A Mixed-Effects Heterogeneous Negative Binomial Model for Postfire Conifer Regeneration in Northeastern California, USA

Abstract

Many western USA fire regimes are typified by mixed-severity fire, which compounds the variability inherent to natural regeneration densities in associated forests. Tree regeneration data are often discrete and nonnegative; accordingly, we fit a series of Poisson and negative binomial variation models to conifer seedling counts across four distinct burn severities and three forest types 10 years after the 23,000-ha Storrie Fire, a large mixed-severity fire in northern California. Despite the accessibility and power of the zero-inflated negative binomial mixture model, a flexible heterogeneous negative binomial model offered a superior fit. Incorporation of a random stand effect further improved model performance. A parametric bootstrap analysis was conducted to examine seedling distributions and stand stocking. Mean simulated seedling densities had an expansive range (272–29,257 ha−1). Stocking analyses suggest a high probability of deficient conifer coverage in the majority of lower-elevation high-severity burn stands. In addition, models were fit to fir and pine seedling counts. Only a minority of postfire stands were likely to be stocked in the pine-only analysis. These models will help land managers prioritize limited resources for artificial reforestation in mixed-severity burned landscapes.