These are statistical model outputs for the distribution of the mule deer (Odocoileus hemionus), completed by CBI. Predictions of relative habitat suitability were generated for the northern Sacramento Valley from a multi-scale MaxEnt (habitat suitability model for presence-only data; Version 3.3.3k, Phillips et. al. 2006) model.

This 90m resolution species distribution model was calibrated within a limited extent within the Great Valley, Northern California Interior Coast Ranges, and Sierra Nevada Foothills ecoregion sections, defined as a 10km buffer of this species range (California Wildlife Habitat Relationships (CWHR), California Department of Fish and Wildlife 2014). Detection data from 1990 on were obtained from BISON (Biodiversity Information Serving Our Nation, http://bison.usgs.ornl.gov, 2016-08-04) and iNaturalist (http://www.inaturalist.org, 2016-12-20), thinned to a minimum nearest neighbor distance of 500m, and divided into a model training set (n = 38) and testing set (n = 13).

The model included the following 4 environmental predictors in order of mean permutation importance:

Forest/woodland/shrub/herbaceous interspersion and juxtaposition index (over 3960m radius moving window);

Distance to perennial wetlands, lakes, ponds, and streams;

Forest/woodland/shrub/herbaceous proximity index (over 3960m radius moving window); and

Mean percent slope (over 540m radius moving window).

This model has a mean 10-fold cross-validated test AUC score of 0.869 (standard deviation 0.056), mean 10% test omission of 0.125, mean difference between training and testing AUC of 0.011, and correctly classified 61.5% of the reserved test occurrences (n= 13; using maximum training sensitivity and specificity threshold).

The model was projected across the northern Sacramento Valley (defined as the Great Valley, Northern California Interior Coast Ranges, and Sierra Nevada Foothills ecoregional sections clipped to within these counties: Butte, Colusa, El Dorado, Glenn, Lake, Napa, Nevada, Placer, Sacramento, Shasta, Solano, Sutter, Tehama, Yolo, and Yuba). Areas outside the limited calibration extent described above should be interpreted with more caution. Both the logistic continuous probability surface and binary layer are available. The binary layer depicting predicted suitable habitat was derived using the maximum training sensitivity and specificity threshold (0.409). Results are preliminary and have not yet been reviewed by expert biologists.