Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
Forest and Wildlife Ecology
Non Technical Summary
This project is designed to provide land managers with guidance for optimizing forest management to recover forest-roosting bats following the white-nosed syndrome epizootic. White-nose syndrome is a fungal disease that causes mass mortality in bats that hibernate in caves or mines. In Wisconsin, two species of cave bats roost in buildings and bat houses during summer, while another two species primarily use forests for day roosts and foraging habitat. These forest-dwelling species, the northern long-eared bat (Myotis septentrionalis; NLEB) and the eastern pipistrelle (Perimyotis subflavus; EAPI), depend on forest resources for critical life-history processes that drive population growth. Conservationists expect a wave of local extirpations among cave-hibernating bats because of the disease, and consequently most hibernating bats are on State or Federal watch lists with varying levels of regulatory classification (e.g. species of concern, threatened, endangered). Recovery of these populations and restoration of the ecological benefits provided by bats will require land managers and conservationists to manage forests for range of factors related to habitat, addressing both mortality sources and optimizing recruitment to achieve positive population growth following the inevitable die-off.Unfortunately, basic life-history information on Wisconsin's threatened bats is relatively unknown. In the context of recovery, managers have little information on the spatial relationships between large essentially permanent winter hibernacula (caves, mines) and dispersed summer roosting sites that are temporally ephemeral (dead trees in an appropriate state of decay). Bats give birth and provision their young while using summer roosting habitat hence quality summer habitat is critically important for recruitment and population growth. Forest management to help recover post-epizootic bat populations could include the retention or creation of dead trees, but how close to hibernacula do they need to be? How soon after tree death is a dead tree useful for roosting and how long does a dead tree provide a suitable roosting resource? How does ecological context affect the potential use of a roost tree?This project takes an interdisciplinary approach to answering these and related questions by using remote sensing of forests (first discipline) and wildlife ecology (second discipline) to mount a rigorous multiscale analysis of the location of roosting habitat relative to landscape position of known bat hibernacula and suitable forest conditions (likely a juxtaposition of bat foraging habitat and presence of suitable dead trees in Wisconsin).
Animal Health Component
100%
Research Effort Categories
Basic
(N/A)
Applied
100%
Developmental
(N/A)
Goals / Objectives
The goal of this research is a validated predictive model of use of forests by bats in Wisconsin to help aid management for the recovery of bat populations following impacts White-Nose Syndrome disease. We hypothesize that: 1) large-scale selection of forest habitat will be driven by landscape context that reflects distances from established permanent hibernacula, forest cover types, and forest fragmentation; 2) fine-scale use of forest habitat (selection of roost trees) will be dependent on probability of large-scale selection and fine-scale measures of density of dead trees, as well local features such as soil type, and proximity to wetlands (because of their influence on arthropod prey).
Project Methods
We propose three discrete steps towards deriving a rigorous predictive model of summer roosting habitat for NLEBs and EAPIs: 1) Estimation of a landscape model of summer habitat selection in Wisconsin forests 2) Estimation of a roost-site selection model conditioned on landscape level selection, and 3) Evaluation and validation of landscape roost site selection models using independently generated data.Step one will be an evaluation of geo-referenced bat detections for NLEBs and EAPIs generated by the WI DNR's bureau of Natural Heritage (please see attached letter of support). The WI DNR database contains roughly 650 and 740 summer detections of NLEB and EAPIs (identified roost trees, capture records, acoustic detections) from throughout the state in addition to geo-referenced records of visits to forested sites where these bats were not found. These detections/non-detections will be treated as experimental observations and reflect mostly a pre-WNS condition. We will use Generalized Linear Models (GLMs) or occupancy modeling (MacKenzie et al. 2005) to model detection/non-detection probabilities as functions of landscape context and remotely sensed habitat features that vary at the scale of discrete forest stands (e.g. species composition, age, stocking density). Occupancy modeling has the advantage of enabling the analyst to estimate separately site-level probabilities of occupancy and detection but it requires repeated observations. Our decision to use occupancy modeling will depend on whether the database contains natural aggregations of observations (detections/non-detection) in space or time that can be treated as repeated observations and whether aggregated data produces a reasonable sample size and gives adequate coverage of Wisconsin's forests.Under a GLM analysis, observations that are within a 2 foraging radii for bats will be considered spatially dependent and we will use random effects to account for unmeasured spatial dependence. We will estimate 2 models, one for each species (NLEB and EAPI). Maps of habitat type, land cover, and fragmentation will be derived from the forthcoming 30-m Wiscland 2 product (dnr.wi.gov/maps/gis/datalandcover.html). Other data sources, such as the annual USDA National Agricultural Statistics Service (NASS) 30-m Cropland Data Layer (CDL) will also be used for cover type mapping (www.nass.usda.gov/Research_and_Science/Cropland/SARS1a.php).Step two will involve subsampling of archived geo-referenced bat observations used in step one to reflect a representative range of predicted landscape-level occupancy or occurence probabilities. The subsampled points will then be used to study the fine-scale influence of dead trees of various ages. We will use high resolution remotely sensed imagery from NAIP (1m, available statewide 2005, 2006 [2m], 2008, 2010, 2013, 2015, imagery before 2010 is true color) to measure the density (and potentially age) of dead trees within 2 foraging radii of each subsample point. We will segment imagery to identify likely trees (Fig 2a) and use machine learning algorithms such as support vector machines for classification. (Cortes and Vapnik 1995). Training data for live vs. dead segments will be developed from manual interpretation of NAIP images and validation will be conducted using field surveys (see below) and independent sources such as Google Earth (Achanta et al. 2012). Dead trees will be categorized into age classes on the basis of when they first appear in the imagery as dead trees. Again, we will use GLMs to examine the influence of dead trees on the probability of bat detection by species. We will accomplish this by modeling the influence of dead tree measurements on residuals for step one models (above) fit to the subsample data and by modeling probability of occupancy in the subsample data with predicted landscape detection rates as a covariate.Step three will involve using the high resolution remotely sensed imagery and the step two models to select an independent set of data points for model verification. We will then visit these sites and survey for bat presence with acoustic detectors using the DNR field protocols used to generate the modeling dataset (described below). DNR monitoring of WNS in known hibernacula and step 1 modeling will be used to guide study design so that field sampling will avoid areas where absence of bats is likely to be a result of disease. We will then fit derived (steps 1 and 2) and null models to the validation dataset and evaluate the validity of our derived models on the basis of comparisons to null models using Akaike's Information Criterion (Burnham and Anderson 2003). We will then refine models as needed and produce a "heat map" of Wisconsin forests showing how relative potential as summer roosting habitat varies with orientation to hibernacula and large- and small-scale ecological drivers.Data generated for the validation data set will incorporate information on the influence of WNS-affected hibernacula. We view this as a natural experiment for validating landscape level effects and we will evaluate models using additional covariates to account for year-specific detections of WNS in discrete hibernacula.Field sampling for bat occupancy and model validation: Once a sample site is identified, we will follow USFWS's range-wide summer survey guidelines for Indiana bat and NLEB (USFWS 2016). We will start occupancy assessments by using ultrasound detectors designed to record bat echolocation calls. Recorded calls of high enough quality can be identified to species. For non-linear projects, USFWS requires four detector nights per 123 acres. If target species NLEB or EAPI are recorded, we will continue assessments by netting for 42 net nights. If target species are captured, Wisconsin Department of Natural Resources will attach radio transmitters and track bats for the duration of the life of the transmitter. We will collect habitat characteristics from roost trees to which we track bats, further adding to the model.