Data from: Predictive habitat occupancy models for North American river otters along inland streams in New Jersey
Data files
Jan 20, 2025 version files 170.40 KB
-
NJ_Otter_Occupancy_Data.csv
167.15 KB
-
README.md
3.24 KB
Abstract
The North American river otter (Lontra canadensis) is a semi-aquatic furbearer species that historically ranged throughout North America. Starting in the mid-1800s and continuing through the early 1900s, the negative effects associated with anthropogenic disturbances (i.e. overharvest, development and ultimately habitat alternation) led to local extinctions. Researchers debate whether current land use patterns are affecting river otter occupancy. New Jersey is the most densely populated state in the United States, thus it provides a perfect study area to test potential anthropogenic effects on river otters. Using occupancy modeling to examine river otter habitat preferences, we measured presence/absence at 244 low order streams from January-April 2011–2012 along with 19 corresponding site/landscape covariates in both Northern and Southern New Jersey. In Southern New Jersey, we detected otters at 83/141 sites (58.9%) with a detection probability of 97.7% across repeat visits and a predicted occupancy of 59.4 ± 0.04%. In Northern New Jersey we detected otters at 31/103 sites (30.1%) with a detection probability of 44.5% across repeat visits and a predicted occupancy of 58.8 ± 0.04%. We determined the influence of habitat covariates on otter occupancy and found that water depth, water quality, stream width and mink presence were positively correlated with otter occupancy. The % Commercial, Industrial, Transportation and Recreational habitat, % low intensity development, bank slope, and distance to lake were negatively correlated with otter occupancy. Knowing the location of occupied stream and latrine sites will assist biologists in their efforts to monitor river otter populations and help estimate river otter density for harvest and conservation efforts.
README: New Jersey River Otter Occupancy
https://doi.org/10.5061/dryad.4b8gthtpv
Description of the data and file structure
We chose fewer sites and smaller distances per bootstrapped replicate in Northern New Jersey to better accommodate the smaller area of the Northern study area. We used Student’s t-test (α = 0.05) to identify the habitat-specific landscape distances that exhibited the strongest correlation to river otter presence at the site. We used the smallest radius within that range as the scale for measuring that landscape variable (see Figure 2 as an example). Using the shortest distance reduced the overlap of landscape buffers between points. In addition to the 8 spatially explicit landscape covariates (% Agriculture; % Commercial, Industry, Transportation, Recreation; % High-Intensity Development; % Low-Intensity Development; % Upland Natural; % Water, % Wetland, % Other) we also measured 8 site level covariates including:
1) Average water depth at a site during the field season: (1) 0-15 cm, (2) 16-30 cm, (3) 31-45 cm, (4) 46-60 cm, (5) 61-75 cm, (6) 76-90 cm, (7) 91-105 cm, (8) >106 cm.
2) Average bank height, measured from the stream bed to the top of the stream bank, for the surveyed portion of the stream: (1) 0-31 cm, (2) 31-62 cm, (3) 62-93 cm, (4) 93-124 cm, (5) 124-155 cm, (6) 155-186 cm, (7) >186 cm.
3) Average bankfull height, measured from the top of the water line to the top of the bank, at a site during the field season: (1) 0-31 cm, (2) 31-62 cm, (3) 62-93 cm, (4) 93- 124 cm, (5) 124-155 cm, (6) 155-186 cm, (7) >186 cm.
4) Average bank slope for the surveyed portion of the stream: (1) <30 degrees, (2) 30-60 degrees, (3) >60 degrees.
5) Average stream width for the surveyed portion of the stream: (1) 0-2 m, (2) 2- 4 m, (3) 4-6 m, (4) 6-8 m, (5) 8- 10, (6) >10 m
6) Distance to closest lake or pond from the survey site: (0) 0-.5 km, (1) .5-1 km, (2) 1-1.5 km, (3) 1.5-2 km, (4) 2-2.5 km, (5) 2.5-3 km, (6) >3 km.
7) Overall water quality for a site, determined by AMNET/FIBI rating: (1) Poor, (2) Fair, (3) Good, (4) Excellent.
8) Overall habitat quality for a site, determined by AMNET/FIBI rating: (1) Poor, (2) Marginal, (3) Sub-optimal, (4) Optimal.This is data collected in 2010-2011 regarding otter occupancy in rivers in New Jersey.
The data is organized by the following columns:
- Site Number - A unique identifier for the stream point measured
- Visit - a value over 6 repeat visits
- Date
- Current Weather
- Previous Weather
- Start Time - of survey
- End Time - of survey
- % Ice Cover
- Ice Cover (Binary) - yes (1) or no (0)
- Snow Cover
- Snow Cover (Binary) - yes (1) or no (0)
- Precipitation
- Precipitation Binary - yes (1) or no (0)
- Approx water depth
- Average Water Depth
- Soil Type
- Bank Slope
- Average Bank Slope
- Bank Height
- Average Bank Height
- Stream Width
- Average Stream Width
- Otter sign
- Otter Sign Type/Location Substrate
- Otter Binary - yes (1) or no (0)
- Mink sign
- Mink Binary - yes (1) or no (0)
- Beaver sign
- Beaver Binary - yes (1) or no (0)
- Muskrat sign
- Muskrat Binary - yes (1) or no (0)
- Notes
Missing data: n/a
Methods
Site Selection
Potential sites were restricted to lower order streams that could be safely waded and/or had accessible stream banks that could be surveyed by foot. To identify a stratified distribution of random stream sites, we looked at streams that were close in proximity to sites classified as either Ambient Biomonitoring Network (AMNET, NJDEP 2011) or Index of Biotic Integrity (IBI)/Fish Index or Biotic Integrity (FIBI) sites (NJDEP 2012a) because the New Jersey Department of Environmental Protection (NJDEP) assessed the overall water quality and stream health which would allow us to use corresponding water quality, habitat quality, and fish density data as covariates for our occupancy models. Once the sites were selected, we selected transects that would include a road/bridge crossing of the stream (Roberts and Crimmins 2008, Crimmins et. al 2009) to ensure that we would be surveying areas with some level of human traffic. If some AMNET, FIBI, or otter harvest/sightings points were not on a road/bridge crossing, we moved our stream site to the nearest road/bridge crossing on the stream. We had a total of 244 unique sites surveyed throughout the two-year study (124 in 2011 and 120 in 2012) that were comprised of 147 AMNET, 24 IBI, 25 FIBI, 42 otter harvest/sightings, and 6 random stream sites with 141 located in Southern New Jersey and 103 located in Northern New Jersey.
Field Protocol
To sample the presence of otters, we conducted 6 repeat walking transect surveys 1 January - 30 April 2011 and 2012 every 14 ± 4 d at each survey site. The purpose of these 4 days of flexibility was twofold:1) to allow for the potential of missing survey days due to inclement weather such as snow, ice, and heavy rain and 2) to provide the opportunity for surveyors to be proactive and survey certain sites in advance in anticipation of extended periods of inclement weather. Additionally, depending on the physical characteristics of the sites, surveys were not conducted ≤ 2 d after heavy rain or snow/ice storms to ensure the safety of surveyors and because scat and track signs could be washed away or hidden under snow. We conducted 600 m long stream bank walking sign surveys, usually 300 m on each side of a bridge crossing (Roberts et al. 2008). If there were any impediments on a particular side of the crossing, such as a lake, extremely dense vegetation, or deep water, we walked an additional distance on the other side to make up the difference. We took GPS coordinates at each site and recorded the start/end time for each stream survey. For each replicate visit, we recorded current, overnight, and previous day’s temperature, the presence or absence of precipitation (snow/ice/rain) during the survey, overnight, and the previous day and the presence or absence of snow or ice cover at the sites at the time of the survey. Snow cover can both negatively and positively impact the ability of the surveyors to locate animal signs. We also recorded average site-level habitat characteristics including water depth and bankfull height (i.e. the water level, or stage, at which a stream, river or lake is at the top of its banks and any further rise would result in water moving into the flood plain), which were taken randomly within the transect at each replicate visit while stream width, bank height, and bank slope were only taken once per site.
While visiting each site, we made note of live otter sightings and the presence of otter tracks, scat, latrines, and slides. We also identified and recorded the substrate where that specific otter sign was found. GPS coordinates were taken for otter latrines and otter slides so that the NJDFW could monitor these locations in the future.
Water Quality Assessment of Sites
Water quality and habitat quality assessments were based on previously gathered and readily available AMNET (NJDEP 2011; NJDEP GIS 2013) and FIBI (NJDEP 2012a) data. Because not all survey points were AMNET or FIBI points (i.e. random points and otter harvest/sighting points mentioned above, N = 48), not all points had this background data that we could use. Therefore, we created the following guidelines to provide estimated water and habitat quality values for all survey points:
- For those points that were AMNET or FIBI, we used the corresponding water and habitat quality assessment scores/ratings for those sites.
2. For those points that were characterized as either random or otter harvest/sighting, we chose the closest AMNET or FIBI point to that survey point, with preference given to AMNET or FIBI points that were on the same stream as the survey point.
We also had to consider the year in which the AMNET and FIBI points were surveyed because these assessments are not done annually at every site making some variation regarding the year in which the water quality or habitat quality assessments are made. While AMNET/FIBI samples are taken every year, it takes 5 years to reach all sites. While assigning water and habitat quality values to a survey site, we chose the most recent available values within the five year intervals. Each survey site, was given a water quality rating of excellent, good, fair, or poor (via AMNET or FIBI rating). Further, survey sites were given habitat quality ratings (and corresponding scores) of optimal (160-200), suboptimal (110-159), marginal (60-109), and poor (<60) via the AMNET and FIBI scoring based on individual condition of 10 habitat parameters (NJDEP 2007).
Data Analysis: Occupancy and Predictive Habitat Modeling
We arbitrarily defined the “survey site” as the area within 600 m of the 600 m stream transect (i.e. 600 m upstream and 600 meters downstream totaling 1800 m in length). We quantified the AMNET/FIBI 10 site level habitat configuration and quality variables including: stream width, water depth, bank height, bankfull height, bank slope, distance to the closest lake/pond, AMNET/FIBI water quality, AMNET/FIBI habitat quality, as well as beaver and mink presence we observed in our transects.
For landscape level covariates, we reclassified the 84 land use/land cover types of the New Jersey Land Use Land Cover (NJLULC) dataset to into 8 summary habitat types defined by NJLUC including: 1) High Intensity Development; 2) Low Intensity Development; 3) Commercial, Industrial, Transportation and Recreational (CITR); 4) Agriculture; 5) Upland Natural; 6) Fresh Water; 7) Non-coastal Wetlands; and 8) Other. To quantify habitat composition, we measured the proportion of each habitat type (excluding Other) within the buffered transect area. For each covariate measured at the landscape scale, we determined the buffer radius around the site at which each covariate was most strongly correlated to otter presence among a set of buffer radii ranging from 0.6 km to 16.2 km (being within the range of home range estimates of the river otter) at 600 m increments for all the sites in each sampling region (Holland et al. 2004, Duren et al. 2011). We used bootstrapping to obtain Spearman’s Rank Correlation Coefficients on 10,000 random samples of 10 sites that were 25.2 km apart in Southern New Jersey and 9 sites that were 22.8 km apart in Northern New Jersey.
We estimated site occupancy and detection probability using the modeling approach of Mackenzie et al. (2002), which accounts for the probability of an individual occupying the site and being detected during a survey. We used Akaike’s Information Criterion (AIC) to evaluate and select models (Burnham and Anderson 2002) and performed analyses using the program PRESENCE (Hines 2006). First, we modeled detection probability among survey points considering four explanatory covariates: month, observer, snow cover, presence of precipitation, and additionally constant detection probability to overcome or avoid variation in detection probability, which was considered to be a nuisance parameter. We selected a best model from that analysis to control for detection probability for subsequent modeling of otter occupancy. We modeled otter occupancy for Northern and Southern New Jersey separately using logistic regression with the covariates of site and landscape scale metrics.
Candidate models were constrained by the following rules:
1) Each model, except for the initial individual covariate models, included at least one site-scale and one landscape-scale covariate. Models were not required to have the same number of site and landscape covariates. Finally, to avoid the construction of overly complex models, no model was allowed to have more than 6 total covariates.
2) Covariates that are correlated (|r| ≥ 0.5) were not used in the same model and we randomly chose one of the two correlated variables based on potential predictive relationship discussed in the literature.
3) All covariates were used at least once. Some covariates were used more than others because they repeatedly appeared in higher-ranked models. Covariates that consistently appeared in low-ranked models were used less often. Additional covariates were individually added to high-ranked models to determine their effect on the model. Covariates that did not improve the log-likelihood estimate of the model by a value of >2 were removed.
4) Within the set of candidate models, we also included a global model of all covariates and a null model in which occupancy was held constant.
We calculated the small-sample-corrected information criterion AICc for all models (Hurvich and Tsai 1989) because sample size was small with respect to the number of parameters (K) in the analyses. We performed multiple-model averaging to predict occupancy using those models that had substantial support for fitting the data given the candidate set of models to address model selection uncertainty (Burnham and Anderson 2002). To determine the direction and magnitude of effect sizes for covariates, we calculated the mean standardized partial regression coefficients across all the models containing the variables of interest, and estimated precision using an unconditional variance estimator that incorporates model selection uncertainty (Burnham and Anderson 2002, p. 162).
We ultimately validated the accuracy of our best-fit occupancy models for Southern and Northern New Jersey by constructing an error matrix as part of a classification analysis. This analysis required the use of observed and conditional occupancy values for every site in both sampling regions. We averaged conditional occupancy values for each site in both sampling regions across all models that had the best support for fitting the data (i.e. those models that were used in the model averaging process). Thus, each site in both regions had representative observed and conditional occupancy values. Using the constructed matrix, we determined the overall accuracy and our omission and commission errors for the best-fit occupancy models in both regions (Congalton 1991).