Integrating presence-only and detection/non-detection data to estimate distributions and expected abundance of difficult-to-monitor species on a landscape-scale

Estimating species distribution and abundance is foundational to effective management and conservation. Using an integrated species distribution model that combines presence-only data from various sources with detection/non-detection data from structured surveys, we estimated the distribution and expected abundance of difficult-to-monitor mammals of management concern across New York State, namely, coyotes, bobcats, and black bears. Three distinct landscape-scale camera trap surveys provided detection/non-detection data over nine years between 2013-2021, and we augmented those data with incidental records of our focal species from public repositories. We used an inhomogeneous Poisson point process to construct an integrated model that fit both data types simultaneously. We demonstrate a simple application of spatial point density of all species records in the accessed public databases to inform the thinning process to account for unknown spatial sampling in the presence-only data, often referred to as the “magic covariate”. Using this approach, we examine habitat associations and provide spatially explicit estimates of expected abundance across the entirety of New York state for all three focal species. 

As expected, coyotes were the most widely distributed and abundant species, with a strong positive association with agricultural land uses. Bobcats exhibited low expected abundance throughout the state and showed positive associations with deciduous forest and forest edge, and a negative association with road density. Finally, we observed considerable spatial variation in abundance of black bears with expected numbers increasing in association with various forest cover and composition covariates and decreasing with crop cover. We present insights into habitat associations and provide management implications for each of the species of interest. 

Our integrated modelling method allows for managers to use citizen sightings combined with detection/non-detection surveys to estimate robust indices of abundance for both high- and low-density, and wide-spread versus patchily distributed species. Through comparison with previous studies, we highlight how broad-scale programs, such as the statewide efforts to estimate species distributions undertaken here, can benefit substantively from integrated models that leverage additional data (here, incidental records) from a larger region of space, and thus capture more landscape heterogeneity than is plausible within formalized surveys.

Methods
Landscape-scale structured camera trap surveys

Multiple camera trap surveys spanning nine years documented the occurrence of mammals across much of the state from 2013 – 2021. We conducted winter surveys of the south-central part of the state including the High Allegheny Plateau, the Western Allegheny Plateau, and the Great Lakes ecoregions (Omernik & Griffith, 2014) in 2013 (294 sites), 2014 (608 sites), and 2015 (599 sites). We conducted a second set of large-scale winter camera surveys in the same region in 2019 (584 sites), 2020 (603 sites), and 2021 (601 sites). Winter surveys of the northern part of the state (Northern Appalachians ecoregion), were conducted annually in 2016 (189 sites), 2017 (179 sites), and 2018 (191 sites). All nine winter camera trap surveys from 2013-2021 were leveraged to generate detection/non-detection data for bobcats and coyotes. We additionally conducted summer camera trap surveys for black bears in the High Allegheny Plateau, Western Allegheny Plateau, and Lower New England ecoregions in 2017 (238 sites) and 2018 (242 sites).  The summer surveys were used to generate detection/non-detection data for black bears only. See Table 1 for full details on spatial scale, bait used, and sampling durations. Sites in the winter surveys were sampled using camera traps deployed opposite baited stations attached to trees. A site was defined as a 15 km2 grid cell in the winter surveys, and 25 km2 grid cells in the summer surveys, with the detector (camera traps) deployed as close to center of its respective grid cell as possible.  Weekly detection records were created for each focal species over the sampling period, resulting in 3 occasions for each site-x-year combination. In the summer surveys sites were sampled using a camera trap deployed opposite to a hair snare, which consisted of 2 strands of barbed wire set at 30cm and 60cm off the ground encircling 3-6 trees. Sites were checked to replace SD cards, replenish batteries, and apply new scent and bait attractants every 2 weeks, for a total of 10 weeks. Only one detection was permitted per 2-week period minimizing violations of independence assumptions.  

Presence-only background data

Additional records for the focal species were collated by accessing both online public repositories including the Global Biodiversity Information Facility (https://www.gbif.org/), iNaturalist (Research-Grade; https://www.inaturalist.org/), DataBasin (https://databasin.org/), eMammal (https://emammal.si.edu/), and Movebank ( https://www.movebank.org), as well as requesting data from governmental data stores namely NYSDEC Nuisance Wildlife Complaints. We used the search terms “bobcat”, “Lynx rufus”, “coyote”, “Canis latrans”, “black bear”, and “Ursus americanus”, and gathered all records from the sampling years (2013-2021) where specific coordinate locations were provided. For black bears, we leveraged an additional source of presence-only data, iSeeMammals, an on-going citizen project established for monitoring black bears in New York State in 2017 (Sun et al., 2021). All presence-only records were plotted and manually checked for validity. Records from iNaturalist were restricted to those labels as “research-grade”. Only records from the sampling years for each species were retained (2013-2021 for bobcat and coyote; 2017-2018 for black bear), with a maximum of one record / 10km2 pixel / week to ensure independence of observations (see spatial and observation covariates). Due to licensing issues the presence-only data from all sources apart from iSeeMammals could not be hosted by Dryad, please contact the corresponding author (Josh Twining, [e-mail]) for data queries or see the associated github (https://github.com/jptwining/PO-PA-Integrated-SDM).