The areas depicted in this shapefile were derived for an analysis of potential conservation priority areas on Bureau of Land Management (BLM) lands across 11 western states: Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming. Steps used to derive these core areas are described below. See Dickson et al. (2014) for a more comprehensive description of the methods and data sources below, and for a complete list of literature cited.
Specifying the study and analysis extent
Our study extent included BLM lands in the 11 contiguous western states: Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming . Ownership data were obtained for each state from their respective BLM websites or state geospatial data clearinghouses. Areas included in the analysis extent were located outside of existing special designations, including national monuments, wilderness areas, and wilderness study areas (WSAs). We excluded WSAs because they are provisionally protected until their designation is changed through legislative action by the U.S. Congress. We used land management designations in the U.S. Protected Areas Database and a geographic information system (GIS; ArcGIS v10.1, Esri, Redlands, CA) to identify and remove the aforementioned lands from our analysis extent.
Because we were interested in identifying candidate areas that could meet the criteria for the highest level of protection, i.e., wilderness designation, we also required lands in the analysis extent to be contiguous areas > 20.2 km2(or 5,000 ac, by convention the minimum size for wilderness designation in the U.S.) after removal of areas otherwise occupied by roads, railroads, and electric power transmission lines. Recognizing that no comprehensive dataset exists for all BLM lands that differentiates between maintained and unmaintained roads, we used 2011 TIGER/Line roads data to buffer (5 m per side) and remove all linear road features from the analysis extent. Additionally, we removed railroads and powerlines, also using a 5-m buffer. Contiguous areas < 20.2 km2were removed from consideration in our analysis extent.
We defined our analysis extent using readily available data on current road networks. Nevertheless, our results likely reflect errors in these data that would be difficult to quantify over an 11-state region, so we urge data users to scrutinize areas we assumed to be ‘roadless.’ Users should also be mindful of high value areas within BLMs jurisdiction where errors of commission (i.e., the presence of a road is falsely indicated by the TIGER/Line data) could have resulted in these areas being excluded from our analysis.
Deriving ecological indicators and spatial data layers
Through an extensive literature review and subsequent consultations with experts in the field of conservation indicators, we identified a suite of seven variables that were both readily available and spatially contiguous across our analysis extent to serve as ecologically based indicators of biodiversity, resilience to climate change, and/or landscape connectivity. These variables (i.e., data layers) included estimates of landscape naturalness and permeability, rarity-weighted species richness*, surface water availability, topographic complexity, vegetation community diversity, and ecoregional protection. Within a GIS, each indicator variable was derived as a raster data layer at a 270-m pixel resolution and summarized at three spatial scales (described in following section) across the 11 western states. We used a 270-m resolution because it presented a reasonable trade-off between certainty in the pixel value of a given indicator variable and computational efficiency.
Determining appropriate scales of analysis
We conducted our analyses at each of several spatial scales to explicitly address potential scale-dependencies of outcomes. We chose 20.2 km2 (5,000 ac), 80.9 km2 (20,000 ac), and 263.1 km2 (65,000 ac) (hereafter 20, 80, and 260 km2) as the most appropriate spatial scales for our analysis because 20 km2 is the mandated minimum size for wilderness areas in the U.S. (Wilderness Act 1964) and because we observed 80 and 260 km2 to be natural breaks in the distribution of sizes of contiguous areas > 20.2 km2in our analysis extent. At each scale, each indicator variable was summarized using a single focal statistic based on a circular moving window operation within a GIS. All focal statistics were generated at the extent of the 11 western states in a contiguous fashion before extracting the data to the analysis extent. This step attempted to prevent edge effects and ensure that the ecological context provided by lands adjacent to those in the analysis extent was not lost.
Statistical modeling
We conducted a principal components analysis (PCA) on the indicator variables to eliminate collinearities and to reduce the dimensionality of these data to a smaller set of new variables represented by principal component ‘axes.’These axes simultaneously integrated information from multiple indicators but also reduced computational costs. We standardized each variable to a mean of zero and unit variance prior to statistical modeling. We then used the first four PCA axes in a weighted linear combination (WLC) model (see below) for each scale of analysis. The first four axes explained a cumulative 85-88% of the variance in the data, depending on the scale of analysis. We used these new variables (axes) and the WLC modeling approach to assign a conservation score to each pixel in our analysis extent. This conservation score represented a quantitative measure of conservation value given the input data summarized at a particular spatial scale. WLC models involve assigning a weight to each variable according to its real or perceived importance. The final step of the WLC method typically involves identifying areas (in our case contiguous groups of pixels of a certain size) based on the rank of the conservation scores across the analysis extent. The area with the highest mean score is the ‘best’area. The results of WLC analyses, however, can be highly sensitive to the weights assigned to each variable. To address the potential sensitivity of WLC model results to the weights applied to each variable, we developed an approach to calculate a set (vector) of conservation scores for every pixel based on different weighting schemes. We then calculated the mean conservation score across all weighting schemes as an indicator of overall conservation value and the standard deviation of the conservation score across all weighting schemes as an indicator of the sensitivity of the score to the weighting scheme applied. These steps were repeated for each scale of analysis.
Quantifying and mapping scale-dependent core areas and ‘Conservation Priority Areas’
We defined two distinct types of high value areas based on the conservation scores at each scale of analysis. First, scale-dependent core areas were identified as those places within the analysis extent with high mean conservation scores and low standard deviation (i.e., sensitivity) values at each scale of analysis. These areas were identified by retaining any pixels above the 80thpercentile of mean conservation scores and below the 20thpercentile of sensitivity values (an 80/20threshold). Although any number of threshold values could be applied, we chose this threshold to demonstrate a reasonable and data-driven application of our results. Second, we identified where scale-dependent core areas overlapped at all three scales of analysis, indicating areas that were robust to choice of scale. We refer to these areas as conservation priority areas.
As a large-scale landscape analysis based on discrete variables associated with high climate resiliency, biodiversity, and/or landscape connectivity, this analysis does not address the importance of other values associated with intact, 'roadless' areas across the BLM domain. As a decision support tool for conservation management decisions, the analysis and associated results are intended as additive – not a substitute – to conservation prioritizations that may address other important values such as wilderness characteristics, sensitive species habitat, rare vegetation associations, archeological resources, scenic values, and other unique natural or cultural information.
Full metadata can be viewed upon download in the file named 'metadata1_original.xml'
*We thank NatureServe and its network of Natural Heritage member programs for use of the rarity-weighted richness index data layer.
Dickson, B.G., L.J. Zachmann, and C.M. Albano. 2014. Systematic identification of potential conservation priority areas on roadless Bureau of Land Management lands in the western United States. Biological Conservation, 178:117-127.