Graduation Year

2019

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Biology (Integrative Biology)

Major Professor

Gordon A. Fox, Ph.D.

Co-Major Professor

Luanna Prevost, Ph.D.

Committee Member

Earl McCoy, Ph.D.

Committee Member

Christina Richards, Ph.D.

Committee Member

Marc Lajeunesse, Ph.D.

Committee Member

Eric Menges, Ph.D.

Keywords

fire history, growth rate, mixed models, tree-rings

Abstract

Within a population, individuals frequently differ in the rate at which they grow, and this rate can be impacted by both genetic differences and abiotic factors. Often, dendrochronology is used to elucidate growth trends based on climate or other factors. This dissertation explores new statistical approaches to dendrochronological research.

First, I created a chronology for a population of longleaf pine (Pinus palustris P. Mill.) individuals in a southwest Florida sandhill community. I then used generalized linear mixed models to investigate the effects of fire frequency, year, tree age and size, and elevation on variation in radial growth heterogeneity. I then compared the chronology results to the model results.

Whereas classical dendrochronological approaches typically focus on a single signal of interest, I present an approach using linear mixed models incorporating multiple parameters that may impact growth. These models also indicated that individual tree growth variation tripled after a period of prescribed burns compared to a prior period with no prescribed fire. Growth rate variation has been shown to have important impacts on population dynamics and extinction risk; this dissertation provides evidence that fire may increase this variation. Further, the model estimates for yearly growth correlate with the chronology ring width indices, forming a statistically-based chronology while also explicitly accounting for multiple parameters at once.

Tree-ring data were collected from pines in experimental plots, each with one of four levels of prescribed fire-return interval. The fire-return intervals approximated 1-, 2-, 5-, and 7-year fire frequencies, or the plot was left unburned, for a total of five treatments. Prescribed burns were ongoing from 1976 - 2004. Tree age and basal area increment were calculated from radial tree-ring growth measurements in order to compare these factors with year and burn frequency.

In building the chronology, the trees were systematically sampled across burn plots, size, age, and elevation within the site, then detrended individually using a spline. The chronology showed more variation when trees were young, and there were some marker years consistent among plots. I built models to determine how weather (precipitation or temperature) impacts residual tree growth, and they indicated that wetter than average springs or summers had the strongest impact on growth. I created a generalized linear mixed model containing year and a term for fire, but the impact of fire was small. Overall, I was unable to clearly detect burn years by looking at the growth trace or using statistical models on the chronology growth residuals.

I examined the periods before 1976 and after 1976 in separate analyses using linear mixed-effects models. For both periods, I included burn frequency, tree size and age, and elevation as fixed effects. Individual tree and year were included in the model as random effects in order to quantify the amount of variation in these parameters. Some key results included: the relationship between diameter and elevation has varied and complicated impacts on growth rate; all levels of recent fire history impact growth rate negatively, with back-to-back burns resulting in extremely varied growth rates; individual tree core growth variance tripled within burned plots compared to unburned plots, indicating that longleaf pines exhibit some persistent heterogeneous growth when fire is incorporated into the plots. Importantly, the use of GLMMs provided flexibility to incorporate statistical sampling methods instead of targeted sampling methods, and explicitly addresses age or other factors without dismissing them as “noise”. Because other factors can be addressed, this type of approach can also answer a wider variety of questions instead of focusing on a single, overarching signal of interest.

In the period after prescribed burns began, estimated individual tree growth variances were three times larger than variance estimates for between years. This indicates that longleaf pines exhibit some individual-level, persistent heterogeneous growth when fire is incorporated into the plots, and less heterogeneous growth when fire is excluded. This leads to hypotheses regarding how fire may increase between-individual tree growth variance. Due to the heterogeneous nature of fire, each tree often experiences it differently, and growth may change accordingly. Also, fire commonly reduces competition, which could make pine growth heterogeneity more distinct by underscoring the importance of other factors, like genetics or microhabitats.

In Chapter 4, I compared the classical chronology to the statistically-based chronology created in Chapter 3. I showed that the random effect estimates for year and the ring width index from the chronology were correlated, and showed that the random effect estimates for year form a statistically-based chronology. While the model results closely resemble traditional chronological results, the model approach allows us to more explicitly describe changing tree growth due to factors like fire or water availability. Further, GLMMs provide the opportunity to measure individual variation; demographic heterogeneity has important consequences for populations and is not typically addressed when filtering out noise to produce a signal within a classical dendrochronological approach. While more exploration of these model types is in order, GLMMs are an important new tool for dendrochronologists.

Share

COinS