Optimization of forest sampling strategies for woody plant species distribution modelling at the landscape scale

Hierarchical species distribution models in support of vegetation conservation at the landscape scale

“…The findings presented here are promising and suggest avenues for future research. In particular, hierarchical SDMs could be further evaluated and tested for their use in a variety of local conservation efforts (see Guisan et al., ), such as reintroduction programs (Araújo & Peterson, ), management of endangered plant species (Wang et al., ), species selection in vegetation restoration programs (Gastón et al., ), comparing various scale preferences in partnership with stakeholders (Hughes et al., ), forest management (Mateo et al., ), and, especially, in the development of local restoration programs in the context of climate change (Eitzel et al., ; Jiang & Zhang, ). The consequences of climate change for the application of vegetation restoration, in particular, need further investigation (Harris, Hobbs, Higgs, & Aronson, ).…”

“… Regional scale (Spain) training database: Species presences and absences of the five tree species (Table ) were gathered from the Spanish National Forest Inventory for mainland Spain (76,347 circular plots of 25 m radius). It follows a network with nodes separated by 1,000 m. This database was used to develop models at the regional scale. Landscape scale (Biosphere Reserve) training database: Based on previous work in the study area (Mateo et al., ), an opportunistic sampling along roads and tracks was conducted. A total of 302 circular (10 m radius) plots were sampled at 1,000‐m intervals, with presence and absence of tree species recorded in each plot (Table ).…”

“…In order to represent these factors, we generated different predictor variables at 25 m resolution: northness (orientation along the South–North axis), eastness (orientation along the West–East axis), monthly solar radiation, least surface distance to any stream, compound topographic index, heat load index, tree canopy cover fraction derived from a LIDAR image, elevation, curvature, and slope. To avoid multicollinearity, we used the same approach as for the regional predictors (above) based on pairwise correlations, to finally keep the five most uncorrelated and ecologically meaningful variables for explaining plant species distribution patterns (see; Mateo et al., ): (a) distance to streams: for each pixel, we calculated the least surface distance (m) to any stream (Fleming & Doan, ); (b) canopy cover above three meters derived from a LIDAR image (Centro Nacional de Información Geográfica): this is the fraction of ground area covered by the vertical projection of tree crown perimeters; (c) heat load index (HLI): this was calculated according to the method of McCune and Keon (), where a southwest‐facing slope has warmer temperatures than a southeast‐facing slope; (d) elevation: derived from a 25‐m resolution digital elevation model (DEM, Centro Nacional de Información Geográfica); (e) slope: derived from the DEM. All the independent variables were generated at 25 m resolution and projected at ETRS89 UTM 30N using ArcGIS 10.3 (ESRI) and R 3.3.2 (R Core Team) software.…”

“…However, despite the high potential applicability of SDMs for plant restoration (Gastón et al., ), they have rarely been applied for this purpose (but see; Burnside, Smith, & Waite, ; Gastón et al., ; van Loon et al., ). This lack of restoration use might be caused by limitations in conceptual and methodological issues arising when applying SDMs to conservation programs at the landscape scale, such as the need to integrate hierarchical levels (Gastón & García‐Viñas, ), or the choice of an appropriate sampling strategy to sample the data at the landscape scale (Mateo, Gastón, Aroca‐Fernández García‐Viñas & Saura, ).…”

“…So far, many modeling studies, even when aimed at conservation applications, have been conducted over large extents (regional or continental) and coarse resolutions (km; Araújo, Thuiller, Williams, & Reginster, ; Mateo et al., ). However, modeling at the landscape scale with fine spatial resolution is, needed to accurately guide local vegetation conservation programs (Mateo et al., ). In this regard, a hierarchical modeling approach (Gallien et al., ; Pearson et al., ; Petitpierre et al., ) was shown to provide the appropriate tool to account for those environmental factors driving entire species range together with the resolution needed to account for processes operating over a smaller extent and at finer resolution, i.e., at the landscape scale, such as microtopography (Lenoir, Hattab, & Pierre, ; Vicente et al., ).…”