Landscape Genetics and Landguth, 2010) Together, these studies warrant caution regarding use of partial Mantel tests, careful construction of hypotheses, and the need to use alternative statistical tests to confirm partial Mantel results Barrier Detection As of 2010, approximately half of the landscape genetics studies focused on understanding barriers to gene flow using a variety of approaches (Storfer et al., 2010) Interest in barriers to gene flow is broad, spanning from understanding basic landscape features that generate genetic population structure, to understanding the conservation implications of anthropogenic development (e.g., roads and dams) Early studies sampled on both sides of a putative barrier and compared genetic distance across the barrier with samples taken at similar distances in continuous habitat (Storfer et al., 2007) If significantly decreased gene flow was found across the putative barrier and not the continuous habitat, it was concluded that the landscape feature was, in fact, a barrier An example was a study that showed that a California highway acted as a barrier to gene flow for both bobcats and coyotes (Riley et al., 2006) More recently, methods in landscape genetics and spatial statistics have together allowed both more quantitative and spatially explicit ways of investigating potential barriers such as mountains, as well as ways to detect breaks in genetic continuity caused by less obvious or even cryptic landscape features, such as changes in moisture patterns across a study area (Storfer et al., 2007) One approach for detecting changes in allele frequencies in space is Wombling (Barbujani et al., 1989; Womble, 1951) Wombling has been used to identify spatial locations of abrupt changes in allele frequencies by quantifying local variability in allele frequencies (Cercueil et al., 2007) The software WOMBSOFT (Crida and Manel, 2007) utilizes local polynomial regressions to estimate these genetic changes, as well as a binomial test for significance of each genetic break A recent example is a study of Giant pandas in China, where a Wombling analysis showed genetic breaks between the Xiaoxiangling and Daxiangling Mountains (Zhu et al., 2011) Another approach is using Monmonier’s maximum difference algorithm, which creates a break line through the vectors between a network of sampled populations or individuals (Monmonier, 1973) Essentially, the Monmonier algorithm estimates the largest changes in allele frequencies along this break line, and thereby allows inference of barriers in these locations Noted drawbacks of this method include dependence on the initial sampling design, as well as an a priori definition of the number of genetic groups present in the study area (Guillot et al., 2009) One way to circumvent the necessity of predefining the number of (genetic) populations is by using assignment tests Assignment tests are most commonly used to determine the most likely number of genetic clusters (or K; i.e., populations) based on allele frequencies among samples using a Bayesian approach and Monte Carlo Markov Chain (MCMC) optimization (Pritchard et al., 2000; Guillot et al., 2008) Researchers should be aware that K is estimated in a variety of different ways, depending on the assumptions of the method, such as whether there is admixture or not, and the software 511 used (Rowe and Beebee, 2007; Francois and Durand, 2010; Guillot et al., 2009) Barriers or edges are commonly inferred when genetic clustering coincides with one or more geographic features consistent with those that would impede movement of the study organism (Figure 2) For example, the program GENELAND (Guillot et al., 2008) was developed as a spatially explicit assignment-based approach to determine the number of genetic clusters (i.e., K) present in a dataset Tessellation is used to construct a series of polygons that denotes the boundaries of the genetic clusters in space, independent of the sampling sites (Guillot et al., 2008) Barriers can thus be inferred to coincide with the spatial locations of the polygon edges that delineate the genetic clusters An individual-based assignment test method provides another approach for detecting genetic discontinuities among samples while avoiding population delineation (Manel et al., 2007) This method generates a spatially referenced probability map for finding the genotype of a particular individual using a moving window approach A mean slope for all individual probability maps is calculated, and population boundaries are identified by areas of high mean slope (Manel et al., 2007) A comparison of assignment tests and other edge detection methods, including Wombling and Monmonier’s algorithm, showed that assignment tests generally perform better on both simulated and empirical data (Safner et al., 2011) Reviews of the applications of, and statistical issues with, assignment tests can be found in Manel et al (2005), Guillot et al (2009), and Francois and Durand (2010) Resistance Surfaces Another major way to assess effects of landscape features on the spatial distribution of genetic variation is with the use of resistance surfaces (Spear et al., 2010) Generally, resistance surfaces are estimated by assigning values to each landscape variable or environmental feature contained in a spatial layer of a raster GIS environment (Spear et al., 2010) Higher assigned values translate to greater hypothesized resistance to gene flow by that particular variable, whereas lower values suggest lower resistance Thus, each grid square (e.g., 30 m or km) in a GIS is assigned a cumulative value based on hypothesized habitat use and movement through each of the landscape variables contained in that grid square The resistance surface comprised of all grid squares that make up the intervening habitat matrix between sampling localities across a study area (Figure 3) Rarely researchers have direct estimates of dispersal and habitat use to parameterize such surfaces, and therefore assignment of resistance values to each landscape feature remains a challenge for modeling resistance surfaces (Spear et al., 2010) As a result, the majority of resistance surfaces are constructed using subjective methods; a recent review suggested only approximately one-fourth of studies used nongenetic field data, such as radio-telemetry data or direct movement studies, to inform resistance assignment values (Spear et al., 2010) One way to deal with the problem of subjectively assigning resistance values is by assigning a range of values to each landscape variable, thereby generating a variety of surfaces that comprise alternative hypotheses Then, the multiple surfaces can be statistically