Occupancy modeling of Black‐backed Woodpeckers on burned Sierra Nevada forests

The Black-backed Woodpecker (Picoides arcticus) has been designated by the USDA Forest Service as a management indicator species for snags in burned conifer forests of the Sierra Nevada of California, USA. However, little is known about the characteristics that affect between-fire and within-fire habitat selection by the species in the region. Here we report on the first broad-scale survey of Black-backed Woodpeckers on wildfire-affected forests of the Sierra Nevada. We implemented a Bayesian hierarchical model to: 1) estimate Black-backed Woodpecker occupancy probability in fire areas burned within a time window of 1–10 years; 2) identify relationships between occupancy probability and habitat covariates (fire age, change in canopy cover pre-to-post fire, and snag basal area), elevation, and latitude; and 3) estimate detection probability and relate it to survey interval length and survey type (passive v. broadcast). We included random fire-area effects in our model of occupancy probability to accommodate clusters of non-independent points surveyed within the larger set of fire areas. Mean occupancy probability was estimated to be 0.097. Elevation (after controlling for latitude) had the strongest effect on occupancy probability (higher occupancy at higher elevation) followed by latitude (higher occupancy at northerly sites). Fire age was also important; occupancy probability was about 4× higher on the youngest compared to oldest fires. Although the direction of regression coefficients were in the expected direction (positive), snag basal area and canopy cover change were of minor importance in affecting occupancy probability. There was some indication, however, that the importance of snag basal area increased with fire age. Weak links between occupancy and canopy cover change suggested the species uses a range of burn severities, and such heterogeneity may promote habitat longevity. Our estimate of overall detection probability (across all survey intervals) was 0.772. We found strong effects of survey interval length (higher for longer interval) and, especially survey type (higher for broadcast survey) on detection probability. Our modeling framework and implementation illustrates the flexibility of the Bayesian hierarchical approach for handling complexities such as estimating derived parameters (and variances) and modeling random effects, and should prove generally useful for occupancy studies.