The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman & Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo’package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence’package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif’format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

This dataset contains the modeling results GIS data (maps) of the study “Sustainable Human Population Density in Western Europe between 560.000 and 360.000 years ago” by Rodríguez et al. (2022). The NPP data (npp.zip) was computed using an empirical formula (the Miami model) from palaeo temperature and palaeo precipitation data aggregated for each timeslice from the Oscillayers dataset (Gamisch, 2019), as defined in Rodríguez et al. (2022, in review). The Population densities file (pop_densities.zip) contains the computed minimum and maximum population densities rasters for each of the defined MIS timeslices. With the population density value Dc in logarithmic form log(Dc). The Species Distribution Model (sdm.7z) includes input data (folder /data), intermediate results (folder /work) and results and figures (folder /results). All modelling steps are included as an R project in the folder /scripts. The R project is subdivided into individual scripts for data preparation (1.x), sampling procedure (2.x), and model computation (3.x). The habitat range estimation (habitat_ranges.zip) includes the potential spatial boundaries of the hominin habitat as binary raster files with 1=presence and 0=absence. The ranges rely on a dichotomic classification of the habitat suitability with a threshold value inferred from the 5% quantile of the presence data. The habitat suitability (habitat_suitability.zip) is the result of the Species Distribution Modelling and describes the environmental suitability for hominin presence based on the sites considered in this study. The values range between 0=low and 1=high suitability. The dataset includes the mean (pred_mean) and standard deviation (pred_std) of multiple model runs.

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo’package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence’package in R (Freeman & Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif’format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo’package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence’package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif’format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo’package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence’package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif’format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo’package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence’package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif’format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].

The Maxent modeling algorithm was used to build the species distribution model at 270 m spatial resolution using species occurrence points and environmental layers as predictors (Phillips et al. 2006). Species occurrence points were primarily obtained from CNDDB (California Natural Diversity Database) and other CDFW sources, GBIF (Global Biodiversity Information Facility), PRBO (Point Blue Conservation Science) and Arctos museum databases. Vegetation, distance to water, elevation, and bioclimatic variables (Franklin et al. 2013) were used as predictor variables. The models were run at 270 m spatial resolution with five replications using cross-validation as a method of sample evaluation. Cross-validation involved the partitioning of the sample data into n subsets, fitting the models to n-1subsets, and testing the model on the one subset not used in fitting the model. Initial model runs showed that our models converged around 2,000 iterations and for this reason we ran all models with 2,500 maximum iterations. Maxent was implemented in R using the ‘dismo''package (Hijmans et al. 2011). Model evaluation was carried out using the ‘PresenceAbsence''package in R (Freeman and Moisen 2008). We used AUC as a metric to evaluate model performance. The package also computes threshold values using several accuracy metrics to translate predicted probability maps into binary suitable and unsuitable habitats. We selected the MeanProb, a threshold set based on the mean predicted probability of species occurrences. The output from Maxent are grid datasets in a multiband ‘tif''format with one band for each replication. We averaged the five replicated maps and created a mean grid for each species. The grid was then symbolized to represent low (threshold-50), medium (50-75) and high (75-100) habitat suitability, with pixel values that are below the threshold excluded. Models were reviewed by CDFW species experts; please review the use limitations.For more information see the project report at [https://nrm.dfg.ca.gov/FileHandler.ashx?DocumentID=85358].