Synthesizing multiple data sources to understand the population and community ecology of California trees
- Author(s): Solera, Melissa Viola Eitzel
- Advisor(s): de Valpine, Perry
- et al.
In this work, I answer timely questions regarding tree growth, tree survival, and community change in California tree species, using a variety of sophisticated statistical and remote sensing tools.
In Chapter 1, I address tree growth for a single tree species with a thorough explanation of hierarchical state-space models for forest inventory data.
Understanding tree growth as a function of tree size is important for a multitude of ecological and management applications. Determining what limits growth is of central interest, and forest inventory permanent plots are an abundant source of long-term information but are highly complex. Observation error and multiple sources of shared variation (spatial plot effects, temporal repeated measures, and a mosaic of sampling intervals) make these data challenging to use for growth estimation. I account for these complexities and incorporate potential limiting factors (tree size, competition, and resource supply) into a hierarchical state-space model. I estimate the diameter growth of white fir (Abies concolor) in the Sierra Nevada of California from forest inventory data, showing that estimating such a model is feasible in a Bayesian framework using readily available modeling tools. In this forest, white fir growth depends strongly on tree size, total plot basal area, and unexplained variation between individual trees. Plot-level resource supply variables (representing light, water, and nutrient availability) do not have a strong impact on inventory-size trees. This approach can be applied to other networks of permanent forest plots, leading to greater ecological insights on tree growth.
In Chapter 2, I expand my state-space modeling to examine survival in seven tree species, as well as investigating the results of modeling them in aggregate (at the community level) and comparing with the individual species models.
Declining tree survival is a complex, well-recognized problem, but studies have been largely limited to relatively rare old-growth forests or low-diversity systems, and to models which are species-aggregated or cannot easily accommodate yearly climate variables. I estimate survival models for a relatively diverse second-growth forest in the Sierra Nevada of California using a hierarchical state-space framework. I account for a mosaic of measurement intervals and random plot variation, and I directly include yearly stand development variables alongside climate variables and topographic proxies for nutrient limitation. My model captures the expected dependence of survival on tree size. At the community level, stand development variables account for decreasing survival trends, but species-specific models reveal a diversity of factors influencing survival. Species time trends in survival do not always conform to existing theories of Sierran forest dynamics, and size relationships with survival differ for each species. Within species, low survival is concentrated in susceptible subsets of the population and single estimates of annual survival rates do not reflect this heterogeneity in survival. Ultimately only full population dynamics integrating these results with models of recruitment can address the potential for community shifts over time.
In Chapter 3, I combine statistical modeling with remote sensing techniques to investigate whether topographic variables influence changes in woody cover.
In the North Coast of California, changes in fire management have resulted in conversion of oak woodland into coniferous forest, but the controls on this slow transition are unknown. Historical aerial imagery, in combination with Object-Based Image Analysis (OBIA), allows us to classify land cover types from the 1940s and compare these maps with recent cover. Few studies have used these maps to model drivers of cover change, partly due to two statistical challenges: 1) appropriately accounting for spatial autocorrelation (ideally without throwing away data) and 2) appropriately modeling percent cover which is bounded between 0 and 100 and not normally distributed. I study the change in woody cover in California's North Coast using historical (1948) and recent (2009) high-spatial-resolution imagery. I classify the imagery using eCognition Developer and aggregate the resulting maps to the scale of a Digital Elevation Model (DEM) in order to understand topographic drivers of woody cover change. I use Generalized Additive Models (GAMs) with a quasi-binomial probability distribution to account for spatial autocorrelation and the boundedness of the percent woody cover variable. I explore the relative roles of elevation, topographic slope, aspect (Northness/Eastness), topographic wetness index, profile curvature, historical percent woody cover, and geographical coordinates in influencing current percent woody cover. I estimate these models for scales of 20, 30, 40, 50, 60, 70, 80, 90, and 100 m, reflecting both tree neighborhood scales and stand scales. I find that historical woody cover has a consistent positive effect on current woody cover, and that the spatial term in the model is significant even after controlling for historical cover. Specific topographic variables emerge as important for different sites at different scales, but no overall pattern emerges across sites or scales for any of the topographic variables I tested. This GAM framework for modeling historical data is flexible and could be used with more variables, more flexible relationships with predictor variables, and larger scales. Modeling drivers of woody cover change from historical ecology data sources can be a valuable way to plan restoration and enhance ecological insight into landscape change.
I conclude that these techniques are promising but a framework is needed for sensitivity analysis, as modeling results can depend strongly on variable selection and model structure.