Integrating occupancy modeling and interview data for corridor identification: A case study for jaguars in Nicaragua

Biological Conservation 144 (2011) 892–901

Contents lists available at ScienceDirect

Biological Conservation journal homepage: www.elsevier.com/locate/biocon

Integrating occupancy modeling and interview data for corridor identification: A case study for jaguars in Nicaragua Katherine A. Zeller a,⇑, Sahil Nijhawan a, Roberto Salom-Pérez a, Sandra H. Potosme a, James E. Hines b a b

Panthera, 8 West 40th Street, New York, NY 10018, USA USGS Patuxent Wildlife Research Center, 11510 American Holly Drive, Laurel, MD 20708-4017, USA

a r t i c l e

i n f o

Article history: Received 26 August 2010 Received in revised form 1 December 2010 Accepted 6 December 2010 Available online 5 January 2011 Keywords: Panthera onca Connectivity Conservation planning Detection/non-detection Detection probability Interviews

a b s t r a c t Corridors are critical elements in the long-term conservation of wide-ranging species like the jaguar (Panthera onca). Jaguar corridors across the range of the species were initially identified using a GIS-based least-cost corridor model. However, due to inherent errors in remotely sensed data and model uncertainties, these corridors warrant field verification before conservation efforts can begin. We developed a novel corridor assessment protocol based on interview data and site occupancy modeling. We divided our pilot study area, in southeastern Nicaragua, into 71, 6  6 km sampling units and conducted 160 structured interviews with local residents. Interviews were designed to collect data on jaguar and seven prey species so that detection/non-detection matrices could be constructed for each sampling unit. Jaguars were reportedly detected in 57% of the sampling units and had a detection probability of 28%. With the exception of white-lipped peccary, prey species were reportedly detected in 82–100% of the sampling units. Though the use of interview data may violate some assumptions of the occupancy modeling approach for determining ‘proportion of area occupied’, we countered these shortcomings through study design and interpreting the occupancy parameter, psi, as ‘probability of habitat used’. Probability of habitat use was modeled for each target species using single state or multistate models. A combination of the estimated probabilities of habitat use for jaguar and prey was selected to identify the final jaguar corridor. This protocol provides an efficient field methodology for identifying corridors for easily-identifiable species, across large study areas comprised of unprotected, private lands. Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction The jaguar (Panthera onca) was previously divided into eight different subspecies (Pocock, 1939). However, recent genetic research has shown there is little justification for subspecific division and that dispersal between jaguar populations throughout their range is relatively high (Eizirik et al., 2001; Ruiz-Garcia et al., 2006). Landscape connectivity facilitates this dispersal and may in turn increase a population’s chances of persistence by counteracting genetic drift and inbreeding that can reduce diversity in small, isolated populations (Soulé and Mills, 1998; Young and Clarke, 2000). Given that high levels of genetic variation and exchange of genetic material are vital to the survival of most species (Frankham, 2006; Hedrick, 1995), maintaining existing corridors between jaguar populations should be integral to long-term conservation planning for this species. To determine where jaguar corridors might exist, we conducted a range-wide least-cost connectivity analysis between known

⇑ Corresponding author. Tel.: +1 646 786 0409; fax: +1 646 786 0401. E-mail address: [email protected] (K.A. Zeller). 0006-3207/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.biocon.2010.12.003

jaguar populations (see Rabinowitz and Zeller, 2010). Jaguar experts from Mexico to Argentina applied movement costs to six GIS-based landscape layers. These layers were combined to produce a movement cost surface quantifying the matrix between jaguar populations in terms of difficulty of jaguar movement. Leastcost corridors were then modeled across this cost surface between the 90 known jaguar populations identified in earlier range-wide priority-setting exercises (Sanderson et al., 2002; Zeller, 2007). These corridors identified probable connections between jaguar populations. However, because of errors inherent in remotely sensed data, such as misclassification and resolution issues, changes that may have taken place at the ground level since GIS data were collected, and limitations of least-cost corridor models (Beier et al., 2009; Theobald, 2006), field-based assessments are necessary to further refine corridor boundaries. Field assessments are also essential to confirm the use of the corridor by the species for which it is intended, allowing us to determine the appropriateness of the corridor (Hilty et al., 2006; Noss and Daly, 2006). Because corridors between jaguar populations exist primarily in human-dominated landscapes, usually spanning large spatial scales, it is often not practical to use conventional detection/nondetection techniques such as camera trapping or line transects

893

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

for field assessment. Theft or tampering of cameras, high human traffic, private land ownership, and the need for large amounts of staff time and funding are some of the obstacles to collecting data using these methods. An alternative is interviews with local people. Local people can be good sources of information about the presence or absence of wildlife (Rabinowitz, 1997; White et al., 2005). Well designed interviews can provide a credible, cost-effective alternative to large-scale field surveys for direct observation or sign of a species, especially for species that are rare and difficult to detect in a short survey period (Berg et al., 1983; Lariviére et al., 2000; Pike et al., 1999; van der Hoeven et al., 2004). In recent years, site occupancy modeling has become increasingly useful to ecologists because the only data requirements are based on the detection or non-detection of a species over several sampling occasions (Linkie et al., 2007; MacKenzie et al., 2002; Weller, 2008) and it provides for detection probability. Nocturnal and cryptic species, such as the jaguar, are not always detected when present and failure to incorporate detection probabilities can produce biased estimates, and lead to misguided conservation decisions (Gu and Swihart, 2004; Linkie et al., 2007; MacKenzie et al., 2006). Site occupancy modeling also provides a flexible framework that enables occupancy to be modeled as a function of covariate information (MacKenzie et al., 2002). The use of covariates, such as land cover type, elevation, and human presence, can provide valuable information about the factors influencing habitat use by jaguar and prey species in a human-modified landscape. As a potential indirect method to obtain detection/non-detection data for analysis in the site occupancy framework (MacKenzie et al., 2002) (also see Stanley and Royle, 2005), we propose the use of systematic and objective interviews with knowledgeable local residents. Though interview data may be an efficient way to gather data on species presence, the use of these data in the occupancy model presents a unique suite of challenges. Detections come from several sources whose credibility needs to be determined, and the quality of data collected through interviews is highly dependent on the wording of the questions and the skill of the interviewer (White et al., 2005). Furthermore, certain assumptions that allow for estimation of proportion and probability of area occupied of the MacKenzie et al. (2002) model cannot be met. This paper describes how we addressed those challenges through interview design and interpretation of model parameters. We demonstrate our methodology in a corridor site in eastern Nicaragua and provide an example of how we used the results to identify a biologically robust corridor between two known jaguar populations.

2. Materials and methods 2.1. Study area The study area lies between Wawashan National Park and Cerro Silva Forest Reserve on the Atlantic Coast of Nicaragua. Because the least-cost corridor provided limited connectivity options between the two patches, we wanted to be sure that we did not omit any possible connections that might be more circuitous, yet offer comparable habitat. We performed a closeness analysis of the landscape between the protected areas using FRAGSTATS (v3 McGarigal et al., 2002) and identified forest patches that contributed to the overall connectivity of the region. We combined these forest patches with the GIS-based least-cost corridor to create our study area (Fig. 1). The average annual rainfall and temperature are 4500 mm and 29 °C respectively. Elevation ranges from 0 to 231 m above sea level. Agricultural activities constitute nearly 15% of the study area whereas approximately 45% is covered by primary and secondary broadleaved forests. The study area is inhabited by peo-

Fig. 1. Study area.

ple at moderate densities, with over 35 settlements spanning the municipalities of Laguna de Perlas, Kukra Hill, Bluefields, and El Rama in the autonomous RAAS region of southern Nicaragua. The primary occupations of the residents vary from fishing, ranching and small scale farming, to working on large oil palm plantations. 2.2. Sampling approach We divided the study area into 71, 6  6 km, sampling units, which resulted in a final study area measuring 2556 km2 (Fig 1). Sampling unit size was based upon approximate home range size of jaguars in the region (Carrillo, 2000; Ceballos et al., 2002; Rabinowitz and Nottingham, 1986). We conducted interviews from February 19th to April 26th, 2009. Access issues and time constraints prevented us from conducting interviews in 4 of the 71 sampling units. We treated each interview as a separate replicate for the computation of detection probabilities. We used data on occupancy and detection probabilities from another likelihood-based habitat use study, on jaguars and prey in Costa Rica (Nijhawan, unpublished data), and the equations presented in MacKenzie and Royle (2005), to determine the optimum number of surveys needed at each site. This indicated the need for two to three replicates, or interviews, per sampling unit, which we increased to four to six, to account for variability in local conditions and lack of site-specific data. 2.3. Questionnaire and data collection We developed a questionnaire to gather detection/non-detection data on jaguar and the following prey species: white-lipped peccary (Tayassu pecari), collared peccary (Pecari tajacu), whitetailed deer (Odocoileus virginianus), red brocket deer (Mazama americana), Central American agouti (Dasyprocta punctata), paca (Agouti paca), and nine-banded armadillo (Dasypus novemcinctus). At the beginning of each interview, in an effort to gather unbiased information, we made each interviewee well aware of the fact the interviewers were from a non-governmental organization and that

894

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

all information they provided was anonymous. To ascertain the knowledge and credibility of the interviewees, we asked them to identify wildlife species through a two-step vetting process. First, we asked them to provide a detailed description of the animal and, second, we asked them to identify the animal and its tracks from an array of pictures. If an interviewee was not considered credible, data from that interview was not included in the analysis. This process helped ensure that we did not record any false positives for jaguar and prey detection. The local people selected for interviews acted as our surveyors. Each interviewee was associated with an area that they have ‘surveyed’. We referred to this area, represented by a single or a group of sampling units, as their ‘‘Area of Knowledge’’ (AOK). If an interviewee identified more than one sampling unit as their AOK, we asked the same set of questions for each sampling unit, thereby keeping a distinct sampling effort and detection/non-detection record for each sampling unit. For each interviewee, we determined and recorded frequency of visitation and amount of time spent in a sampling unit (Vongkhamheng, unpublished results). To meet our criteria for survey effort, an interviewee must have either lived in or ‘surveyed’ a sampling unit at least twice per month for a year. For both jaguar and prey, we recorded detections and nondetections for sightings that occurred within the last year and only for the sampling unit(s) identified by an interviewee as their AOK. We recorded detection for a sampling unit if any of the following were successfully established: direct sighting of the animal (alive or recently killed), direct observation of sign (tracks, burrows, or vocalizations), and in the case of jaguars, direct observation of a jaguar kill. Interviewers recorded the same jaguar sighting from two different interviewees only once. For prey, we also asked the interviewees how frequently the sightings occurred. We classified these responses into four sub-categorical variables representing frequency of sightings: undetected (not seen), rare (observed once a year), moderate/sometimes (observed twice a year to once per month/2–12 times a year) and frequent (observed more than once per month/13 times a year or more). 2.4. Data analysis and interpretation of estimates We used PRESENCE (v3.0; Hines, 2010) and ArcView (v9; Environmental Systems Research Institute, Redlands, CA) software to analyze the data. Interview responses resulted in detection/non-detection matrices for each species with a maximum of seven replicates per sampling unit including several missing values. MacKenzie and Nichols (2004) noted that the closure assumption of the occupancy model could be relaxed if changes to the population are random within the survey period. We were unable to meet the population closure assumption because interview data recorded jaguar and prey sightings over 1 year, during which considerable random fluxes in the occupancy of a corridor may have occurred. Additionally, violation of the independence assumption is unavoidable when detections are temporally spaced over a year. Relaxation of these conditions changes the interpretation of the occupancy parameter, psi, from true occupancy, or ‘‘proportion of area occupied’’ to ‘‘proportion of area used’’. Because we were interested in the use of a corridor by the species, not occupation of it, this was sufficient to meet the goals of our corridor assessment efforts. Henceforth, in this context, we are working with a state variable that should be interpreted as ‘‘habitat use (W)’’ not ‘‘occupancy’’ (MacKenzie and Nichols, 2004). 2.4.1. Single state and multistate models Naïve occupancy (proportion of sampling units where a species was detected) for jaguar and white-lipped peccary was 0.57 and

0.38 respectively. We analyzed jaguar and white-lipped peccary sightings using a single state model (MacKenzie et al., 2002). Data for the other prey species indicated detection in 82–100% of the sampling units. The high naïve occupancy of these species may be due to the fact that they were widespread in the study area. Alternatively, because prey species like collared peccary, deer, agouti, paca, and armadillo use much smaller areas than our sampling unit size, it is possible that models for these species provide a blanket 100% probability of habitat use for all sampling units even though they might have only used a section of a sampling unit. To help address this issue, we analyzed data for remaining prey species using multistate models (MacKenzie et al., 2009; Nichols et al., 2007). Interviewee – provided data on observation frequency were used as the different ‘states’. We assumed the more frequently a species or its sign was observed, the higher its relative abundance in a sampling unit. Thus, we converted the four sub-categorical frequency variables mentioned above into three relative abundance states: undetected (State 0), rare to moderately present (State 1), and frequently present (State 2). Realistically, there may be ambiguity not only in the detection or non-detection of the species from the field observations, but also in assignment of the correct state (MacKenzie et al., 2009; Nichols et al., 2007). The reported states are hierarchical in terms of certainty of the information about the true state, where the lowest observed state has the greatest uncertainty and the highest observed state has no ambiguity about its veracity. This means that if State 0 is observed, there is a chance that the species was present in State 1 or 2, but was undetected. However, if a species is detected in State 2 there is no uncertainty associated with it being the true state. We used multistate models developed by Nichols et al. (2007) to estimate the probability of presence in each of the three states. The states are considered to be mutually exclusive, therefore, their probabilities of habitat use must sum to one. 2.4.2. Covariates and a priori hypotheses Covariates that were considered to influence the use of a sampling unit by our target species were selected for analysis. Habitat covariates included proportion of forest, grassland, agriculture/ shrub, open areas (bare ground and pastures), early-stage forest re-growth, wetland, and water. Topographic and anthropogenic covariates included mean elevation of a sampling unit and distance to protected areas. Mean elevation was centered around the sample average to improve its interpretation in models. Roads were not used as a covariate because of lack of accurate information. Because prey species, especially smaller prey, were usually observed very close to settlements, we used number of settlements in a sampling unit as a covariate only for jaguar models. We modeled detection probability as a function of interviewee-specific covariates such as length of residency (or visitation), survey effort, as well as a function of habitat covariates. We only used one covariate if two or more covariates were found to be correlated. With the exception of Foster et al. (2010), limited research has been conducted on jaguar habitat use outside of protected areas in Central America. Therefore, we fit jaguar models as functions of all covariates. For prey species, we selected covariates most likely to influence distribution based on prior knowledge of their ecology rather than a data dredging approach (Anderson et al., 2001). For white-lipped peccary, we tested covariates related to their affinity for forest and water (Fragoso, 1999; Keuroghlian and Eaton, 2008; Sowls, 1997) and avoidance of anthropogenic disturbance (Altrichter and Boaglio, 2004; Fragoso, 1999; Sowls, 1997). For collared peccary, we tested covariates related to their affinity for water, wetlands (Keuroghlian and Eaton, 2008; Reid, 1997), forest (Altrichter and Boaglio, 2004; Fragoso, 1999), and habitat types such as agriculture and open areas owing to their broad habitat preferences (Naranjo, 2002; Sowls, 1997). For red brocket deer,

895

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

we tested covariates related to their association with forest and water (Emmons and Feer, 1997; Reid, 1997). Other prey species like white-tailed deer, agouti, paca and armadillo are known to be habitat generalists (Emmons and Feer, 1997; Méndez, 1984; Reid, 1997). For these species, we fit general models (including all the habitat covariates). We modeled habitat use and detection probabilities for single state and multistate models as a function of covariates using logit and multinomial-logit link (mlogit) functions respectively (MacKenzie et al., 2006, 2009). Each state was modeled separately with its own set of habitat covariates. mlogit models are a simple extension of logit models. For example, data for a species is recorded in m states such that m = 2, . . . , M. The logit of the probability of use of site j in state m is expressed as:

Z mj ¼ logitðWmj Þ ¼ amj þ

n X

bmji X mji

i¼1

which is a linear function of the n covariates (X1 . . . Xn) associated with site j, with one intercept term a and n b coefficients to be estimated. The lowest state is usually the reference state, in our case that would be State 0 or non-detection. The probability of membership in other states is compared to the probability of membership in the reference state. Therefore, the probability of habitat use in site j in state p (m = 2 . . . M) is expressed as:

Wj ðm ¼ pÞ ¼

1þ

expðZ pj Þ PM m¼2 expðZ mj Þ

Note that, when m = 2, the mlogit and logit regression models become one and the same. 2.4.3. Conditional probability (W-cond) Logit-based (or mlogit-based) model-estimated W is unconditional, meaning it is calculated as a function of modeled habitat covariates regardless of detection of the species in a sampling unit. The only conditioning is on the set of sampling unit characteristics. Conditional probability (W-cond) is computed taking detection history into account and is only appropriate for sampling units where data was collected. For single state models, W-cond = 1 for sampling units with at least one detection because species presence was unambiguously established. Conditional probability for sites with no detections is computed by factoring in the probability of not detecting the species (1  p):

W-cond ¼

WQ ð1  W þ W  Q Þ

where Q = (1  p)k, k = number of surveys. W-cond for states in a multistate model can be computed by an expression (similar to the one given above) where the denominator includes all possible iterations of the observed detection history and the numerator gives the probability associated with one of the states. For example, let po,r be the probability of detection when the observed state is o and the true state is r. For a detection history of 10101-,

W-condstate2 ¼

Wstate2 ð1  p1;2  p2;2 Þ4 Wstate2 ð1  p1;2  p2;2 Þ4 þ Wstate1 ð1  p1;1 Þ4 þ ð1  Wstate2  Wstate1 Þ

The additional term in the denominator in the expression above represents a scenario under which the species was absent from the unit (State 0). Missing observations are omitted in the calculation. 2.4.4. Model selection and model averaging We ranked models using the small-sample correction to Akaike’s Information Criterion (AICc). We computed Akaike weights (w) to compare weight of evidence among models in the candidate set (Burnham and Anderson, 2002). We eliminated models where b coefficients of covariates were statistically insignificant (at 95% confidence interval). However, we kept covariates with statistically insignificant b coefficients if they had been previously shown to influence species distribution and their sign agreed with the existing ecological knowledge about the species. Whenever there were a number of candidate models with relevant covariates and with similar AIC weights, we applied a model averaging technique to estimate probabilities of habitat use and detection (Buckland et al., 1997; Burnham and Anderson, 2002). 2.5. Predictive maps and corridor identification We used W-cond in State 2 (W-condstate2) for species with multistate models. We divided prey species into two groups based on their size. Group I included the smaller species, agouti, paca, and armadillo. Group II included the larger species, red brocket deer, white-tailed deer, collared peccary and white-lipped peccary. We multiplied probabilities of habitat use of species in Group I to obtain W-condGI = (Wagouti  Wpaca  Warmadillo), the probability that all of the smaller prey species use a sampling unit. We then calculated the probability that at least two of the larger prey species use a sampling unit to obtain W-condGII = 1  (Probability of none of the species using a sampling unit)  (Probability of at most one species using a sampling unit). We multiplied W-condGI and W-condGII to get the probability that all of the small prey species and at least two of the large prey species use a sampling unit. We examined various combinations of thresholds for the prey product (W-condGI  W-condGII) with thresholds of W-cond for jaguar to ascertain which grid cells would form a conservative, yet continuous corridor. After several iterations, every cell where prey product was greater than 0.9 (90% probability), and jaguar probability of habitat use was greater than 0.75 (75% probability) was considered for inclusion in the final corridor. We compared these results with the GIS-based corridor by calculating the area of overlap.

3. Results

W  condstate2 ¼

third and fifth surveys, or the unit may have been in State 1 with the species undetected in the second and fourth surveys and State 1 observed in the first, third and fifth surveys. Similarly, for detection history 0000- -,

Wstate2 ð1  p1;2  p2;2 Þ2 ðp1;2 Þ3 Wstate2 ð1  p1;2  p2;2 Þ2 ðp1;2 Þ3 þ Wstate1 ð1  p1;1 Þ2 ðp1;1 Þ3

The denominator in the expression of W-condstate2 implies that the unit may have been in State 2, with the species undetected in the second and fourth surveys and State 1 observed in the first,

We conducted 160 interviews over a 10-week period. Residency of the interviewees in the study area ranged from 4 months to 80 years with a median residence of 15 years. Males comprised 97.22% of the interviewees, the youngest interviewee was 16 years old, the oldest was 80 years old and the mean age of the interviewees was 40 years.

896 Table 1 Top models and untransformed coefficients of covariates for jaguar, white-lipped peccary, collared peccary and red brocket deer. Jaguar and white-lipped peccary data were analyzed using single state models. Other prey species were modeled using multistate models (State 2 represents the highest relative abundance state, State 0 represents absence). wb

0.00 1.74 2.96 3.80 7.08

0.53 0.22 0.12 0.08 0.02

W(forest + c.elev + forestc.elev), p(.) W(forest + c.elev + forestc.elev), p(l.stay) W(forest + c.elev + forestc.elev + dist_PA), p(.) W(forest + forestc.elev), p(.)

0.00 0.91 1.90 2.66

0.44 0.28 0.17 0.12

Collared peccarye 1 2 3

Naïve estimated: 0.98 W[W(1), W(2) = shrub], p(.) W[W(1), W(2) = forest + shrub], p(.) W[W(1), W(2) = forest + water], p(.)

0.00 0.50 1.23

0.43 6 0.34 7 0.23 7

Red brocket deere 1

Naïve estimated: 0.82 W[W(1), W(2) = forest + wetland + 10wetlandforest], p(.) W[W(1), W(2) = forest + water + 10waterforest], p(.) W[W(1), W(2) = forest + wetland], p(.)

0.00

Models

Jaguar 1 2 3 4 5

Naïve estimated: 0.57 W(.), p(c.elev) W(c.elev), p(.) W(c.elev), p(effort) W(c.elev), p(water) W(open), p(water)

White-lipped peccary 1 2 3 4

Naïve estimated: 0.38

2 3

Kc Untransformed coefficients of covariates (standard errors) Intercept

Forest

Water

c.elev

Dist.PA

Additional

3 3 4 4 4

2.06 3.54 3.21 4.25 3.24

– – – – –

– – – 0.44 (1.04) 1.62 (0.97)

0.04 0.15 0.14 0.18 –

– – – – –

Effort: 0.34 (0.34)

5 6 6 4

2.43 2.51 1.73 2.10

9.79 (4.54) 9.79 (4.30) 10.13 (4.93) 7.42 (3.24)

– – – –

0.16 (0.09) 0.16 (0.09) 0.16 (0.09) –

– – 0.06 (0.08) –

Forestc.elev: Forestc.elev: Forestc.elev: Forestc.elev:

35.61 (6.67) 27.94 (3.37) 23.24 (4.34)

– 7.17 (3.34) 7.68 (2.82)

– – 6.07 (3.40)

– – –

– – –

Shrub: 20.99 (9.65) Shrub: 6.05 (2.58)

0.65 8

2.05 (6.52)

15.40 (7.30)

–

–

–

2.93 (3.78)

13.77 (6.01)

34.32 (15.98)

–

–

Wetland: 45.08 (21.71); 10forestwetland: 12.27(6.35) 10forestwater: 9.23 (4.66)

2.17

0.22 8

3.24

0.13 7

0.58 (2.98)

8.23 (3.42)

–

–

–

Wetland: 10.92 (8.65)

(0.87) (2.58) (1.94) (5.48) (1.28)

(1.26) (1.23) (1.54) (0.99)

(0.01) (0.10) (0.08) (0.20)

Open: -13.66 (5.75)

0.65 0.63 0.65 0.18

(0.32) (0.30); l.stay: 0.01 (0.01) (0.33) (0.07)

Covariates: forest, proportion of forest in a sampling unit; open, proportion of bare ground and pasture in a sampling unit; water, proportion of water bodies and wetland in a sampling unit; wetland, proportion of wetland in a sampling unit; dist.PA, distance to the edge of the nearest protected area from centroid of a sampling unit; l.stay, length of residency of an interviewee in the study area; effort, sampling frequency of interviewee expressed as a ratio of number of days spent in a sampling unit annually to total number of days in a year; c.elev, mean elevation in a sampling unit centered around sample mean; shrub, proportion of shrubby and agricultural vegetation in a sampling unit. a Difference in AICc value relative to the top model. b AICc weight. c Number of parameters in the model. d Proportion of sampling units where the species was detected. e For multistate models, b coefficients are for State 2.

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

D AICca

Species

897

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901 Table 2 Habitat use estimates for more common prey species. Models fit as a function of habitat covariates for these species were misleading due to the reported commonness in the study area. Therefore, the simple model, W[W(1), W(2)],p(.), was selected as the final model. W(2) is the probability of habitat use in State 2. Species

Naïve estimate

W(2) (S.E. (W(2)))

White-tailed deer Central American agouti Paca Nine-banded armadillo

0.99 1.00 1.00 0.99

0.99 0.98 0.99 0.94

(0.01) (0.01) (0.01) (0.14)

Forty-four of the 71 (62%) sampling units qualified for inclusion in the corridor, totaling an area of 1584 km2, 521 km2 of which overlapped with the least-cost-corridor from the GIS analysis. The final corridor boundary was drawn from the protected area boundaries around all adjacent qualifying sampling units (Fig. 3a). This resulted in the inclusion of 33 qualifying sampling units and four sampling units that did not qualify. The final corridor measured 1332 km2. The overlap between the least-cost corridor and the final corridor boundary was 443 km2, or 33% (Fig. 3b).

3.1. Jaguar 4. Discussion Mean elevation and open areas, were included in the top models for the jaguar (Table 1). The b coefficients of both the covariates were negatively correlated with W suggesting jaguars avoid higher elevation areas and areas with no vegetative cover. The top models also indicated that detection probability is higher in lower elevation areas and may increase with an increase in interviewee survey effort and the proportion of water. 3.2. Prey Three covariates appeared important in habitat use models for white-lipped peccary: proportion of forest, mean elevation and distance to protected areas (Table 1). The b coefficients in the top models suggest the probability of habitat use by white-lipped peccary increased with increase in the proportion of forest in a sampling unit and decreased as the distance from protected areas increases, supporting our a priori predictions. One of the top models suggested interviewees’ length of residence in the study area might increase the chances of detecting white-lipped peccary. For collared peccary, three multistate models received the most support (Table 1). The covariates selected suggest that in the highest state, State 2, habitat use by collared peccary was inversely proportional to the proportion of agriculture, and directly proportional to forest and water, both in the form of permanent water sources and wetlands. For red brocket deer, top multistate models suggest the probability of habitat use in State 2 was positively correlated with proportion of forest and water; however, wetland or wetter areas were preferred more than permanent water bodies. The negative coefficient on the interaction terms, forest and water, suggests the red brocket deer was less likely to be present if one of the two habitat types were absent from a sampling unit. White-tailed deer, agouti, paca and armadillo (naïve occupancies >98%, Table 2), were detected in the highest state, State 2, in nearly 90% of the sampling units. The simple model was selected as the best model for these species. Detection probabilities were estimated as very high (>70% in State 2) except for armadillo (Table 3). 3.3. Predictive maps and corridor identification Using the averaged top models for jaguar, white-lipped peccary, collared peccary, and red brocket deer, and the simple models for the remaining prey species, we developed probability of habitat use maps (Fig. 2).

Corridors are often large and it is not uncommon for a project to lack the personnel and monetary needs to cover such extensive areas. There is also a general lack of methodological examples in the literature, no widely accepted protocols, and few practical examples of field assessment of wildlife corridors. The protocol presented here is a cost and time effective way to assess corridors for one or more easily recognizable target species. The technique can be used in areas that traverse private lands and has the added advantage of incorporating detection probabilities and habitat covariates into predictive models. It should be noted that this protocol was designed to gather and provide information about jaguars and their prey in the context of assessing and refining a corridor specific to jaguar dispersal. This protocol would likely not be appropriate for use within core populations, areas where human presence is scarce, or studies where true occupancy or abundance estimates are of key interest. We have identified two main improvements to this protocol. First, detection probability for jaguars was only 28%, which may suggest the need for more interviews per sampling unit (at least six), to achieve an optimal sampling effort (MacKenzie and Royle, 2005). Second, we suggest including more interviewee specific characteristics so that variation in detection probability, due to different sources, can be explained. In the surveys we are currently conducting, we have started collecting additional interviewee attributes such as, reason for visitation, mode of transportation, length of time spent in wildlife habitat and seasonal differences in visitation. Furthermore, in an effort to maintain independent interviewee responses, we refrain from interviewing people who travel together, hunt together, or work together. To our knowledge, Karanth et al. (2009) is the only other example of using interview data within an occupancy model. The use of interviews for estimating W needs to be measured against robust data collection methods such as line transects and camera traps. We are currently conducting such a study in Belize as well as testing the usefulness of this protocol in other corridors to determine its applicability in a wide range of scales, habitats, levels of fragmentation, and species presence. The data for white-lipped peccary indicated, as expected, a preference for forested habitats and close proximity to protected areas, though their high preference for water bodies was not reflected in our data (Fragoso, 1999; Keuroghlian and Eaton, 2008; Sowls, 1997). The models for collared peccary and red brocket deer showed, as predicted, a positive correlation between probability

Table 3 Model averaged detection probability of jaguar and prey species. Jaguar and white-lipped peccary data were analyzed using single state models. Other prey species were modeled using multistate models. p(o,r) represents detection probability where o is the observed state and r is the true state. Probability of detection

Jaguar

White-lipped peccary

Collared peccary

Red brocket deer

White-tailed deer

Paca

Agouti

Armadillo

p(1,1)/p p(1,2) p(2,2)

0.28 (0.04) – –

0.21 (0.05) – –

0.59 (0.10) 0.28 (0.03) 0.42 (0.04)

0.24 (0.08) 0.49 (0.04) 0.13 (0.03)

0.80 (0.20) 0.23 (0.03) 0.73 (0.03)

0.87 (0.02) 0.27 (0.03) 0.64 (0.03)

0.46 (0.18) 0.16 (0.02) 0.75 (0.03)

0.87 (0.16) 0.43 (0.01) 0.43 (0.01)

898

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

Fig. 2. Estimated probabilities of habitat use. We calculated W-cond values for single state models for jaguar and white-lipped peccary, and W-condstate2 for the remaining prey species. We used model-estimated W for the sampling units that were not sampled.

of use and proportion of forest cover and water. The simple model was the top model for white-tailed deer, paca, agouti and armadillo, all known generalist species (Emmons and Feer, 1990; Méndez, 1984; Reid, 1997). The highest ranked model for jaguars did not include any habitat covariates. Other models indicated a weak negative correlation with open areas and elevation suggesting a lack of specific habitat preference. Jaguars in this area may be responding more strongly

to prey locations and movement than to a particular set of habitat variables. A telemetry study on jaguars in the Venezuelan Llanos found no strong preference of habitat use versus availability, and that jaguars foraged in areas where prey was abundant, as opposed to feeding opportunistically (Scognamillo et al., 2003), a behavior which has been echoed in other jaguar studies (Polisar et al., 2003; Rabinowitz and Nottingham, 1986). Furthermore, it has been shown that large predators, such as jaguars, can utilize sub-optimal

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

899

Fig. 3. Corridor selection: (a) sampling units that qualified for inclusion in the jaguar corridor; (b) final corridor shown with GIS-based least-cost corridor from Rabinowitz and Zeller (2010).

habitats given sufficient prey and vegetative cover (Saberwal et al., 1994; Soni, 2000; Vijayan and Pati, 2002). Currently, there are no data on how long jaguars might spend during a dispersal event (with the exception of Quigley and Crawshaw, 2002) or their resource needs during dispersal. Therefore, we had limited guidance on which to base our corridor thresholds. We were faced with two choices – assume jaguars could, and would travel the length of the corridor within a few days and use solely habitat use estimates for jaguar, or, assume jaguars might need a food source during dispersal and incorporate prey presence as a factor for corridor identification. The inclusion of prey is not only a more conservative corridor estimate, but it widens the applicability of the corridor to a suite of species, thereby increasing its contribution to biological conservation. Furthermore, the ability of jaguars to use multiple prey items along their movement paths may increase the chance of dispersal success for jaguars by potentially reducing livestock depredation (Polisar et al., 2003). We recognize the thresholds on probability of habitat use by prey and jaguar for corridor inclusion are slightly arbitrary however, in the absence of genetic data documenting movement and breeding between populations, use of the area by the target species for which the corridor is intended is one of the best signs of functional connectivity (Hilty et al., 2006; Noss and Daly, 2006), and high proba-

bility of presence of a variety of prey species of different sizes further indicates a relatively healthy corridor. Other corridors throughout jaguar range will likely have different levels of jaguar and prey presence, therefore the threshold for including a sampling unit within a corridor may vary depending on the region. This underscores the need for a more scientifically rigorous designation of the minimum requirements a sampling unit needs to have to be included in a corridor. The results of our analyses can be used to promote corridor conservation and inform management decisions because they present a strong argument, backed by scientific data, for a jaguar corridor. The resultant maps can also be used as the foundation from which future corridor implementation can be based and against which future conservation strategies can be measured. For example, Fig. 3a shows how targeted restoration efforts in a few non-qualifying sampling units in the southwestern part of the study area might result in widening the corridor and making connectivity through this area more robust. In addition, immediate conservation efforts are likely needed in the sampling units that did not qualify for corridor inclusion, yet are still within the corridor due to their location. We are currently using the protocol presented here to assess jaguar corridors throughout Central America and Colombia. We have collected data in nine study areas covering over 20,000 km2

900

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901

and have begun addressing site-specific threats to corridors in Costa Rica, Panama, Honduras, and Belize (see Salom-Peréz et al., in press, for a discussion on corridor implementation strategies). The results of our work emphasize the importance of field-verifying coarse-scale GIS-based corridor exercises. Though the leastcost corridor may provide adequate connectivity for jaguars, our research identified a more functional connection to the east, which only overlapped with the least-cost corridor by 33%. The risk in not conducting a corridor assessment is that a poorly designed corridor will have less chance of success, resulting in a waste of precious conservation resources (Noss and Daly, 2006). The protocol presented in this paper can be adapted to other wide-ranging carnivores, providing an efficient tool for large-scale connectivity assessment and conservation planning.

Acknowledgements We are grateful to the Wildlife Conservation Society, the Liz Claiborne and Art Ortenberg Foundation, Environmental Systems Research Institute, Inc., the United States Department of State, and T. Kaplan and Panthera for funding and support of this work. We would also like to thank J.D. Nichols for his input in the protocol development, and J. Smith, L. Hunter, H. Quigley, M. Linkie and the journal reviewers for critical review of this manuscript. We are indebted to A. Rabinowitz for his jaguar corridor vision and encouragement, J. Polisar for valuable input during the early stages of this work, and F. Diaz and L. Maffei for their assistance in the field.

References Altrichter, M., Boaglio, G.I., 2004. Distribution and relative abundance of peccaries in the Argentine Chaco: associations with human factors. Biological Conservation 116, 217–225. Anderson, D.R., Burnham, K.P., Gould, W.R., Cherry, S., 2001. Concerns about finding effects that are actually spurious. Wildlife Society Bulletin 29, 311–316. Beier, P., Majka, D.R., Newell, S.L., 2009. Uncertainty analysis of least-cost modeling for designing wildlife linkages. Ecological Applications 19, 2067–2077. Berg, R.L., McDonald, L.L., Strickland, M.D., 1983. Distribution of mountain lions in Wyoming as determined by mail questionnaire. Wildlife Society Bulletin 11, 265–268. Buckland, S.T., Burnham, K.P., Augustin, N.H., 1997. Model selection: an integral part of inference. Biometrics 53, 603–618. Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A Practical Information-theoretic Approach, second ed. Springer, New York. Carrillo, E., 2000. Ecology and Conservation of White-lipped Peccaries and Jaguars in Corcovado National Park, Costa Rica. PhD Thesis, University of Massachusetts. Ceballos, G., Chávez, C., Rivera, A., Manterota, C., Wall, B., 2002. Tamaño poblacional y conservación del jaguar en la reserva de la biosfera Calakmul, Campeche, México. In: Medellín, R., Equihua, C., Chetkiewicz, C.L.B., Crawshaw, P.G., Jr., Rabinowitz, A., Redford, K.H., Robinson, J.G., Sanderson, E.W., Taber, A.B. (Eds.), El Jaguar en el Nuevo Milenio. Fondo de Cultura Económica, Universidad Nacional Autónoma de México, and the Wildlife Conservation Society, México, pp. 403–418. Eizirik, E., Kim, J., Menotti-Raymond, M., Crawshaw Jr., P.G., O’Brien, S.J., Johnson, W.E., 2001. Phylogeography, population history and conservation genetics of jaguars (Panthera onca, Mammalia, Felidae). Molecular Ecology 10, 65–79. Emmons, L.H., Feer, F., 1997. Neotropical Rainforest Mammals: A Field Guide, second ed. The University of Chicago Press, Chicago. Foster, R., Harmsen, B.J., Doncaster, C.P., 2010. Habitat use by sympatric jaguars and pumas across a gradient of human disturbance in Belize. Biotropica 42, 724– 731. Fragoso, J.M.V., 1999. Perception of scale and resource partitioning by peccaries: behavioral causes and ecological implications. Journal of Mammalogy 80, 993– 1003. Frankham, R., 2006. Genetics and landscape connectivity. In: Crooks, K.R., Sanjayan, M. (Eds.), Connectivity Conservation. Cambridge University Press, Cambridge, pp. 72–96. Gu, W., Swihart, R., 2004. Absent or undetected? Effects of non-detection of species occurrence on wildlife–habitat models. Biological Conservation 116, 195–203. Hedrick, P.W., 1995. Gene flow and genetic restoration: the Florida panther as a case study. Conservation Biology 9, 996–1007. Hilty, J.A., Lidicker Jr., W.Z., Merenlender, A.M., 2006. Corridor Ecology: The Science and Practice of Linking Landscapes for Biodiversity Conservation. Island Press, Washington, DC.

Hines, J.E., 2010. Program PRESENCE (Version 3.0). . Karanth, K.K., Nichols, J.D., Hines, J.E., Karanth, K.U., Christensen, N.L., 2009. Patterns and determinants of mammal species occurrence in India. Journal of Applied Ecology 46, 1189–1200. Keuroghlian, A., Eaton, D.P., 2008. Fruit availability and peccary frugivory in an isolated Atlantic forest fragment: effects on peccary ranging behavior and habitat use. Biotropica 40, 62–70. Lariviére, S., Jolicoeur, H., Crête, M., 2000. Status and conservation of the gray wolf (Canis lupus) in wildlife reserves of Québec. Biological Conservation 94, 143– 151. Linkie, M., Dinata, Y., Nugroho, A., Haidir, I.A., 2007. Estimating occupancy of a data deficient mammalian species living in tropical rainforests: sun bears in the Kerinci Seblat region, Sumatra. Biological Conservation 137, 20–27. MacKenzie, D.I., Nichols, J.D., 2004. Occupancy as a surrogate for abundance estimation. Animal Biodiversity and Conservation 27, 461–467. MacKenzie, D.I., Royle, J.A., 2005. Designing efficient occupancy studies: general advice and tips on allocation of survey effort. Journal of Applied Ecology 42, 1105–1114. MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, A., Langtimm, C.A., 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83, 2248–2255. MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Hines, J.E., Bailey, L.L., 2006. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Elsevier, San Diego, CA. MacKenzie, D.I., Nichols, J.D., Seamans, M.E., Gutiérrez, R.J., 2009. Modeling species occurrence dynamics with multiple states and imperfect detection. Ecology 90, 823–835. McGarigal, K., Cushman, S.A., Neel, M.C., Ene, E., 2002. FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps. . Méndez, E., 1984. Mexico and Central America. In: Hals, L.K. (Ed.), White-tailed Deer Ecology and Management. Stackpole Books, Harrisburg, PA, pp. 513–524. Naranjo, E.J., 2002. Population Ecology and Conservation of Ungulates in the Lacandon Forest, México. PhD Thesis, The University of Florida, Gainesville. Nichols, J.D., Hines, J.E., MacKenzie, D.I., Seamans, M.E., Gutierrez, R.J., 2007. Occupancy estimates and modeling with multiple states and state uncertainty. Ecology 88, 1395–1400. Nijhawan, S., Unpublished results. Analysis of Interview Data from the Barbilla Corridor, Costa Rica, using the Occupancy Framework. Panthera Internal Report, New York. Noss, R.F., Daly, K.M., 2006. Incorporating connectivity into broad-scale conservation planning. In: Crooks, K.R., Sanjayan, M. (Eds.), Connectivity Conservation. Cambridge University Press, Cambridge, pp. 587–619. Pike, J.R., Shaw, J.H., Leslie Jr., D.M., Shaw, M.G., 1999. A geographic analysis of the status of mountain lions in Oklahoma. Wildlife Society Bulletin 27, 4–11. Pocock, R.I., 1939. The races of jaguar (Panthera onca). Novitates Zoologicae 41, 406– 422. Polisar, J., Maxit, I.E., Scognamillo, D., Farrell, L., Sunquist, M.E., Eisenberg, J.F., 2003. Jaguars, pumas, their prey base, and cattle ranching: ecological interpretations of a management problem. Biological Conservation 109, 297–310. Quigley, H.B., Crawshaw, P.G., 2002. Reproducción, creimiento y dispersión del jaguar en la región del pantanal de brasil. In: Medellín, R., Equihua, C., Chetkiewicz, C.L.B., Crawshaw, P.G., Jr., Rabinowitz, A., Redford, K.H., Robinson, J.G., Sanderson, E.W., Taber, A.B. (Eds.), El Jaguar en el Nuevo Milenio. Fondo de Cultura Económica, Universidad Nacional Autónoma de México, and the Wildlife Conservation Society, México, pp. 289–302. Rabinowitz, A., 1997. Wildlife Field Research and Conservation Training Manual. Wildlife Conservation Society, Bronx, New York. Rabinowitz, A., Nottingham, B., 1986. Ecology and behavior of the jaguar in Belize, Central America. Journal of Zoology 210, 149–159. Rabinowitz, A., Zeller, K.A., 2010. A range-wide model of landscape connectivity and conservation for the jaguar, Panthera onca. Biological Conservation 143, 939– 945. Reid, F.A., 1997. A Field Guide to the Mammals of Central America and Southern Mexico. Oxford University Press, Oxford. Ruiz-Garcia, M., Payan, E., Murillo, A., Alvarez, D., 2006. Microsatellite characterization of the jaguar (Panthera onca) in Colombia. Genes & Genetic Systems 81, 115–127. Saberwal, V.K., Gibbs, J.P., Chellam, R., Johnsingh, A.J.T., 1994. Lion–human conflict in the Gir Forest, India. Conservation Biology 8, 501–507. Salom-Peréz, R., Polisar, J., Quigley, H., Zeller, K.A., in press. Iniciativa del Corredor del Jaguar: un corredor biológico y un compromiso a largo plazo para la conservación. Mesoamericana. Sanderson, E.W., Redford, K.H., Chetkiewicz, C.B., Medellin, R.A., Rabinowitz, A.R., Robinson, J.G., Taber, A.B., 2002. Planning to save a species: the jaguar as a model. Conservation Biology 16, 58–71. Scognamillo, D., Maxit, I.E., Sunquist, M.E., Polisar, J., 2003. Coexistence of jaguar (Panthera onca) and puma (Puma concolor) in a mosaic landscape in the Venezuelan llanos. Journal of Zoology 259, 269–279. Soni, V.C., 2000. Study of Satellite Population of Asiatic Lion, a Report. Gir National Park and Sanctuary, Forest Department, Gujarat State, India. Soulé, M.E., Mills, L.S., 1998. No need to isolate genetics. Science 282, 1658– 1659. Sowls, L.K., 1997. Javelinas and the Other Peccaries: Their Biology, Management and Use, second ed. Texas A&M University Press, College Station, TX.

K.A. Zeller et al. / Biological Conservation 144 (2011) 892–901 Stanley, T.R., Royle, J.A., 2005. Estimating site occupancy and abundance using indirect indices. Journal of Wildlife Management 69, 874–883. Theobald, D., 2006. Exploring the functional connectivity of landscapes using landscape networks. In: Crooks, K.R., Sanjayan, M. (Eds.), Connectivity Conservation. Cambridge University Press, Cambridge, pp. 416–443. Van der Hoeven, C.A., De Boer, W.F., Prins, H.H.T., 2004. Pooling local expert opinions for estimating mammal densities in tropical rainforests. Journal for Nature Conservation 12, 193–204. Vijayan, S., Pati, B.P., 2002. Impact of changing cropping patterns on man-animal conflicts around Gir protected area with specific reference to Talala SubDistrict, Gujarat, India. Population and Environment 23, 541–559. Vongkhamheng, C., Unpublished results. A Protocol for Questionnaire Survey in Nam Et-Phou Louey Tiger Conservation Landscape, Northern Lao PDR.

901

Weller, T.J., 2008. Using occupancy estimation to assess the effectiveness of a regional multiple-species conservation plan: bats in the Pacific Northwest. Biological Conservation 141, 2279–2289. White, P.C., Jennings, N.V., Renwick, A.R., Barker, N.H.L., 2005. Questionnaires in ecology: a review of past use and recommendations for best practice. Journal of Applied Ecology 42, 421–430. Young, A.G., Clarke, G.M., 2000. Conclusions and future directions: what do we know about the genetic and demographic effects of habitat fragmentation and where do we go from here? In: Young, A.G., Clarke, G.M. (Eds.), Genetics, Demography and Viability of Fragmented Populations. Cambridge University Press, Cambridge, England, pp. 361–366. Zeller, K.A., 2007. Jaguars in the New Millennium Data Set Update: The State of the Jaguar in 2006. Wildlife Conservation Society, Bronx, NY.