Riparian land-cover data and model code for: Multiple-region, N-mixture community models to assess associations of riparian area, fragmentation, and species richness
Fogarty, Frank (2022), Riparian land-cover data and model code for: Multiple-region, N-mixture community models to assess associations of riparian area, fragmentation, and species richness, Dryad, Dataset, https://doi.org/10.25338/B8VD1R
The associations of habitat area and fragmentation with species richness long have been major topics within community ecology. Recent discussion has focused on properly assessing fragmentation independent of habitat area (fragmentation per se), and on whether fragmentation has significant negative or positive associations with species richness. We created a novel, multiple-region, N-mixture community model (MNCM) to examine the relations of riparian area and fragmentation with species richness of breeding birds in mountain ranges within the Great Basin, Nevada, USA. Our MNCM accounts for imperfect detection in count data at the survey-point level while allowing comparisons of species richness among regions in which those points are embedded. We used individual canyons within mountain ranges as regions in our model and measured riparian area and the normalized landscape shape index, a metric of fragmentation that is independent of total riparian area. We found that riparian area, but not its fragmentation, was a primary predictor of canyon-level species richness of both riparian obligates and all species. The relation between riparian area and riparian-obligate species richness was nonlinear: canyons with ≥ 25 ha woody riparian vegetation had relatively high species richness, whereas species richness was considerably lower in canyons with <25 ha. Our MNCM can be used to calculate other metrics of diversity that require abundance estimates. For example, Simpson’s evenness of riparian obligate species had a weak negative association with riparian area and was not associated with fragmentation. Projections of future riparian contraction suggested that decreases in species richness are likely to be greatest in canyons that currently have moderate (~10-25 ha) amounts of riparian vegetation. Our results suggest that if a goal of management is to maximize the species richness of breeding birds in montane riparian areas in the Great Basin, it may be more effective to focus on total habitat area than on fragmentation of patches within canyons, and that canyons with at least moderate amounts of riparian vegetation should be prioritized.
We used data from 23 canyons in the Shoshone Mountains, Toiyabe Range, Toquima Range, and Monitor Range (central Great Basin, Nevada, USA).
We derived riparian area and fragmentation at the canyon level from 2013 National Agricultural Inventory Program (NAIP) images. We did not attempt to distinguish the cause of fragmentation or the historical extent of riparian cover in these canyons. There were no substantial disturbances or changes in land use in these canyons from 2001-2018 that would drive changes in the extent of riparian area or its fragmentation. It is possible that changes occurred gradually, but we believe that the slow rate of growth of woody vegetation minimized these changes over the temporal extent of our work. To calculate riparian area, we delineated a buffer (500 meters from the center line of the canyon bottom) that was large enough to contain all riparian vegetation in most canyons. We mapped riparian cover within the buffer in QGIS version 3.0.2 (QGIS.org 2018) and classified that cover as wet meadows (little to no perennial woody vegetation; dominated by grasses, sedges, and forbs) or woody riparian (extensive cover of perennials, primarily Betula occidentalis, Populus spp., Prunus virginiana, Salix spp., or Rosa woodsii). We selected these two classes because they are not used equally by riparian obligates. Although some riparian obligates forage in wet meadows, the meadows do not contain the woody vegetation in which most species nest. We were not confident in our ability to discern structural classes (e.g., trees versus shrubs) or individual woody species in the NAIP images and therefore did not further classify the images. Wet meadows were present in 9 of 23 canyons, and the cumulative area of those meadows in a given canyon was 0.3-13.1 ha. Woody riparian vegetation was present in all canyons, with a canyon-level mean cumulative area of 18.0 ha (range 0.3-96.5 ha). Overall, riparian vegetation was a small percentage of the land cover in our canyons (mean 2.7%, range 0.1-21.9%).
We selected the Normalized Landscape Shape Index (nLSI) (McGarigal et al. 2012) as our measure of fragmentation. This metric is not highly correlated with total area of patches of a given class (e.g., riparian vegetation) when that total area is <30% of the defined landscape (Wang et al. 2014), as is the case in our study system. nLSI for a sampling unit (canyons in our case) is equal to the cumulative perimeter of patches minus the minimum possible perimeter of those patches. nLSI then is standardized by dividing each value by the difference between the maximum and minimum possible cumulative perimeter of those patches in the sampling unit. The nLSI of a sampling unit with a single square patch would be 0, the nLSI of a sampling unit with a checkerboard distribution of patches would be 0.5, and the nLSI of a sampling unit with maximally disaggregated patches would be 1. Other fragmentation metrics that are not highly correlated with total area of a focal class of patches are less intuitive measures of variance, require a minimum number of patches, or require user delineation of core focal areas. We calculated nLSI of woody riparian in each canyon with the program FragStats (McGarigal et al. 2012). We did not calculate nLSI for wet meadows because nLSI cannot be calculated if the area of the focal class is zero in some sampling units, as was the case for several of our canyons. The correlation between all pairwise combinations of nLSI, woody riparian area, and wet meadow area in our canyons was <|0.56|, which was sufficiently low to include all three variables in a given model (Dormann et al. 2013).