes
rangewide, with 187 of those routes in the United States. BBS trends (1996-2002) are negative
throughout much of its range, although trends are statistically significant only in FWS Region 6
(-0.2, P = 0.03) and FWS Region 1 (4.4, P = 0.01).
In general, species are considered adequately monitored by the BBS if the standard error
(SE) of the estimated rangewide trend is < 0.90 and if there is no reason to believe that bias (e.g.,
roadside, detectability, & survey timing) is especially large (Bart and Francis In Prep.). Using
BBS data, LBCUs have a SE of 1.10 (Bart and Frances In Prep.), and therefore the BBS does not
adequately monitor LBCU trends. An increase in the number of BBS routes with LBCU could
potentially lower the SE below the 0.90 threshold. However, since BBS routes are surveyed in
June when LBCUs are largely inconspicuous (Gratto-Trevor, pers. comm.), this results in bias in
BBS for LBCUs that an increase in the numbers of routes would not decrease. We believe this
bias is substantial and that the BBS does not adequately reflect LBCUs trends. Our goal here is
to design a monitoring scheme that provides estimates of population size and the data necessary
to design a program to monitor trends in abundance rangewide.
Objectives
Objective 1 - Determine present LBCU distribution and population size rangewide during a two-year
survey.
This information will be used for the development of management and conservation
strategies, which will include the development of a monitoring plan after identifying populations
or ecoregions of concern. At present, population size estimates are available for only a few areas
(e.g. British Columbia, Alberta, and southeast Colorado). These estimates are derived from very
different methods, with large variations in reliability.
Final: October 2, 2003
4
Objective 2 – Develop a plan to monitor long-term trends in population size and distribution.
Regional and rangewide trend estimates are available from the BBS from 1966 - 2002.
However, since the reliability and bias associated with BBS data are unknown, LBCU may
require an additional survey to monitor trends.
Methods
This proposal is for a two-year survey for Objective 1, and development of an alternate
monitoring survey in Objective 2.
Objective 1:
The general approach will be to: 1) specify a spatial sampling frame covering the
geographic range of LBCUs; 2) stratify sample units on the basis of factors thought to influence
LBCU populations; 3) select a random sample of units from each stratum; 4) specify a unit for
subsampling; and 5) collect data for population estimation. Under this nested, stratified random
sampling design, adapted from Saunders (2001), it will be possible to estimate population sizes
and error terms for the strata and geographic area covered by the sampling frame.
Sampling frame:
LBCUs need large, open (> 250 m wide), contiguous patches of suitable grassland and
rangeland habitats, particularly large patches of native grasslands (e.g., Buchloe-Bouteloua
grassland) with a low vegetative profile for nesting. LBCUs rarely nest in seeded pastures,
although they will use this habitat for other activities. Many other grassland types are used for
nesting, and are considered suitable habitat. Close proximity to water is not critical for adults, at
least not early in the breeding season. In southwest Idaho, water was within 5 - 10 km of the
core nesting grounds during March and April (Redmond and Jenni 1986). Some considerations
to reduce bias include that LBCUs have a broad but patchy distribution, may be semi-colonial,
and generally nest in low densities (Dugger and Dugger 2002).
Final: October 2, 2003
5
We will delineate the geographic range of LBCUs using GIS and combining land cover
maps that identify potentially suitable nesting habitat (as described above), BBS data, and other
data on LBCU occurrence (e.g. Breeding Bird Atlases). Land cover maps are available for the
western United States at 30-m resolution from the National Land Cover Data
(http://landcover.usgs.gov/nationallandcover.html), or at 1 km resolution for western North
America from the USGS Land Cover Characteristics Database
(http://edcdaac.usgs.gov/glcc/nadoc2_0.html). BBS data on LBCUs are available from the
Patuxent Wildlife Research Center (http://www.mbr-pwrc.usgs.gov/bbs/). Townships falling
within areas containing potentially suitable nesting habitat or where LBCUs historically have
occurred will be considered the population of sample units that comprise the sampling frame.
Stratification of sample units:
The goal of stratification is to partition sample population, in this case the population of
sample units (i.e., townships) comprising the sampling frame, into strata that are relatively
homogeneous with respect to the variable being measured (i.e., population size). In the present
case, we will use data from land cover maps, and possibly the BBS, to construct strata into which
sample units can be placed a priori. One possible approach would be to mimic the scheme used
by Saunders (2001), in which stratification was based on the percentage of native prairie and/or
other suitable LBCU habitat in a township. An alternative approach would be to stratify on the
basis of the existence of contiguous patches of grassland >250 m wide. We will consider these
approaches and others that might better exploit the land cover and historical bird data available
to us; it may be possible to post-stratify once we get on the ground and can determine what is
really available.
Selection of sample units:
Sample units within each stratum will be selected using a simple random sampling
approach (without replacement), where each unit has an equal probability of becoming part of
the sample population. The number of sample units comprising the sample population will be
Final: October 2, 2003
6
constrained by the number of personnel that can be hired and distances between selected sample
units. For the strata and population estimation methods used by Saunders (2001), it was
estimated that a total of 412 sample units would need to be sampled to detect a 20% change in
the LBCU nesting population with a power of 90% (a desirable target). As Saunders rightfully
points out, however, such a target is unrealistic for LBCUs. We will attempt to maximize the
precision of our LBCU population estimates by sampling as many units as possible within a
season, and by optimally allocating those units among strata using estimated variances and
standard statistical formulae (e.g., Cochran 1977). We will use the strata-specific variance
estimates reported by Saunders (2001), if appropriate for the strata we ultimately define, or will
estimate strata-specific variances using count data from BBS routes in those strata. After the
first year of data are collected, the optimal allocation of units among strata will be recomputed
using actual strata-specific variance estimates. A new set of sample units will be selected for
surveys in year 2 to maximize information on LBCU distribution and to improve occupancy and
sightability models (see below). The second year will also be necessary to quantify year effects,
such as amount of rainfall.
Specification of subsampling units:
Within each sample unit, a single 32-km route will be established using the criteria
specified in Saunders (2001:4) and using suitable habitat in proportion to their availability, and
points will be designated for sampling at 0.8-km intervals along the route (i.e., 40 points/route).
Because any particular route represents only one of the many routes (and hence points) that
could have been established, a route is best considered a subsample of size one from the sample
unit. As such, it will not be possible (due to lack of replication) to use data from routes to
estimate error terms for population size for sample units (i.e. error terms for population size can
not be determined at the route level but data must be combined over many routes to determine
population size and error terms). Moreover, because points comprising a route are not truly a
random sample of the population of points existing in a sample unit, parameter estimates derived
from routes may be biased. Because it is not practical or operationally feasible to randomly
Final: October 2, 2003
7
sample points in a township, we must assume that parameter estimates obtained from a route are
representative of the sample unit itself. We believe this assumption is reasonable because the
area sampled by the route represents 21.8% of the area of the township itself (more if the actual
search radius exceeds the minimum 400-m radius assumed), a fairly large percentage.
Data collection on subsampling units:
The objective of this study is to estimate the population size (N) of LBCUs. To
accomplish this, two pieces of information are required: n, a random variable representing the
number of LBCUs counted in an area of interest, and π , the probability of detecting a LBCU in
an area of interest. Then, given the count data and detection probability, population size can be
estimated as N = n/π . An important point to note here is that N = n only ifπ =1. That is, all
LBCUs must be detected in an area of interest with 100% certainty for the count to equal the true
population size. Anything less then perfect detection will result in a count n that is, to an
unknown degree, a biased estimate of N. Consequently, it is essential in surveys where
population size is the parameter of interest to conduct the survey in a manner that allows
detection probability (π ) to be estimated.
In this study, count data from points along a route will be used to estimate the LBCU
population size for the route, which as described above will be taken as the population estimate
for the sample unit that contains the route. Routes will be surveyed in a manner coinciding with
the relatively narrow time window (2-4 weeks) corresponding to the arrival and pre-incubation
period of LBCUs, when males are most conspicuous in their aerial display flights (Redmond et
al. 1981). Pilot data on Marbled Godwits (Limosa fedoa) and Willets (Catoptrophorus
semipalmatus) indicate extremely low detectability rates during the incubation period, and pre-incubation
surveys appear to be the most representative of how many were actually nesting in
the area (Gratto-Trevor, pers. commun.). We believe that LBCUs behave similarly and LBCU
surveys will be carried out during the pre-incubation period. At lower elevations and latitudes
(e.g., north Texas and southwest Idaho) the survey period will begin in early March, whereas at
higher elevations and latitudes (e.g., Montana's Centennial Valley), surveys will begin in early
Final: October 2, 2003
8
May, and end 30 May in Canada.
Routes will be surveyed by teams of two observers who will stop at points spaced 0.8-km
apart to record all LBCU seen or heard within 5 minutes. Surveys will start ½ hour before
sunrise and will continue for 4 - 5 hours. Routes will be placed on secondary roads and other
rights-of-way, with parallel routes a minimum of 2-km apart. If no roads are located in a
selected area, a 24-km off-road route will be selected. Data on LBCUs seen or heard at a point
will be collected so that detection probabilities can be estimated from the count data.
Specifically, at each stop there will be a primary observer and a secondary observer, and the
individuals in these roles will alternate between stops. The primary observer will be responsible
for detecting LBCUs by sight or sound, determining by laser rangefinder the distance band (0 -
400 m, 400 - 800 m, >800 m) in which the LBCU occurs, and communicating this information to
the secondary observer. The secondary observer will record the information, including the 1-
minute time interval in which each LBCU was detected, and will record all LBCUs and the
distance band that were not detected by the primary observer. The time of detection will be
based on the first observation, whether the first or second observer saw it. Collecting data in this
manner will allow estimation of detection probabilities and population sizes using two distinctly
different methods: the double-observer approach of Nichols et al. (2000) and the removal-model
approach of Farnsworth et al. (2002). Furthermore, because the data will be collected by
distance band, total area sampled along a route can be calculated so the population size estimate
for the route can be converted to a density estimate.
The detection probability (π ) can be decomposed into the product ψ p , whereψ is the
probability a LBCU is present in an area and p is the probability a LBCU is detected in an area
given it is present. In large-scale monitoring studies it is often of interest to construct a model
for ψ using habitat or other covariates, so that model can be used to predict the probability an
animal occupies a particular area based on habitat (or other covariates) alone. This "occupancy
model" can then be used to adjust rangewide population size estimates. The model could also be
used as planning tool for future monitoring activities, and to identify potentially important areas
for management activities. It is common to model ψ using logistic regression; however, this
Final: October 2, 2003
9
approach assumes that sample units are correctly classified as occupied or unoccupied.
Unfortunately, when the conditional detection probability p is less than 1, this assumption is
violated because there will exist occupied sample units that are classified as unoccupied because
the LBCUs present went undetected. Modeling p as a function of weather, habitat, observer, or
other covariates is likewise often of interest in monitoring studies. The resulting "sightability
model" for p can then be used to identify the best times for surveying or to adjust population size
estimates to account for variable sightability conditions.
MacKenzie et al. (2002) offer a solution to the problem of estimating ψ when p < 1.
They constructed a model that allows joint estimation of ψ and p, and allows these parameters to
be modeled as functions of covariates. This allows the possibility of constructing a robust
occupancy model for ψ as well as a sightability model for p, both of which can be exploited for
planning, research, and management. In this study, assuming sufficient personnel resources and
adequate travel funds, a small subset of units from our stratified random sample may be selected
for multiple visits (approximately 5 visits/sample unit) so that ψ and p can be modeled using the
MacKenzie et al. (2002) approach. Covariates for modeling ψ should be time-constant and site-specific,
and might include things such as % native grassland (grazed and ungrazed), % tame
pasture, % cultivated (irrigated or dryland), % other dominant grassland and rangeland species
present (e.g. cheat grass, Bromus secalinus ) and distance to other features (e.g., riparian,
farmsteads, irrigation canals, industrial activity, gravel pits), distance and type of nearest water;
and vegetation structure and composition. Covariates for modeling p will include things such as
time of day, precipitation, wind speed, and observer.
Statistical Analysis:
Population size and density estimates for sample units will be combined to obtain point
and error estimates for strata and the entire geographic range using the standard formulae of
sampling theory (Cochran 1977). If data can be obtained to estimate ψ and p under the
MacKenzie et al. (2002) model, occupancy and sightability models will be constructed and used
Final: October 2, 2003
10
to adjust point estimates.
Objective 2:
The general approach to accomplishing Objective 2 will be to use the information
obtained from Objective 1 to determine the priority and need for continuing monitoring to
promote the conservation of LBCU. The data gathered in the rangewide survey (in Objective 1)
will be used to further evaluate the need for supplemental monitoring and the geographic
priorities for this monitoring. After this evaluation, the data from Objective 1 could be used to
design a long-term monitoring program.
Implementation
Implementation of this survey will use existing resources, volunteers, and agency
biologists, along with the new funding received to hire new personnel. Many state and federal
biologists, in both the U.S. and Canada, have expressed an interest in cooperating with the
project, and would be interested in conducting routes in their states and provinces. A minimum
of four biologists will be needed to complete the surveys, coordinate the volunteers and state
biologists, analyze the data, and complete the reports (Redmond and Jones 2002). This project
was selected for implementation by the FWS Region 6 Science Support Team for funding in
FY03/04 (Redmond and Jones 2002) and the survey will occur in 1 Mar – 30 May, 2004 and
2005.
Many state and provincial natural resources agencies have expressed an interest in
estimating LBCU population numbers for their state. We will work with them to establish the
sample sizes required to achieve this goal, and will identify sample units in their jurisdiction that
were not selected for the rangewide survey so they can survey those routes. Some volunteers,
particularly in the more populous states, could be used for a number of surveys. In the event
cooperators and volunteers are only able to collect count data from routes, estimated detection
probabilities and sightability models from other routes could be used to adjust those data to yield
population size estimates.
Final: October 2, 2003
11
Acknowledgements
We thank the many biologists that assisted us with the ideas and implementation of this study
proposal including Brad Andres, Jon Bart, J. Scott Dieni, John Dinan, Suzanne Fellows, Susan
Haig, Sandra Hagen, Bill Howe, Cheri Gratto-Trevor, Jeff Marks, Lew Oring, Fritz Prellwitz,
Richard Quinlan, Christopher Rustay, and Sue Thomas. For reviews of previous versions, we
thank Brad Andres, Jon Bart, J. Scott Dieni, Suzanne Fellows, David Klute, Don Paul, Rick
Lanctot, Robert Russell, and Jason Woodward. We thank all the biologists and naturalists that
sent us data on Long-billed Curlew locations and arrival dates. Thanks to Doug Backlund and
Bob Gress for the photos and to BNA and Alan Poole for the range map.
References
Bart, J. and C. Francis. In prep. Quantitative Goals for Avian Monitoring Programs. Manuscript.
Brown, S., C. Hickey, B. Harrington, and R. Gill (eds.). 2001. United States Shorebird
Conservation Plan. Manomet Center for Conservation Sciences, Manomet, Massachusetts. 70 pp.
Cochran, W.G. 1977. Sampling theory. John Wiley & Sons, Inc. New York.
Dugger, B.D. and K.M. Dugger. 2002. Long-billed Curlew (Numenius americanus). In Birds
of North America, No. 628 (A. Poole and F.B. Gill, eds.). The Birds of North America, Inc.,
Philadelphia, PA.
Farnsworth, G.L., K.H. Pollock, J.D. Nichols, T.R. Simons, J.E. Hines, and J.R. Sauer. 2002. A
removal model for estimating detection probabilities from point-count surveys. Auk 119:414-
425.
MacKenzie, D.I., J.D. Nichols, G.B. Lachman, S. Droege, J. A. Royle, and C.A. Langtimm.
Final: October 2, 2003
12
2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology
83:2248-2255.
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.
Redmond, R. L. and S. L. Jones. 2002. Develop Long-Billed Curlew (Numenius americanus)
Rangewide Monitoring Guidelines. FY 2003 Fish and Wildlife Science Support Proposal, FWS
Region 6.
Redmond, R. L. and D. A. Jenni. 1986. Population ecology of the Long-billed Curlew
(Numenius americanus) in western Idaho. Auk 103: 755-767.
Redmond, R.L., T.K. Bicak, and D.A. Jenni. 1981. An evaluation of breeding season census
techniques for Long-billed Curlews (Numenius americanus). Stud. Avian Biol. 6: 197-201.
Saunders, E. J. 2001. Population estimate and habitat associations of the Long-billed Curlew
(Numenius americanus) in Alberta. Alberta Species at Risk Report No. 25. Edmonton, AB.
http://www.ab-conservation.
com/your_dollars_at_work/wildlife_fish/birds/project_details.asp?project=151
U. S. Fish and Wildlife Service. 2002. Birds of Conservation Concern-2002. USDOI, Fish and
Wildlife Service, Administrative Report, Arlington, VA.
Photo by Doug Backlund