Effects of Hurricanes Katrina and Rita on Louisiana Black Bear Habitat

Effects of hurricanes Katrina and Rita on Louisiana black bear habitat

Jennifer L. Murrow1 and Joseph D. Clark1,2,3

1Department of Forestry, Wildlife and Fisheries, University of Tennessee, 274 Ellington Plant Sciences, Knoxville, TN 37996, USA 2US Geological Survey, Southern Appalachian Research Branch, University of Tennessee, 274 Ellington Plant Sciences, Knoxville, TN 37996, USA

Abstract: The Louisiana black bear (Ursus americanus luteolus) is comprised of 3 subpopulations, each being small, geographically isolated, and vulnerable to extinction. Hurricanes Katrina and Rita struck the Louisiana and Mississippi coasts in 2005, potentially altering habitat occupied by this federally threatened subspecies. We used data collected on radio- telemetered bears from 1993 to 1995 and pre-hurricane landscape data to develop a habitat model based on the Mahalanobis distance (D2) statistic. We then applied that model to post- hurricane landscape data where the telemetry data were collected (i.e., occupied study area) and where bear range expansion might occur (i.e., unoccupied study area) to quantify habitat loss or gain. The D2 model indicated that quality bear habitat was associated with areas of high mast- producing forest density, low water body density, and moderate forest patchiness. Cross- validation and testing on an independent data set in central Louisiana indicated that prediction and transferability of the model were good. Suitable bear habitat decreased from 348 to 345 km2 (0.9%) within the occupied study area and decreased from 34,383 to 33,891 km2 (1.4%) in the unoccupied study area following the hurricanes. Our analysis indicated that bear habitat was not significantly degraded by the hurricanes, although changes that could have occurred on a microhabitat level would be more difficult to detect at the resolution we used. We suggest that managers continue to monitor the possible long-term effects of these hurricanes (e.g., vegetation changes from flooding, introduction of toxic chemicals, or water quality changes).

Key words: forest, habitat modeling, Hurricane Katrina, Hurricane Rita, Louisiana black bear, Mahalanobis distance, Ursus americanus luteolus, wetland

Ursus 23(2):192–205 (2012)

Impacts of hurricanes have been evaluated for a dramatic population reduction (Burkey 1989, Lande variety of animal taxa including fish (Dolloff et al. 1993, Simberloff 1995). 1994), herpetofauna (Reagan 1991, Schriever et al. The Louisiana coast is home to 1 of 3 small, 2009), birds (Wiley and Wunderle 1993, Torres and isolated subpopulations of Louisiana black bears Leberg 1996, White et al. 2005, Brown et al. 2011), (Ursus americanus luteolus), the other 2 subpopulations and mammals, mostly ungulates (Craig et al. 1994, being located along the Tensas River in northeastern Saı¨d and Servanty 1995, Swilling et al. 1998, Labisky Louisiana and along the upper Atchafalaya River et al. 1999, Lopez et al. 2003, Widmer et al. 2004, in central Louisiana. The coastal subpopulation was Storms et al. 2006). Effects have varied from bene- estimated at 77 individual bears (Triant et al. 2004) and ficial to detrimental depending on storm severity, is restricted to a narrow bandofforesthabitatbetween species habitat requirements, and species status. coastal marsh to the south and urban and agricultural Small, isolated populations that are highly vulnera- development to the north (Nyland 1995, Wagner 1995). ble to stochastic environmental events warrant Because of the small size and vulnerability of these 3 particular attention because they often lack the subpopulations, Louisiana black bears were listed as capability to move to more suitable habitat and do threatened under the US Endangered Species Act (US not have the demographic resilience to withstand a Fish and Wildlife Service 1992), requiring federal agencies to conserve habitat and protect this species from further decline (US Fish and Wildlife Service [email protected] 1995, 1999).

192 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 193

Hurricane Katrina struck the Louisiana and The area was similar to much of coastal Louisiana in Mississippi coasts on 29 August 2005, followed by that it was undergoing active subsidence. Hurricane Rita on 24 September. The 2 storms We also characterized a broader landscape that caused the loss of approximately 562 km2 of wetland included unoccupied bear habitat (hereafter, unoc- (Barras 2006). Hurricanes can affect forest ecology cupied) from eastern Texas to western Florida, USA (Stoneburner 1978, Breininger et al. 1999, Widmer et (28u539–32u189N, 87u279–95u389W). This area was al. 2004), and Chambers et al. (2007) reported a primarily made up of the Outer Coastal Plain Mixed dramatic increase in non-photosynthetic land cover Forest Province but also encompassed portions of classifications (e.g., felled trees) in southern Mis- the Lower Mississippi Riverine Forest Ecosystem sissippi following the 2 hurricanes, suggesting that Province (Bailey 1988). The Lower Mississippi considerable tree fall may have occurred in some Riverine Forest Province is characterized by broad forest types. The coastal bear population is small, floodplains and low terraces made up of alluvium coastal forest habitat is extremely limited in quantity and loess (Bailey 1988). Most of the area was flat, and highly fragmented, and no work has been with an average southward slope of ,127 mm/km. performed to evaluate effects of hurricanes on large The only significant changes in elevation were sharp carnivores. Thus, our objectives were to estimate terrace scarps and natural levees that rose a few how hurricanes Katrina and Rita affected black bear meters above adjacent bottomlands or stream habitat quantity and quality. To perform that channels. Historically, this province was covered by evaluation, our approach was to use pre-hurricane bottomland deciduous forest with an abundance of landscape and bear radiotelemetry data to develop a ash (Fraxinus spp.), elm (Ulmus spp.), cottonwood habitat model, estimate changes in land cover type (Populus deltoides), sugarberry, sweetgum (Liquid- based on data collected after the hurricanes, and ambar styraciflua), water tupelo (Nyssa aquatica), apply the habitat model to the post-hurricane oak (Quercus spp.), and baldcypress (Taxodium landscape to quantify and characterize bear habitat distichum). Much of the province has since been loss or gain to evaluate the potential impacts to this converted to agriculture (Stavins 1986, Neal 1990). threatened subpopulation of bears.

Study areas Methods We evaluated 2 landscape extents in our analysis. Telemetry data First, we modeled habitat quality within occupied Coastal Louisiana black bears were radiomoni- bear range (hereafter, occupied) in an area consist- tored weekly from aircraft during 1991–1995 ing of the 36.5-km2 Bayou Teche National Wildlife (Wagner 1995, 2003; Pace et al. 2000). Locations of Refuge (NWR) and surrounding private land near those bears were estimated during daylight hours Franklin, Garden City, and Centerville, Louisiana, and plotted directly on 1:24,000 USGS quadrangle USA (29u279–29u589N, 91u149–91u559W). This area maps. Median location error was estimated to be was within the Outer Coastal Plain Mixed Forest 281 m (147 m inner quartile, 417 m outer quartile, Province within the Mississippi Alluvial Valley and n 5 106; Wagner 2003). was gently sloping with elevations ,90 m (Bailey We retained individual bears in the data set only if 1988). Average annual temperatures ranged from 16 .30 radiolocations had been collected within a 2- to 21uC. Rainfall was abundant and well distributed year period comprised of .10 of the 12 calendar throughout the year; precipitation ranged from 102 months. We estimated straight-line distances be- to 153 cm annually. Bayou Teche NWR was tween consecutive weekly locations (5–13 days apart) surrounded by private lands that were a mix of to calculate average weekly movement. We used the forested, agricultural, and industrial lands used for Animal Movement Extension (Hooge and Eichen- hunting, sugarcane farming, and industry such as oil, laub 1997) in ArcViewH GIS (Environmental Sys- gas, salt, and carbon black production. Marshes, tems Research Institute, Redlands, California, USA swamps, and lakes were numerous. Live oak [use of trade, product, or firm names does not imply (Quercus virginiana), water oak (Quercus nigra), endorsement by the United States government]) to sugarberry (Celtis laevigata), and American elm calculate 75% and 95% kernel home ranges (Worton (Ulmus americana) were common overstory species. 1989) for the animals meeting our inclusion criteria.

Ursus 23(2):192–205 (2012) 194 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark

Habitat variables until every cell was assigned an average forest We created 13 habitat variables for potential density. Water body density (WATER) was deter- inclusion in the model from land cover data, water mined with a combination of an NHD line layer, body data, and roads data with ArcInfoH GIS which represented rivers and streams, and an NHD (Environmental Systems Research Institute, Red- water bodies layer (water 5 1, non-water 5 0). We lands, California, USA) and Fragstats software calculated road density (ROADS) based on the (McGarigal et al. 2002). All 13 variables were 30-m TigerH primary roads data whereby pixels containing resolution continuous-valued grid metrics. We ob- roads were assigned values of 1 and the remaining tained pre- and post-hurricane land cover data based pixels had values of 0. We used the same neighbor- on 30-m resolution Landsat Thematic Mapper (TM) hood analysis technique to calculate WATER and and Landsat Enhanced Thematic Mapper satellite ROADS at the 30-m2 resolution. imagery (TM 5 or 7, National Oceanic and We used Fragstats software (McGarigal and Atmospheric Administration 2009). The pre-hurri- Marks 1995, Haines-Young and Chopping 1996, cane Landsat data were collected from 5 November Turner et al. 2001) to calculate 9 additional variables 2004 to 15 August 2005, and the post-hurricane data that quantified broad-scale patterns of forest– were collected from 22 November 2005 to 2 March wetland edge and patch configuration that were 2006. At-satellite reflectance was performed on each potentially important for bear movement and scene, and the tasseled cap transformation was metapopulation dynamics. All 9 variables were applied (Crist and Cicone 1984). Each Landsat TM calculated using the forest–non-forest land cover scene was geo-referenced by the US Geological data. Contagion (CONTAG) was a measure of both Survey Earth Resources Observation and Science dispersion and interspersion of patches and was Data Center and tested for spatial accuracy to within inversely related to edge density (McGarigal and 2 pixels (60 m). Water body data was obtained from Marks 1995). Connectivity (CONNECT) was a the most current National Hydrography Dataset measure of the connectedness of the landscape and (NHD; US Geological Survey 2009) and road had a threshold based on 25% of the estimated mean information from TIGERH line data from the distance of weekly female bear movements. That redistricting 2000 census (US Bureau of the Census threshold was arbitrary but we considered it 2009). reasonable to depict what was easily accessible by We combined land cover types to form a binary female bears within normal movement patterns. If 2 classification consisting of forest (1) and non-forest forest patches were within that threshold, those 2 (0) categories, excluding areas that were permanent- patches were considered connected. Splitting index ly inundated. We defined the forest category as (SPLIT) was a measure of the patchiness of the deciduous forest, mixed forest, or palustrine forest landscape; SPLIT increased as the landscape was Landsat cover types, which have been shown to be increasingly subdivided into smaller patches. Cohe- important to Louisiana black bears (Nyland 1995, sion (COHESION) and percent of like adjacent Benson and Chamberlain 2007). The non-forest (PLADJ) were connectivity measures and required category represented all other land cover types no threshold. Fractal dimensions (FRACTAL) and including urban and barren areas. Using that data landscape shape index (LSI) were measures of layer, we calculated forest density (FOREST) based landscape shape, including land cover aggregation on a neighborhood analysis within a circular moving or clumpiness. Patch density (PD) and patch richness window (any cell center encompassed by the circle (PR) were measures of landscape-level forest patch was included in the processing of the neighborhood), characteristics, and Simpson’s diversity index (SIDI) the diameter of which was equivalent to the average was an ecological measure of patch richness. We weekly movements of radiocollared bears (1,500 m; used the same sized neighborhood analysis to see Results). We used the neighborhood analysis to perform these calculations for these 9 variables at characterize the habitat conditions at a particular the 30-m2 resolution. pixel within the context of the surrounding habitats. We then assigned the average values of all cells in the Statistical model window (i.e., forest density) to the original center 30- Mahalanobis distance (D2) has been used to assess m2 grid cell (i.e., number of cells classified as forest/ habitat for a wide range of plant and animal species total number of cells). We continued this analysis (e.g., Clark et al. 1993, Knick and Dyer 1997, Corsi

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 195 et al. 1999, Farber and Kadmon 2003, Buehler et al. The use of individual radiolocations can bias habi- 2006, Thompson et al. 2006, Griffin et al. 2010). We tat analyses due to disparate numbers of telemetry chose D2 for habitat modeling because it requires locations among animals, telemetry error resulting in only presence data and the method performed well misclassification of individual locations, and tempo- among several presence-only modeling techniques ral biases in the telemetry data, so our use of home evaluated (Alldredge et al. 1998, Tsoar et al. 2007). ranges helped to reduce those effects (Kauhala and In addition, D2 does not require that assumptions of Tiilikainen 2002, Moser and Garton 2007). We only multivariate normality be met, variables can be in used home ranges of female bears for model different units of measurement, and co-varying development and testing because male home ranges habitat variables do not bias the statistic value were large and were likely to include extensive (Rao 1952, Clark et al. 1993, Knick and Rotenberry amounts of area not actually used by the bears, 1998). which could result in an overly optimistic character- Mahalanobis distance is a multivariate measure of ization of model performance. dissimilarity, with small values of D2 (distances) Before developing the model, we reduced the representing landscape conditions similar to those number of variables considered. We did this for associated with the location data, whereas larger simplicity, ease of interpretation, and because some values represent increasingly different conditions variables may have poor or redundant explanatory (Rao 1952). This technique predicts habitat suitabil- power. We selected variables for elimination by ity based on species presence data (e.g., radio- identifying means, variances, or ranges that differed locations) and geographic variables (e.g., land cover little among bear home ranges and the overall study data in a GIS grid) with the following equation: area (i.e., poor explanatory power). We retained variables if we believed they represented unique ’X 2~ {_ {1 {_ landscape characteristics (e.g., ROADS, and D Àx Àu Àx Àu , SPLIT). where x is a vector of landscape characteristics, uˆ is the mean vector of landscape characteristics esti- Model evaluation 2 mated from the animal location data, and g21 is the After constructing the final D model with SAS/ inverse of the variance–covariance matrix of the STATH software (SAS Institute, Inc. 2004), we used landscape variables (Rao 1952). For example, a principal components analysis to assess the consider a model consisting of the 2 variables of relationship between and among the final variables. forest density and road density. If the mean forest We did so by calculating the eigenvalues for each density and road densities within all home ranges eigenvector based on the correlation matrix and were 0.75 km2/km2 and 440 m/km2, respectively (uˆ), identifying the significant components of the model and the values for a particular pixel were 0.70 and (Jackson 1991, Morrison et al. 1992). To determine 420, respectively (x), then the difference (x 2 uˆ) the number of components to interpret from the would be 0.05 and 20, respectively. Those numbers principal components analysis, we used the broken- would constitute a row vector for that pixel, which stick method as a stopping rule (Jackson 1993). The would then be multiplied by the inverse of the broken-stick distribution is essentially the expected variance–covariance matrix of those 2 variables. distribution of eigenvalues based on random vari- Finally, the resulting row vector would be multiplied ables. We evaluated any principal components that by the corresponding column vector (x 2 uˆ)9, thus explained .10% of the total variance and retained producing a single squared value (D2) for that GIS those components if observed eigenvalues were pixel. The same calculation would be made for every higher than the corresponding random broken-stick pixel on the landscape map. The resulting D2 statistic components. for each pixel provides a unitless index of similarity Although defining study area boundaries is not to the multivariate landscape conditions associated necessary to calculate D2, we did so for purposes of with the location data (Knick and Rotenberry 1998). model testing. We defined our occupied study area We used 75% kernel home ranges as our sampling as the area that was physically available (land cover units rather than individual radiolocations and types excluding open water) for home range estab- evaluated their placement on the landscape as our lishment by radiocollared bears. To quantify this habitat selection criterion (Type II; Johnson 1980). area, we merged radiolocations that were included in

Ursus 23(2):192–205 (2012) 196 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark the kernel home range estimates and calculated a removed per iteration (i.e., we randomly removed 3 minimum convex polygon (MCP) around those home ranges and recalculated the model; we did this locations using ArcViewH. We then applied a buffer, 10 times). Because of our relatively small sample which corresponded to the radius of the average sizes and the arbitrary nature of selecting a num- female home range, to the outside periphery of the ber of classes or bins for ratio comparisons, we MCP to delineate the study area boundaries. We calculated the Boyce ratio for 3, 4, and 5 equally created random home ranges with the same size distributed bins of the scaled D2 values. The bin distribution as the bear home ranges until .50% of values were independently established by equally the area within our study area boundaries, as defined dividing mean D2 values from the 250 randomly above, was sampled (Katnik and Wielgus 2005). We located home ranges into 3, 4, and 5 equal parts, and then plotted cumulative frequency distributions of identifying the corresponding D2 value. An increas- D2 values for the original bear home ranges and the ing relationship across binned categories and Spear- random home ranges, using the Kolmogorov-Smir- man-rank correlations (rs . 0.5) between bin ranks nov test to detect differences. If differences were and area-adjusted frequencies for model sets indi- detected, we established a cutoff value for suitable cates a strong relationship and supports the validity habitat by determining the D2 value where the of the model. difference between the 2 cumulative frequency Additionally, we tested the original D2 model with distributions was greatest (Browning et al. 2005). independent bear presence data collected at barbed- We considered pixels with Mahalanobis distance wire hair-sampling sites for a mark–recapture study values below the cutoff to represent suitable bear in Pointe Coupee Parish in the Upper Atchafalaya habitat and used that cutoff to classify each random River basin (Lowe 2011). Pixels containing hair- and bear home range. Home ranges with mean D2 sampling sites where bears were detected were paired values below the cutoff were classified as being in with the D2 values for those pixels to determine the suitable habitat and those above the cutoff were not. proportion above and below our cutoff. We used a To assess model consistency, identify outliers in Wilcoxon rank-sum test to compare the D2 values of the data used to construct the model, and evaluate a random sample of 104 locations, generated within predictive performance, we used a cross-validation an area the size of the average female home range technique (Larson 1931, van Manen et al. 2002) and centered on the hair-sampling sites, to the D2 whereby we excluded 1 bear home range from the values of the hair stations. data set, recalculated D2, and assessed whether the mean D2 value within the excluded home range was Model application above or below the cutoff for that particular model. We used the finalized D2 model to estimate the We repeated that procedure until each home range overall loss of suitable bear habitat. To do that, we had been excluded and tested, calculating the overall recalculated all model variables with the post- proportion of correctly classified home ranges for hurricane landscape data and recalculated the D2 each (Verbyla and Litvaitis 1989). Although our statistic using the pre-hurricane model coefficients. sample size for developing the habitat model was To identify shifts in habitat quality, we examined the small, Katnik and Wielgus (2005) found that the change in mean D2 scores across each bear home Type I error rates for comparing random to range and each randomly placed home range. We observed home ranges with the methods we em- also estimated the area above and below our ployed was low (,0.04 for tests using 10 home suitability cutoffs, both before and after the hurri- ranges 10 km2 in size). canes and for both the occupied and unoccupied We also performed a model validation based on study areas. the Boyce index (Boyce et al. 2002, Hirzel et al. 2006). To do so, we first scaled the D2 scores into 10 value categories based on equal percentiles. For Results example, D2 values for the lowest one-tenth percen- The telemetry data set consisted of 601 locations tile of the study area were categorized as 10.0, the representing 13 (2M:11F) individuals after we next tenth percentile as 9.0, and so on. We used those applied our criteria for data inclusion in the analysis. scaled D2 scores to calculate the Boyce index based The average 95% kernel home-range was 11.8 km2 on a 10-fold cross-validation with 3 home ranges (n 5 11, SE 5 2.6) for females and 51.1 km2 (n 5 2,

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 197

Table 1. Mean (xÀ), standard error (SE), and range of Table 2. Eigenvalues and observed proportion of Mahalanobis distance values (D2) associated with variance explained by 6 principal components bear home ranges and study area for landscape- associated with the variables used to model bear scale variables included in the model of occupied home ranges in coastal Louisiana, 1993–95. The bear habitat in coastal Louisiana, 1993–95. Variables expected proportion of variance is based on the are forest density (FOREST), road density (ROADS), broken-stick distribution. water body density (WATER), contagion (CONTAG), splitting index (SPLIT), and landscape shape Principal Observed Expected index (LSI). component proportion of proportion of vector Eigenvalue variance variance Bear home ranges (n = 10) Study area x 1 4.46 0.7424 0.4083 Variable x (SE) Range (SE) 2 1.02 0.1705 0.2417 3 0.40 0.0658 0.1583 FOREST 0.73 (0.14) 0.55–0.95 0.39 (0.34) 4 0.11 0.0180 0.1028 ROADS 0.03 (0.04) 0.00–0.11 0.03 (0.06) 5 0.01 0.0018 0.0611 WATER 0.06 (0.01) 0.04–0.08 0.12 (0.13) 6 0.01 0.0012 0.0278 CONTAG 41.67 (25.76) 9.04–86.61 57.08 (27.67) SPLIT 2.01 (0.55) 1.13–2.64 1.66 (0.73) LSI 4.15 (2.20) 1.43–7.21 2.87 (1.38) Of the 13 variables considered, 6 were used in the final model: CONTAG, LSI, SPLIT, FOREST, WATER, and ROADS (Table 1). Areas selected by SE 5 36.2) for males. The 75% home ranges for the bears were generally associated with areas of 11 females averaged 5.07 km2 (SE 5 1.66), ranging variable landscape complexity, limited permanent from 1.16 to 19.76 km2. Average weekly movements standing water, and extensive amounts of palustrine of bears with consecutive weekly locations from April or deciduous forest. The principal components to November was 1,454 m (n 5 12, SE 5 146.8) for all analysis indicated the first 2 components explained bears and 1,353 m (n 5 10, SE 5 153.8) for females. 91.3% of the model variation (74.2 and 17.1%, The 75% kernel home ranges of the 11 females and the respectively), but the observed distribution of eigen- 250 randomly generated home ranges covered 47 km2 values was less than the distribution based on random (5%) and 670.5 km2 (68%) of the occupied area, variables for the second component (Table 2). respectively. Based on these data, we used 1,900 m The model variables with the strongest correlation (95% kernel female home range radius) as our buffer and greatest loadings (.0.4) in the first principal for study area delineation for model testing (983 km2), component were evenly distributed between FOR- used 375 m (25% of mean weekly female movement) EST, CONTAG, SPLIT, and LSI (Table 3), all of as the cutoff for the CONNECT habitat variable, and which were related to landscape shape or degree of used 1,500 m (average weekly movement of bears) as fragmentation of forest cover. the diameter of the circular moving window for the Calculation of the Boyce index based on 10-fold neighborhood analyses. cross-validation indicated a positive correlation for We identified 1 outlier home range during the the 3-, 4-, and 5-binned values (mean rs 5 1.0, 0.8, initial cross-validation calculations that dispropor- and 0.9, respectively), although the 4-binned values tionately impacted the model parameters when were not statistically significant (P , 0.001, P 5 included and resulted in .50% of the home ranges 0.200, and P 5 0.037, respectively). Graphically, all to be classified as unsuitable bear habitat. This animal binned values showed the expected relationship if had a home range that included (11%) an industrial habitat selection was occurring (Fig. 1). complex consisting of a salt mine and an electronics The mean D2 value for the occupied study area materials plant. Therefore, we considered that indi- was 1,460.3 (SD 5 3,518.2), whereas mean values for vidual to be an outlier, removed her home range from individual home ranges ranged from 20.4 to 143.0 (Àx the data set, and reparameterized the model. The 5 80.8, SE 5 45.2). The random home ranges had second cross-validation resulted in 7 of the 10 test mean D2 values ranging from 13.7 to 27,362.6 (Àx 5 home ranges being correctly classified. However, 1,417.1, SE 5 2,750.1) and differed from the original mean D2 values for the 3 outlier home ranges bear home ranges (D 5 0.80, P , 0.01), suggesting exceeded the model cutoff by an average of only that bear habitat selection differed from random. 2.3% (SD 5 0.02). Therefore, we did not exclude these The largest separation between the 2 cumulative home ranges from the model parameterization. frequency distributions occurred at a D2 value of

Ursus 23(2):192–205 (2012) 198 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark

Table 3. Principal component loading values of the Following the hurricanes, ,0.5% of the land cover first 2 principal components calculated for a classifications changed on the occupied study area 2 Mahalanobis distance (D ) model of occupied bear (Table 4). Of the cover types making up .1% of the habitat in coastal Louisiana, 1993–95. Variables are study area, open water increased by the greatest forest density (FOREST), road density (ROADS), water body density (WATER), contagion (CONTAG), proportion (1.7%). Forest classifications (deciduous, splitting index (SPLIT), and landscape shape mixed, and palustrine forest) decreased by 2.8%. index (LSI). Mean D2 values increased in 8 of the 10 bear home Principal Principal ranges following the hurricanes, but no home ranges Variables component 1 component 2 would have been reclassified as being non-habitat. 2 FOREST 20.45a 0.02 Average D values within the home ranges increased ROADS 0.39 20.16 from 80.8 to 86.2. Mean D2 values in the occupied a WATER 0.08 0.97 study area before the hurricanes changed from CONTAG 20.47a 20.04 SPLIT 0.47a 0.09 1,460.3 (SD 5 3,518.2) to 1,495.4 (SD 5 3,715.1; 2 LSI 0.45a 20.14 Fig. 3) afterwards. Suitable habitat (i.e., D , 143) 2 aVariables with principal components loadings .0.4. decreased from 348 km before the hurricanes to 345 km2 afterward (0.9%). In the unoccupied study 2 143, with the mean D2 value of all 10 bear home area, suitable habitat decreased from 34,383 km to 2 ranges falling below that cutoff (Fig. 2) compared 33,891 km (1.4%; Fig. 4). with 49 of the 250 random home ranges (19.6%). From the 104 hair-sampling stations in Pointe Coupee Parish where bear presence was documented, Discussion 81 (78%) were located in pixels with D2 values less Habitat quality can be affected in divergent ways than the 143 cutoff, 45 of which occurred where D2 following hurricanes depending on the species of values were ,40. The 104 random locations created interest. For example, Florida scrub jay (Aphelocoma for comparison had an average D2 value of 336.3, 61 coerulescens) coastal populations experienced a 10– (58.7%) of which fell below the cutoff. The Wilcoxon 30% increase in extinction rates (Breininger et al. rank-sum test confirmed that the random locations 1999), whereas Florida key deer (Odocoileus virgi- came from a different distribution than the active hair nianus clavium) experienced an increase in produc- sampling locations (U 5 8,507.5, P , 0.001). tivity because of the increase in available browse

Fig. 1. Ratio of observed and expected frequency of home ranges within 3 (black), 4 (dashed), and 5 (gray) classes (bins) of bear habitat index values based on 10-fold cross-validation of American black bear habitat index values, coastal Louisiana, 1993–95.

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 199

Fig. 2. Cumulative frequency distributions of black bear home ranges’ (solid line) and random home ranges’ (dashed line) average Mahalanobis distance statistic values. The arrow line marks the largest separation of the 2 distributions (143). following hurricanes (Lopez et al. 2003). The hurricanes. Additionally, the effects of the hurricanes ecological literature suggests that habitat generalists on forested habitats as detected by the TM data were may fare better when environmental perturbations minimal. Consequently, we speculate that probabili- occur (Devictor et al. 2008). Black bears are con- ties of population persistence were also not dramat- sidered habitat generalists (Pelton 2003), so it is ically altered for the Louisiana black bear. perhaps not surprising that our analysis indicated that Our analysis was performed at a coarse resolution, bear habitat was only minimally affected by the 2005 a necessity based on the scaling of the landscape TM

Table 4. Land-cover type percentages and area of the 2005 pre- and post-hurricane data used for the model of occupied bear habitat calculated in coastal Louisiana, 1993–95. Percentage of Pre-hurricane Post-hurricane Difference Landsat cover type (value) cover type data (km2) data (km2) (% change) High developed (2) ,1% 1.24 1.25 0.01 (0.8) Medium developed (3) ,1% 2.27 2.20 20.07 (23.1) Low developed (4) 3% 26.53 26.59 0.06 (0.2) Developed open space (5) ,1% 0.91 0.91 0 (0.0) Cultivated/row crops (6) 15% 143.03 143.03 0 (0.0) Pasture/hay (7) 1% 7.54 7.54 0 (0.0) Grasslands (8) ,1% 3.42 3.52 0.10 (2.9) Deciduous forest (9) 1% 9.32 9.07 20.25 (22.7) Evergreen forest (10) ,1% 0.30 0.27 20.03 (210.0) Mixed forest (11) ,1% 0.15 0.15 0 (0.0) Scrub–shrub (12) ,1% 3.28 3.42 0.14 (4.3) Palustrine forested (13) 38% 374.93 374.64 20.29 (20.1) Palustrine scrub–shrub (14) 2% 21.60 20.84 20.76 (23.5) Palustrine emergent (15) 25% 244.16 245.21 1.05 (0.4) Estuarine scrub–shrub (17) ,1% 0.82 0.77 20.05 (26.1) Estuarine emergent (18) 7% 71.63 71.46 20.17 (20.2) Unconsolidated shore (19) 1% 11.00 10.25 20.75 (26.8) Barren (20) ,1% 0.52 0.52 0 (0.0) Water (21) 6% 58.62 59.63 1.01 (1.7) Palustrine aquatic bed (22) 0% 1.37 1.36 20.01 (20.7)

Ursus 23(2):192–205 (2012) 200 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark

Fig. 3. Mahalanobis distance values (D2) after hurricanes Katrina and Rita on unoccupied bear habitat along the Gulf Coast in Texas, Louisiana, Mississippi, Alabama, and Florida, 2005. The occupied study area adjacent to Bayou Teche National Wildlife Refuge is inset. data available to us. Ramsey et al. (2009) found that also reflected greater loss in this specific area Landsat TM was similar to radar satellite data in its compared with the broader region, but to a lesser ability to detect changes in forest structure following extent. Brown et al. (2011) reported reduced canopy Hurricane Katrina in the Pearl River Basin. Landsat closure on their Pearl River study area following the TM data have also been successfully used to identify hurricanes, with increased understory density, most- forests heavily damaged by Hurricane Hugo (Cablk ly consisting of blackberries (Rubus spp.). This et al. 1994), and TM images have provided good suggests that forest cover with moderate wind class separation following hurricanes when classes damage likely retains habitat value for bears because dominated areas .30 m (Ramsey and Laine 1997). production of berries and other soft mast foods However, Wang and Xu (2010) found that some would probably increase due to the creation of more components of the TM Tasseled Cap transformation light gaps in the forest canopy. Thus, the loss of may not as readily detect changes in forest areas with small patches of trees within a larger forest matrix lesser degrees of damage and with minimal leaf would not necessarily degrade bear habitat. Regard- wetness impacts. Thus, it was possible for small (,30 less, further evaluation to resolve the discrepancies x 30 m) or slightly to moderately damaged forest between the remotely sensed data and field observa- cover types to retain their pre-hurricane classifications of forest condition in areas affected by the tions. The loss of forest cover from these 2 hurricanes is needed. hurricanes was perhaps greatest in the Pearl River Some variables in the model were simple, espe- Basin of eastern Louisiana where Chapman et al. cially the binary land-cover classification that we (2008) reported post-Katrina annual tree mortality used. We combined forest cover types to improve rates of 20.5%. Our pre- and post-land cover data performance of the model across a broader land-

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 201

Fig. 4. Black bear habitat lost (D2 values falling below 143 in gray) after hurricanes Katrina and Rita along the Gulf Coast of Louisiana, 2005. The occupied study area is depicted with a black border. scape where we lacked home range data or where The principal component analysis indicated that bears did not occur. For example, had we defined forest density and fragmentation were underlying estuarine shrub–scrub as a separate cover type in the drivers of black bear habitat selection. A post-hoc model, the statistical procedure may well have stepwise logistic regression of the 6 variables resulted identified proximity to that cover type as a habitat in only FOREST and LSI variables being retained requirement because the target data set was located and suggested that bears preferred to locate their near the coast. Consequently, the model would have home ranges in areas where forests were more predicted that habitat was poor further inland, prevalent (FOREST b 5 7.355, 95% CI 5 21.591– which we knew to be untrue. Therefore, we chose 13.119) and aggregated (LSI b 5 1.005, 95% CI 5 more generalized variables to avoid this problem of 0.411–1.599). When we inspected home range model overspecificity. Regardless, cross-validation placement on the landscape, we noticed that all and independent model testing in the Upper bears selected areas of generally high forest density, Atchafalaya River Basin indicated that model but individuals varied in the amount of fragmenta- performance was good. Bears select habitat based tion and edge they tolerated. Furthermore, bears on landscape characteristics (Schoen 1990), so it is whose home ranges centered on deciduous forests not surprising that our model performed well even tended to tolerate not only higher fragmentation but though the variables were relatively coarse. Al- also the lowest forest densities. These observations though it was necessary to use a cutoff value to are again consistent with bears being habitat discriminate between habitat and non-habitat, we generalists and suggest that the effect of forest caution that the D2 values in our model represent a fragmentation on bear ecology in coastal Louisiana continuum and an absolute threshold value is an may depend in part on the quality of the habitat oversimplification. being fragmented.

Ursus 23(2):192–205 (2012) 202 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark

We only evaluated immediate and direct effects of appreciation to R. Pace, M. Chamberlain, J. Ertel, the hurricanes. However, significant longer-term and the other researchers and managers that came impacts to wildlife as a result of hurricanes have before us who collected the bear data, performed been documented following minimal immediate earlier analyses, and permitted their use in our study. effects (Wauer and Wunderle 1992, Wiley and Wunderle 1993, Swilling et al. 1998). Future indirect effects could include vegetation changes due to Literature cited saltwater flooding following storm surges (Conner ALLDREDGE, J., D.L. THOMAS, AND L.L. MCDONALD. 1998. et al. 2005), the introduction of toxic chemicals Survey and comparison of methods for study of (Suedel et al. 2006), water quality changes due to resource selection. Journal of Agricultural, Biological, floodwater pumping (Ray 2006), and land subsi- and Environmental Statistics 3:237–253. dence (Burkett et al. 2003). Also, fallen trees are BAILEY, R.G. 1988. Ecogeographic analysis: A guide to the ecological division of land for resource management. susceptible to insect infestations, and increased fuel Miscellaneous Publication 1465, USDA Forest Service, loads increase the chances of wildfire (Sheikh 2005). Washington, DC, USA. These types of long-term impacts would be especially BARRAS, J.A. 2006. Land area change in coastal Louisiana important for small, vulnerable populations. Thus, after the 2005 hurricanes — A series of three maps. US although existing damage from hurricanes Katrina Geological Survey Open-File Report 06-1274, Wash- and Rita has not resulted in a significant reduction in ington, DC, USA. suitable bear habitat, there may be future impacts. BENSON, J.F., AND M.J. CHAMBERLAIN. 2007. Space use and We suggest that managers continue to monitor habitat selection by female Louisiana black bears in possible longer-term changes in bear habitat due to the Tensas River Basin of Louisiana. Journal of hurricanes. Wildlife Management 71:117–126. Despite our findings of little overall immediate BOYCE, M.S., P.R. VERNIER, S.E. NIELSEN, AND F.K.A. SCHMIEGELOW. 2002. Evaluating resource selection fun- impact, the hurricanes served to illustrate the ctions. Ecological Modelling 157:281–300. precariousness of small, isolated populations. The BREININGER, D.R., M.A. BURGMAN, AND B.M. STITH. 1999. outcome could have been far different for the coastal Influence of habitat quality, catastrophes, and popula- bear population had the trajectories or intensities of tion size on extinction risk of the Florida scrub-jay. the hurricanes been otherwise. Specifically, if Hur- Wildlife Society Bulletin 27:810–822. ricane Katrina had made a more westerly landfall or BROWN, D., T. SHERRY, AND J. HARRIS. 2011. Hurricane Hurricane Rita had made a more easterly landfall, Katrina impacts the breeding bird community in a the damages could have been much more significant bottomland hardwood forest of the Pearl River basin, to bears. The Louisiana black bear recovery plan Louisiana. Forest Ecology and Management 261:111– (US Fish and Wildlife Service 1995) specifies that at 119. least 2 geographically distinct subpopulations should BROWNING, D.M., S.J. BEAUPRE, AND L. DUNCAN. 2005. Using partitioned Mahalanobis D2 to formulate a GIS- be viable for delisting to occur. That multi-subpop- based model of timber rattlesnake hibernacula. Journal ulation approach to recovery is warranted to help of Wildlife Management 69:33–44. spread the risk should weather or other environ- BUEHLER, D.A., M.J. WELTON, AND T.J. BEACHY. 2006. mental perturbation negatively affect any single Predicting cerulean warbler habitat use in the Cumber- population segment. land Mountains of Tennessee. Journal of Wildlife Management 70:1763–1769. BURKETT, V.R., D.B. ZILKOSKI, AND D.A. HART. 2003. Sea- Acknowledgments level rise and subsidence: Implications for flooding in We would like to acknowledge and thank the US New Orleans, Louisiana. Pages 63–70 in K.R. Prince Fish and Wildlife Service, especially D. Fuller, who and D.L. Galloway, editors. US Geological Survey Subsidence Interest Group Conference. US Geological funded and supported this project. The Louisiana Survey Open File Report, 03-308, Galveston, Texas, Department of Wildlife and Fisheries helped coor- USA. dinate the contracting, for which we are grateful. We BURKEY, T.V. 1989. Extinction in nature reserves: The are deeply indebted to D. Walther, M. Davidson, C. effect of fragmentation and the importance of migra- Lowe, and the Black Bear Conservation Committee tion between reserve fragments. Oikos 55:75–81. for helping us better understand bear habitat issues CABLK, M.E., B. KJERFVE, W.K. MICHENER, AND J.R. in southern Louisiana. Finally, we express our JENSEN. 1994. Impacts of hurricane Hugo on a coastal

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 203

forest: Assessment using Landsat TM data. Geocarto Center — Biological Science Office, US Geological International 9:15–24. Survey, Anchorage, Alaska, USA. CHAMBERS,J.Q.,J.I.FISHER,H.ZENG,E.L.CHAPMAN,D.B. JACKSON, D.A. 1993. Stopping rules in principal compo- BAKER, AND G.C. HURTT. 2007. Hurricane Katrina’s carbon nents analysis: A comparison of heuristical and footprint on US Gulf Coast forests. Science 318:1107. statistical approaches. Ecology 74:2204–2214. CHAPMAN, E.L., J.Q. CHAMBERS, K.F. RIBBECK, D.B. JACKSON, J.E. 1991. A user’s guide to principal compo- BAKER, M.A. TOBLER,H.ZENG, AND D.A. WHITE. nents. John Wiley and Sons, New York, New York, 2008. Hurricane Katrina impacts on forest trees of USA. Louisiana’s Pearl River basin. Forest Ecology and JOHNSON, D.H. 1980. The comparison of usage and Management 256:883–889. availability measurements for evaluating resource CLARK, J.D., J.E. DUNN, AND K.G. SMITH. 1993. A habitat preference. Ecology 61:65–71. use model for black bears in Arkansas using a KATNIK, D.D., AND R.D. WIELGUS. 2005. Landscape geographic information system. Journal of Wildlife proportions versus Monte Carlo simulated home ranges Management 57:519–526. for estimating habitat availability. Journal of Wildlife CONNER, W.H., W.D. MIXON II, AND G.W. WOOD. 2005. Management 69:20–32. Maritime forest habitat dynamics on Bulls Island, KAUHALA, K., AND T. TIILIKAINEN. 2002. Radio location Cape Romain National Wildlife Refuge, SC, following error and the estimates of home-range size, movements, Hurricane Hugo. Forest Ecology and Management and habitat use: A simple field test. Annales Zoologici 212:127–134. Fennici 39:317–324. CORSI, F., E. DUPRE´ , AND L. BOITANI. 1999. A large-scale KNICK, S.T., AND D.L. DYER. 1997. Distribution of black- model of wolf distribution in Italy for conservation tailed jackrabbit habitat determined by GIS in south- planning. Conservation Biology 13:150–159. western Idaho. Journal of Wildlife Management 61: CRAIG, P., P. TRAIL, AND T.E. MORRELL. 1994. The decline 75–85. of fruit bats in American Samoa due to hurricanes and ———, AND J.T. ROTENBERRY. 1998. Limitations to overhunting. Biological Conservation 69:261–266. mapping habitat use areas in changing landscapes using CRIST, E.P., AND R.C. CICONE. 1984. A physically-based the Mahalanobis distance statistic. Journal of Agricul- transformation of thematic mapper data — The TM tural, Biological, and Environmental Statistics 3:311– tasseled cap. IEEE Transactions on Geoscience and 322. Remote Sensing 22:256–263. LABISKY, R.F., K.E. MILLER, AND C.S. HARTLESS. 1999. DEVICTOR, V., R. JULLIARD, AND F. JIGUET. 2008. Distri- Effects of Hurricane Andrew on survival and move- bution of specialist and generalist species along spatial ments of white-tailed deer in the Everglades. Journal of gradients of habitat disturbance and fragmentation. Wildlife Management 63:872–879. Oikos 117:507–514. LANDE, R. 1993. Risk of population extinction from DOLLOFF, C.A., P.A. FLEBBE, AND M.D. OWEN. 1994. Fish demographic and environmental stochasticity and ran- habitat and fish populations in a southern Appalachian dom catastrophes. American Naturalist 142:911–927. watershed before and after Hurricane Hugo. Transactions LARSON, S. 1931. The shrinkage of the coefficient of of the North American Fisheries Society 123:668–678. multiple correlation. Journal of Educational Psycholo- FARBER, O., AND R. KADMON. 2003. Assessment of gy 22:45–55. alternative approaches for bioclimatic modeling with LOPEZ, R.R., N.T. SILVY, R.F. LABISKY, AND P.A. FRANK. special emphasis on the Mahalanobis distance. Ecolog- 2003. Hurricane impacts on key deer in the Florida ical Modelling 160:115–130. Keys. Journal of Wildlife Management 67:280–288. GRIFFIN, S.C., M.L. TAPER,R.HOFFMAN, AND L.S. MILLS. LOWE, C.L. 2011. Estimating population parameters of the 2010. Ranking Mahalanobis distance models for Louisiana black bear in the upper Atchafalaya Basin. predictions of occupancy from presence-only data. Thesis, University of Tennessee, Knoxville, Tennessee, Journal of Wildlife Management 74:1112–1121. USA. HAINES-YOUNG, R., AND M. CHOPPING. 1996. Quantifying MCGARIGAL, K., AND B.J. MARKS. 1995. FRAGSTATS: landscape structure: A review of landscape indices and Spatial pattern analysis program for quantifying land- their application to forested landscapes. Progress in scape structure. US Forest Service General Technical Physical Geography 20:418–445. Report PNW-351, Washington, DC, USA. HIRZEL, A.H., G. LE LAY,V.HELFER,C.RANDIN, AND A. ———, S.A. CUSHMAN, M.C. NEEL, AND E. ENE. 2002. GUISAN. 2006. Evaluating the ability of habitat suit- FRAGSTATS: Spatial pattern analysis program for ability models to predict species presences. Ecological categorical maps. University of Massachusetts, Am- Modelling 199:142–152. herst, Massachusetts, USA. HOOGE, P.N., AND B. EICHENLAUB. 1997. Animal move- MORRISON, M.L., B.G. MARCOT, AND R.W. MANNAN. 1992. ment extension to ArcView. Version 2.0. Alaska Science Wildlife-habitat relationships: Concepts and applica-

Ursus 23(2):192–205 (2012) 204 EFFECTS OF HURRICANES ON BEARS N Murrow and Clark

tions. University of Wisconsin Press, Madison, Wis- SHEIKH, P.A. 2005. The impact of Hurricane Katrina consin, USA. on biological resources. CRS Report for Congress, MOSER, B.W., AND E.O. GARTON. 2007. Effects of telemetry RL33117, Washington DC, USA. location error on space-use estimates using a fixed- SIMBERLOFF, D. 1995. Habitat fragmentation and popula- kernel density estimator. Journal of Wildlife Manage- tion extinction of birds. Ibis 137:105–111. ment 71:2421–2426. STAVINS, R. 1986. Conversion of forested wetlands to NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION. agricultural uses: An econometric analysis of the impact 2009. Land cover data for Katrina impacted areas. of federal programs on wetland depletion in the lower Charleston, South Carolina, USA. http://www.csc. Mississippi alluvial plain, 1935–1984. Environmental noaa.gov/crs/lca/katrina/, accessed 20 January 2009. Defense Fund, New York, New York, USA. NEAL, W.A. 1990. Proposed threatened status for the STONEBURNER, D.L. 1978. Evidence of hurricane influence Louisiana black bear. Federal Register 55:25341–25345. on barrier island slash pine forests in the northern NYLAND, P.D. 1995. Black bear habitat relationships in Gulf of Mexico. American Midland Naturalist 99:234– coastal Louisiana. Thesis, Louisiana State University, 237. Baton Rouge, Louisiana, USA. STORMS,D.,D.SAI¨D,H.FRITZ,J.HAMANN,C.SAINT-ANDRIEUX, PACE, R.M., III., D.R. ANDERSON, AND S. SHIVELY. 2000. AND F. KLEIN. 2006. Influence of hurricane Lothar on red Sources and patterns of black bear mortality in and roe deer winter diets in the Northern Vosges, France. Louisiana. Proceedings of the Annual Conference of Forest Ecology and Management 237:164–169. the Southeastern Association of Fish and Wildlife SUEDEL, B.C., J.A. STEEVENS, AND D.E. SPLICHAL. 2006. A Agencies 54:365–373. pilot study of the effects of post-Hurricane Katrina PELTON, M.R. 2003. Black bear (Ursus americanus). floodwater pumping on the chemistry and toxicity of Pages 554–557 in G.A. Feldhamer, B.C. Thompson, violet marsh sediments. Environmental Laboratory and J.A. Chapman, editors. Wild mammals of North Technical Notes (ERDC/EL TN-06-1). US Army America. Johns Hopkins University Press, Baltimore, Engineer Research and Development Center, Vicks- Maryland, USA. burg, Mississippi, USA. RAMSEY, E.W., III., AND S.C. LAINE. 1997. Comparison of SWILLING, W.R., JR., M.C. WOOTEN, N.R. HOLLER, AND Landsat thematic mapper and high resolution photog- W.J. LYNN. 1998. Population dynamics of Alabama raphy to identify change in complex coastal wetlands. beach mice (Peromyscus polionotus ammobates) follow- Journal of Coastal Research 13:281–292. ing Hurricane Opal. American Midland Naturalist RAMSEY, E.W., A. RANGOONWALA,B.MIDDLETON, AND Z. 140:287–298. LU. 2009. Satellite optical and radar data used to track THOMPSON, L.M., F.T. VAN MANEN,S.E.SCHLARBAUM, AND wetland forest impact and short-term recovery from M. DEPOY. 2006. A spatial modeling approach to identify Hurricane Katrina. Wetlands 29:66–79. potential butternut restoration sites in Mammoth Cave RAO, C.R. 1952. Advanced statistical methods in biometric National Park. Restoration Ecology 14:289–296. research. John Wiley and Sons, New York, New York, USA. TORRES, A.R., AND P.L. LEBERG. 1996. Initial changes in RAY, G.L. 2006. Characterization of post-Hurricane habitat and abundance of cavity-nesting birds and the Katrina floodwater pumping on marsh infauna. Envi- Northern Parula following Hurricane Andrew. Condor ronmental Laboratory Technical Notes (ERDC/EL 98:483–490. TN-06-4). US Army Engineer Research and Develop- TRIANT, D.A., R.M. PACE, III, AND M. STINE. 2004. ment Center, Vicksburg, Mississippi, USA. Abundance, genetic diversity and conservation of REAGAN, D.P. 1991. The response of Anolis lizards to Louisiana black bears (Ursus americanus luteolus)as hurricane-induced habitat changes in a Puerto Rican detected through noninvasive sampling. Conservation rain forest. Biotropica 23:468–474. Genetics 5:647–659. SAI¨D, S., AND S. SERVANTY. 1995. The influence of TSOAR, A., O. ALLOUCHE,O.STEENITZ,D.ROTEM, AND R. landscape structure on female roe deer home-range KADMON. 2007. A comparative evaluation of presence- size. Landscape Ecology 20:1003–1012. only methods for modelling species distribution. SAS Institute Inc. 2004. SAS/STATH user’s guide, version Diversity and Distributions 13:397–405. 9.1. SAS Institute Inc., Cary, North Carolina, USA. TURNER, M.G., R.H. GARDNER, AND R.V. O’NEILL. 2001. SCHOEN, J.W. 1990. Bear habitat management: A review Landscape ecology in theory and practice: Pattern and and future perspective. International Conference on process. Springer-Verlag, New York, New York, USA. Bear Research and Management 8:143–154. US Bureau of the Census. 2009. Census 2000 TIGER/Line SCHRIEVER, T.A., J. RAMSPOTT, B.I. CROTHER, AND C.L. Data. http://www.esri.com/data/download/census2000_ FONTENOT. 2009. Effects of Hurricanes Ivan, Katrina, tigerline/index.html, accessed 13 February 2009. and Rita, on a southeastern Louisiana herpetofauna. US Fish and Wildlife Service. 1992. Endangered and Wetlands 29:112–122. threatened wildlife and plants; determination for

Ursus 23(2):192–205 (2012) EFFECTS OF HURRICANES ON BEARS N Murrow and Clark 205

threatened status for U. a. luteolus (Louisiana black Dissertation, Louisiana State University, Baton Rouge, bear). Federal Register 57:588–595. Louisiana, USA. ———. 1995. Louisiana black bear (Ursus americanus WANG, F., AND Y.J. XU. 2010. Comparison of remote luteolus) recovery plan. US Fish and Wildlife Service, sensing change detection techniques for assessing Southeast Regional Office, Atlanta, Georgia, USA. hurricane damage to forests. Environmental Monitor- ———. 1999. Final environmental assessment and land ing and Assessment 162:311–326. protection plan for the Louisiana black bear habitat WAUER, R.H., AND J.M. WUNDERLE,JR. 1992. The effect of protection project. US Fish and Wildlife Service, Hurricane Hugo on bird populations on St. Croix, US Southeast Regional Office, Atlanta, Georgia, USA. Virgin Islands. Wilson Bulletin 104:656–673. US Geological Survey. 2009. National hydrography WHITE, T.H., J.A. COLLAZO, F.J. VILELLA, AND S.A. dataset. Reston, Virginia, USA. http://nhd.usgs.gov/, GUERRERO. 2005. Effects of Hurricane Georges on accessed 13 February 2009. habitat use by captive-reared Hispaniolan parrots VAN MANEN, F.T., J.D. CLARK, S.E. SCHLARBAUM,K. (Amazona ventralis) released in the Dominican Repub- JOHNSON, AND G. TAYLOR. 2002. A model to predict the lic. Ornitologica Neotropical 16:405–417. occurrence of surviving butternut trees in the southern WIDMER, O., S. SAID,J.MIROIR,P.DUNCAN, J.M. Blue Ridge Mountains. Pages 491–497 in J.M. Scott, GAILLARD, AND F. KLEIN. 2004. The effects of hurricane P.J. Heglund, and M.L. Morrison, editors. Predicting Lothar on habitat use of roe deer. Forest Ecology and species occurrences: Issues of accuracy and scale. Island Management 195:237–242. Press, Covelo, California, USA. WILEY, J.W., AND J.M. WUNDERLE,JR. 1993. The effects of VERBYLA, D.L., AND J.A. LITVAITIS. 1989. Resampling hurricanes on birds, with special reference to Caribbean methods for evaluating classification accuracy of islands. Bird Conservation International 3:319–349. wildlife habitat models. Environmental Management WORTON, B.J. 1989. Kernel methods for estimating the 13:783–787. utilization distribution in home-range studies. Ecology WAGNER, R.O. 1995. Movement patterns of black bears in 70:164–168. south central Louisiana. Thesis, Louisiana State University, Baton Rouge, Louisiana, USA. ———. 2003. Developing landscape-scaled habitat selec- Received: 22 December 2011 tion functions for forest wildlife from Landsat data: Accepted: 18 June 2012 Judging black bear habitat quality in Louisiana. Associate Editor: J. McDonald

Ursus 23(2):192–205 (2012)