We focus our comparison on the Columbia torrent salamander, Rhyacotriton kezeri, which is endemic to coastal southwest Washington and northwest Oregon and has a restricted geographic range compared to other amphibians (Whitton et al. 2012), and its congener the southern torrent salamander, R. variegatus, which has a somewhat larger yet still restricted range extending south to northern California from their shared range boundary in coastal Oregon (Fig. 1). Both species are of conservation concern (ODFW 2008; CDFW BDB 2011), with R. kezeri proposed for listing under the U.S. Endangered Species Act (WSDOT 2017). Forest disturbances are key concerns for both species, as their ranges are highly managed for wood production in the Pacific Northwest, where timber harvest and associated roads may degrade and fragment stream breeding and upland dispersal habitats (Bury and Corn 1988; Corn and Bury 1989). Other main land uses in these species' ranges include paved roads and rural development. Also, fire has periodically affected forest habitat in these species' ranges. For example, with the 1987 Silver Fire Complex (566 km2) and 2002 Biscuit Fire (2000 km2) in the R. variegatus range in Oregon, and four fires from 1933 to 1951 collectively known as the Tillamook Burn (1400 km2) in the R. kezeri range in Oregon. Approximately 28% of the historical area of old-growth conifer forest remains in this region of the U.S. as of 2000, which is estimated to be 46,656 km2 (Strittholt et al. 2006). More recently, climate change projections have raised concerns for R. variegatus, as its southern and interior distribution may be limited by historically warm, dry climates that may become warmer and dryer in the future (Bury 2015). A comparative landscape genetic study of R. kezeri and R. variegatus may help inform management efforts for these species of concern by revealing landscape-scale associations with genetic structure, which can be used to identify forest landscape planning priorities in the area.
To compare the habitat fragmentation with genetic diversity for each species, we used land cover data based on a 16-category scheme from the National Land Cover Database (Homer et al. 2007). We trimmed this layer to rectangles bounded by the maximum spatial extent of the localities for each genetic cluster identified per species, plus a 10-km buffer, and combined the three forest categories into one. Next, we calculated four class metrics based on all forest patches (contiguous forest areas) within each rectangle (landscape) using FRAGSTATS v4.2 (McGarigal et al. 2012). The first metric is the proportion of the landscape covered by forest, ranging from no forest (0) to entirely forested (1). Next, we examined patch density, the number of patches per 100 hectares, which ranges from no forest (0) to a value constrained by the cell size of the landscape data and is informative as a comparative measure because resolution is held constant across landscapes. Additionally, we used two metrics based on the cumulative distribution of patch sizes. The landscape division index, defined as the probability that two randomly selected pixels of the same patch type are in different patches, ranges from entirely forested (0) to almost entirely non-forest with only a single small forest patch as it approaches 1. Finally, we calculated the splitting index, which is the number of patches of equal size that could yield the same landscape division index, ranging from entirely forest (1) to the number of cells in the landscape squared. Thus, higher values reflect increased forest fragmentation for all of these metrics. The results were compared qualitatively with the genetic diversity measures for each genetic cluster.
As in forest fragmentation analysis, we created rectangular resistance surfaces for each genetic cluster bounded by the maximum spatial extent of the localities plus a 10-km buffer for each of ten different landscape and climate variables that were important in previous amphibian landscape genetic studies (see Spear and Storfer 2008, 2010; Trumbo et al. 2013; Zancolli et al. 2014; Emel and Storfer 2015). Three of ten landscape variables analyzed were the discrete variables of land cover (described above; Homer et al. 2007), road cover (North American Detailed Streets, ESRI, Redlands, CA), and stream cover (Streamnet GIS Data 2003), which we downloaded as vector data and converted into raster data. The seven remaining variables examined were continuous. These variables included percent canopy cover based on tree canopy density (Homer et al. 2007), annual frost-free period (Rehfeldt 2006), annual growing season precipitation (Rehfeldt 2006), compound topographic index (Moore et al. 1993; Gessler et al. 1995), heat load index (McCune and Keon 2002), roughness (Blaszczynski 1997; Riley et al. 1999), and slope. Canopy cover, frost-free period, and growing season precipitation were downloaded in the form of raster files. Compound topographic index, heat load index, roughness, and slope were calculated from the National Elevation Dataset digital elevation model (Gesch et al. 2002; Gesch 2007) using the Geomorphometry and Gradient Metrics toolbox v2.0-0 for ArcGIS (Evans et al. 2014). Compound topographic index is an estimate of soil wetness based on the expected movement of water downhill, heat load index reflects the amount of solar radiation reaching a surface, and roughness quantifies the change in elevation per unit area (Evans et al. 2014). To test for nonlinear relationships between these continuous variables and genetic distance, after rescaling all variables to range from 0 to 1 we created two transformed resistance surfaces for each variable: \(\textT1=100^\left(original cost surface\right)\), and \(\textT2=100-100^(1-original cost surface)\). T1 represents a scenario in which the slope of the resistance function is shallow at low resistance and steep at high resistance, whereas T2 has the opposite effect on the resistance function (Trumbo et al. 2013).
All resistance layers created for the landscape variables were at 30-m resolution with the exception of frost-free period and growing season precipitation, which were at 0.008333 decimal degrees, or approximately 825-m resolution. We created resistance surfaces for each variable using ArcGIS 10.0 (ESRI, Redlands, CA). For canopy cover, frost-free period, growing season precipitation, and topographical roughness, each raster cell reflected the value measured for that variable. We predicted that high canopy cover, high compound topographic index, long frost-free period, and high growing season precipitation would promote gene flow, hence we subtracted the value of each cell from the maximum value across all regions for each variable, so that low values would result in high cost. For heat load index and slope, we used the raw value as the cost, because we predicted high values of these variables to be associated with low gene flow. Because the effect of land cover type on these salamanders beyond simply the degree of canopy cover is unknown, we examined categorical land cover in addition to percent canopy cover described above by assigning a resistance of 10 or 100 for developed or barren land, 5 or 50 for non-forested vegetated land (e.g. grassland/herbaceous), and 1 for forested habitat, with the expectation that genetic distance will increase for paths crossing non-forested or developed land compared to forest. For stream cover, we first converted the vector data, which represents streams as linear elements, to raster data at 30-m, where each cell denotes the presence or absence of a stream. We then developed three alternative cost ratios for the resistance associated with dispersal over land versus dispersal via stream corridors because it was unknown a priori how strongly land restricts movement in Rhyacotriton over stream, but likely that it would be important for shaping gene flow. In all cases, stream pixels were assigned a resistance value of 1, while land pixels were assigned a value of 2, 10, or 100, because movement on land would presumably be at least twice as costly as movement within streams, but potentially quite a bit more costly (see Spear and Storfer 2008, 2010). We followed a similar procedure for road cover, except that we assigned non-road pixels a resistance of 1, and road pixels a resistance of 2, 10, or 100 because presumably roads are barriers to gene flow for these species. Thus, the best-supported cost ratios for these linear variables were determined via model selection (see below). Finally, for all variables, pixels containing high order rivers (Gary et al. 2010) or ocean were assigned a high cost of 20,000 to prevent the creation of unrealistic paths that would assume movement through these water bodies.
Our comparison of two closely related, ecologically similar species suggests that landscape genetic patterns can be consistent across broad spatial scales, yet ecological differences in habitat characteristics and disturbance can differentially affect landscape genetic structure at small scales. Overall, we determined that land cover and roads are the strongest predictors of genetic distance in these two torrent salamanders, but within the genetic clusters in each species there is variation in the relative importance of these and other variables related to minimizing desiccation. Our results also suggest that decreased genetic connectivity and lower genetic diversity in R. kezeri may be associated with disturbances that reduce forest cover, however, similar patterns of forest fragmentation were found in the northern portion of the study areas for both species. Forest cover can be affected by timber harvest, other land uses, landslides, and wildfire, and in the area analyzed, timber harvest is the most pervasive disturbance to landscape-scale forest cover (Nickerson et al. 2011). Given the smaller geographic range of R. kezeri compared to R. variegatus, the effects of fragmentation in a large proportion of this area may have stronger negative effects on observed genetic diversity. Furthermore, the specific landscape context, variation in degree of human development, biotic interactions, or evolutionary history may play a role in shaping genetic diversity.
 38c6e68cf9