Projecting richness from inventories
Where insufficient data exists for species specific modelling, projecting richness from inventories may be the only possibility. This requires a lower resolution than many other approaches (as inventories must be drawn from an area) and can also not reflect biogeographic differences, meaning that where island biogeography is important it cannot account for that. Projecting richness using this approach relies on site-based inventories of species present. To do this we first used the same dataset as above, we created a 10km2 fishnet as a polygon grid. The fishnet was trimmed to continental boundaries using the same clip as previously used. This was then imported along with point data into QGIS 3.26.3 and the sampling point tool used to intersect grids with point data, with the FID of each grid used in an additional column as a unique ID for the grid. We then reimported this into ArcMap 10.8 and used the summary statistics tool to calculate the number of points and species per grid cell. This was then reconnected to the original grid using joins and relates, and cells with at least 30 unique records were selected as inventories, the latitudes and longitudes of the centroids added, and this data exported to a CSV. To better reflect appropriate data from adequately inventoried but potentially less diverse sites we then added the publically available data from the Darkcide database (Tanalgo et al 2022) once we removed any listings with only a single species present (which may reflect species specific inventories, for example a number of sites only listedOtomops species). The final dataset had 417 inventories for the region which were then used for modelling.
This data could then be modelled for richness. This has the advantage that we can use the same variables as used for species specific modelling, including not requiring the use of a landcover map as a filter and thus enabling more nuance. Variables selected are noted in supplemental data, these were chosen to reflect climate parameters, and continuous metrics of habitat structure. Variables included actual evapotranspiration, annual mean potential evapotranspiration, aridity, two metrics of distance to bedrock (bdticm, bedroom from ISRIC world soil grids, as well as Estimated soil organic carbon stock as a measure of fertility), continental moisture index, continentality, embergers pluviothermic quotient, growing degree days 5, potential evapotranspiration of the driest quarter (resources during the most limiting time of year) potential evapotranspiration seasonality, thermicity (from Envirem) and bio 3,4,5,6,12,13,14 and 15 from Worldclim as well as vegetation canopy height.
Therefore for both forms of modelling (richness modelling based on inventories and species modelling) we used canopy height of all vegetation, a range of soil variables, and variables representative of climate, moisture and seasonality. Richness inventories were then grouped into classes (i.e. under 5 species, 5-10 species, etc) and modelled as individual classes using Maxent, all outcomes had an AUC of over 0.9. Within Maxent we modelled each richness level, ran three replicates (and used an average) and used default parameters. The average was then reclassified in ArcMap 10.8 using the 10 percentile cloglog threshold as a minimum bound of suitability, and then using an equal division between the threshold and the maximum value of 1 to reflect the maximum and minimum values of the richness level, with areas “unsuitable” given a value of 0. The mosaic to new raster tool was then used with the selection set to “maximum” to give the maximum number of species any given area was suitable for based on model outcomes.