Elf owl - Species Distribution Model, DRECP

These data are statistical model outputs for elf owl (Micrathene whitneyi) species distribution, completed by CBI. Predictions of habitat occupancy were generated from Maxent models for the DRECP.

This species distribution model was produced for a limited extent within the DRECP region, defined as a union of 10 km buffers of the elf owl’s CWHR range (California Department of Fish and Wildlife, Biogeographic Data Branch, 2008), Colorado River, and detections, at 270 m resolution with 54 detections points obtained Feb. 2013 from CNDDB (California Department of Fish and Wildlife, Biogeographic Data Branch),  California Department of Fish and Wildlife, Wildlife and Habitat Data Analysis Branch (Elf Owl [ds14], http://bios.dfg.ca.gov) and ORNIS (http://www.ornisnet.org/).

The model was built with the following 4 environmental predictors (provided to CBI by Frank Davis’ Biogeography Lab at UC Santa Barbara, created for the CA Energy Commission’s project “Cumulative Biological Impacts Framework for Solar Energy in the CA Desert”, 500-10-021) in order of importance:

Maximum temperature of warmest period (°C, x10);
Temperature seasonality (C of V, x100);
Topographic relief in the 270m cell estimated as the standard deviation of elevations from 30m digital elevation model;
Perennial water features, as indicated by the USGS NHD feature codes 39004, 39009, 39010, 39011, 39012, 45800, 46006, and 46602.  Categorical presence/absence, indicating the presence of any perennial water feature within each 270m pixel.

This model has a 10-fold cross validated AUC score of 0.852 (standard deviation 0.073). 
 
Both continuous probability surfaces and binary layers are available. The binary layer depicting predicted suitable habitat was derived using the maximum training sensitivity and specificity threshold (0.350) as well as a 1 km buffer around isolated occurrences west of the Colorado River, where the thresholded version of the model showed no habitat, due to the coarseness of the environmental data layers.