Douglas Fir Habitat Suitability (CCSM4, CNRM-CM5, and MIROC-ESM), Golden Gate Biosphere Network

This ensemble model of habitat suitability for Douglas fir (Pseudotsuga menziesii) was projected across the California Floristic Province using 2070-2099 CCSM4, CNRM-CM5, and MIROC-ESM RCP 8.5 climate outputs from the California Basin Characterization Model (BCMv65, Flint and Flint 2014). The baseline ensemble model was generated with the flexSDM R package (Velazco et al. 2022) from 6 individual presence-absence species distribution modeling algorithms (generalized additive (GAM), generalized boosted regression (BRT), generalized linear (GLM), neural networks (ANN), random forest (RF), and support vector machine (SVM)). This ensemble model was created from all models with an area under the receiver operating characteristic (AUC) >= 0.7 using the weighted average method and the true skill statistic (TSS) performance metric.

This habitat suitability model used 749 presence and 773 absence points collected between 1980 and 2023 from multiple sources, cleaned for accuracy and quality, and filtered to 1-km minimum nearest neighbor distance. The model calibration extent was limited to a 5-km buffer of the known species range within the California Floristic Province. The model has a spatial resolution of 270m and is based on baseline (1981-2010) climate conditions and 10 predictors: winter precipitation, summer maximum temperature, winter minimum temperature, actual evapotranspiration, solar insolation index, topographic wetness index, terrain ruggedness, soil available water capacity, soil percent clay, and soil porosity.

Model performance was evaluated with both threshold-independent and threshold dependent metrics, and the model-specific threshold maximizing the sum of sensitivity and specificity was used for calculating threshold-dependent model evaluation metrics. This model had an AUC of 0.83, sensitivity of 0.84, specificity of 0.68, TSS of 0.52, Jaccard index of 0.64, and Sorensen similarity index of 0.77.

Values below the maximum sum of sensitivity and specificity threshold (0.3586) were assigned a value of 0. A land cover mask was used in post processing to convert the suitability value of high intensity development, barren, cultivated crops, and open water (NLCD 2021) land covers to 0 and the output was clipped to the Golden Gate Biosphere Network terrestrial boundary.