Work

The fate of California’s forests under climate change, novel fire regimes, and management interventions

Year: 2026
Type: dissertation
Abstract: The fire regimes, or the frequency, severity, spatial pattern, and seasonality of wildland fires of forested ecosystems in western North America are changing rapidly. In the mixed conifer forests of the Sierra Nevada, shifting fire regimes can be attributed to a combination of the following: the removal of indigenous land stewardship, over a century of fire suppression, and hotter, drier, conditions due to anthropogenic climate change. The removal of fire from the system led to large patches of dense, homogenous forest with high fuel loads unlike the historical forest mosaic containing a variety of gaps and openings, and these large dense patches are more likely to burn at high severity. Since many forests types in the Sierra Nevada are adapted to low severity, high frequency fire, the departure from historic fire regimes have led to shifts in forest structure and composition, which raise important questions about the future of forest ecosystems under a changing climate. In areas where fires repeatedly burn at high severity in short time frames, there is a risk of conversion to non-forested types since trees may not have time to grow to maturity and provide seed source for the next generation. In this dissertation, I investigated forest responses to shifting fire regimes as well as management interventions (namely fuel treatments) taken to mitigate high severity fire effects throughout mixed conifer forests of the Sierra Nevada. In chapter one, I examined the effectiveness of fuel treatments installed before three 2021 fires (the Caldor, Dixie, and Tamarack fires) which span multiple forest types in the central and northern Sierra Nevada. Within each fire perimeter, a variety of fuel treatment types were implemented in different combinations and at different frequencies, including thinning, masticating, piling, pile burning, and broadcast burning. I used a transect-based field protocol to collect data on four trees every 25 meters running from a untreated area to a treated area within each of the three fires. I then implemented generalized additive mixed models in a Bayesian framework to determine 1) which, if any, fuel treatment types reduced the probability of tree mortality, and 2) if tree injury surrogates (bole char, crown scorch, and crown torch) explain the mechanism of mortality. I found thinning and broadcast burning reduced tree mortality the most, and that conditioning on fire injury components such as bole char, crown scorch, and crown torch explained most of the treatment effect. In addition, I found that following piling with a pile burn reduced tree mortality more than just piling alone. Despite the challenge of using observational data to quantify the effects of non-random fuel treatments, the Bayesian modeling framework allowed me to account for multiple treatment types without reducing statistical power and can be easily implemented by other researchers investigating fuel treatment effectiveness across large spatial scales. In chapter two, I expanded upon an existing Random Forest modeling framework for postfire live tree identification using bands from Sentinel-2 imagery, which was originally created to identify live trees at one year post fire. To see if the model could distinguish any post-fire delayed tree mortality (tree mortality between 2 and 5 years postfire). I assessed the model’s ability to identify live trees at one, three, and five years postfire by training the model on NAIP imagery at those time steps. I used four sample fires in Sequoia National Forest and Kings Canyon National Park in the Southern Nevada to train the model on a total of 6000 photo interpreted NAIP pixels. I found the model could reliably predict delayed mortality at one, three, and five years post-fire with high accuracy, although delayed tree mortality was not particularly prominent across the fires. I also created RdNBR maps at each time step and thresholded them into binary live/dead using a high severity threshold cutoff to demonstrate a quick managers might currently assess tree mortality and found that the RdNBR method overestimated live trees when compared to my Random Forest model. As with most spectral-based remote sensing studies, the model could not reliably between tree recovery and shrub regeneration, suggesting that incorporation of 3D structure data could help to improve this gap. In chapter three, I sampled regeneration plots in three older fire perimeters (Freds (2004), Long (2011) and Caples (2019)) within the 2021 Caldor fire to investigate the drivers of conifer regeneration in short interval reburns. Here, short interval reburns are defined as two fires occurring within a 30 year period. Among the four sampled fires, between 64% and 88% of plots had stand replacement density (175 seedlings/ha) of seedlings regardless of age. I used generalized linear mixed models to predict seedling density as a function of reburn interval (years between fires), distance to seed source, shrub cover, severity of the first fire, and several other forest structure and topo climatic variables, separating models by pine species and shade tolerant species. Reburn interval was a strong driver of post-fire seedling density in the case of pine species but not shade tolerant species. In the case of zero year shade tolerant species, reburn interval interacted with shrub cover to influence seedling density such that increased shrub cover corresponded with slightly increased seedling density, suggesting a nurse effect. Overall, a higher shade tolerant species density compared to pine species density post-fire was reflective of the biological legacy of shade tolerant species dominance, and two fires were quick succession are not enough to overturn that legacy. The results of this study suggest that incorporation of reburn history into existing post-fire regeneration planning tools provides important context that can further aid managers seeking to identify areas with high reforestation need.
Language: English
Open Access status: green
OpenAlex ID: W7221484959
Cites: 181
Cited by:
Related to: 10