Least Bell's Vireo Habitat Suitability Model - California - 2019 - USGS [ds2949]

This habitat model was developed to identify suitable habitat for the federally-endangered least Bells vireo (Vireo bellii pusillus) across its current and historic range in California. The vireo disappeared from most of its range by the 1980s, remaining only in small populations in southern California. Habitat protection and management since the mid-1980s has led to an increase in southern California vireo populations with small numbers of birds recently expanding into the historic range. Predictions from this model will be used to focus surveys in the historic range to determine where vireos are recolonizing and to track the status and distribution of populations over time.

We used the Partitioned Mahalanobis D2 modeling technique to construct alternative models with different combinations of environmental variables. We developed calibration models for the current range in southern California using vireo locations recorded from 1990 to present. We selected spatially non-redundant observations reflecting average, below average and above average rainfall conditions. For each rainfall condition, we selected three to four years of spatially non-redundant location data from the period 1990-2013. We used this dataset to randomly select 70 percent of the observations for a calibration dataset and used the remaining 30 percent of observations as a validation dataset. We used supplementary validation datasets with observations from 2016, 2017 and 2018 representing average, above average, and below average rainfall conditions, respectively.

We cross-walked and merged detailed digital vegetation maps for southern California and utilized the Fire Resource Assessment Program 2015 Vegetation Map as a base map for the rest of California. We used the Klausmeyer et al. (2016) Groundwater Dependent Ecosystems map to capture riparian areas not mapped with other source layers. We selected riparian vegetation types used by vireos to develop a grid of riparian points spaced 150m apart and buffered with 500m of adjacent non-riparian habitats. We calculated environmental variables at each grid point in the center of a 150m x 150m cell for the grid of points in this modeling landscape. Variables reflect various aspects of topography, climate, and land use (percent riparian vegetation and urbanization at 150m, 500m and 1km scales). We developed several Normalized Difference Vegetation Index (NDVI) variables based on means, maximums and percentages of pixels with a minimum specified value at the 150m and 500m spatial scales.

We developed alternative calibration models with different combinations of environmental variables reflecting hypotheses about least Bells vireo habitat relationships. Due to spatial unevenness in vireo location data, we divided southern California into ten sampling regions and randomly subsampled 70 locations from each region. We repeated this process 1,000 times using a total of 2,270 spatially precise and non-redundant vireo locations in the calibration dataset. We model-averaged the results from sampling iterations to create a calibration model with partitions for each set of variables. We compared among these calibration model-partitions using the randomly selected validation dataset of 972 observations and the 2016, 2017 and 2018 validation datasets of 610, 1,066, and 882 observations, respectively. We created a presence and pseudo-absence dataset for evaluating each model-partitions performance with the combined 3,530 observations in the validation datasets and 3,566 pseudo-absence points randomly selected from a grid of points encompassing the vireos current range in southern California.

For every model-partition, we calculated Habitat Similarity Index (HSI) predictions for presence and pseudo-absence points ranging from Very High (0.75-1.00); High (0.500.74); Low to Moderate (0-0.49). Suitable habitat is identified as grid cells with HSI equal to or greater than 0.5. We calculated Area Under the Curve (AUC) values from a Receiver Operating Curve (ROC) to determine how well models distinguish between the combined presence and pseudo-absence points. We selected a set of best performing calibration model-partitions based upon median HSI calibration and validation values and AUC results. We then used these models to predict suitable habitat for the riparian grid across California, including the current and historic range. We qualitatively evaluated how well the model-partitions predicted suitable habitat in the historic range across California for historic and recent vireo records in the California Natural Diversity Database. We also used e-Bird observations to qualitatively assess how well the model predicted habitat at recently observed vireo locations in the historic range. Several top-performing model-partitions for southern California did not predict suitable habitat in the historic range. These models included climate variables, elevation, and NDVI variables, which vary widely between the current and historic ranges. Model 30 Partition 1 is the best model-partition for predicting habitat in both the current and historic ranges across California. This model-partition has an AUC of 0.98 and median calibration and random validation HSI values of 0.70. Supplementary validation datasets for 2016, 2017 and 2018 have median HSI values of 0.66, 0.64, and 0.63, respectively. This model includes the following variables: median slope, percent flat land, and percent riparian vegetation at the 150m-scale and distance from water (m). We mapped HSI predictions from this model for each cell in the 150m-scale grid across the California riparian study area to create the habitat suitability map.

The authors of these data require that users direct any questions pertaining to appropriate use or assistance with understanding limitations and interpretation of the data to the individuals/organization listed in the Point of Contact section.

Disclaimer: The State makes no claims, promises, or guarantees about the accuracy, completeness, reliability, or adequacy of these data and expressly disclaims liability for errors and omissions in these data. No warranty of any kind, implied, expressed, or statutory, including but not limited to the warranties of non-infringement of third party rights, title, merchantability, fitness for a particular purpose, and freedom from computer virus, is given with respect to these data.