Search

Search Results

• 1. [Article] Exploration of statistical methods for synthesizing the effects of variable-retention harvesting on multiple taxa

  Variable-retention harvesting was proposed to reduce loss of biodiversity and ecosystem processes associated with late-seral Douglas-fir (Pseudotsuga menziesii) forests in the Pacific Northwest. The Demonstration ...

  Citation × Citation Citation Abstract Copy Title: Exploration of statistical methods for synthesizing the effects of variable-retention harvesting on multiple taxa Author: Lam, Tzeng Yih Variable-retention harvesting was proposed to reduce loss of biodiversity and ecosystem processes associated with late-seral Douglas-fir (Pseudotsuga menziesii) forests in the Pacific Northwest. The Demonstration of Ecosystem Management Options experiment was established to test this hypothesis. Analysis presents various challenges to drawing statistical inferences about treatment effects. This dissertation explored novel statistical methods for understanding the response of multiple forest taxa to variable-retention harvesting. Excessive zero counts are common among terrestrial small mammal species that are captured infrequently. Zero-inflated and hurdle models are appealing tools for analyzing these data. A simulation was performed to understand the properties and robustness of these models. When true mean abundance was low, the estimated parameters from these models were highly unstable. Goodness of fit criteria could not discern among the processes generating the data. The Poisson and negative binomial Generalized Linear Models (GLMs) were fitted to four small mammal species with different rates of capture. Predictors included several variables representing vegetation structure. These models and overdispersed Poisson were then specified as Generalized Linear Mixed Models (GLMMs) to account for nesting and blocking in the experimental design. The fitted GLMs indicated that predictors were not consistent among models for the infrequently captured species. Differences in estimated coefficients between GLMs and GLMMs were noticeable. The overdispersed Poisson GLMM was suggested to be most suitable. Structural Equation Modeling (SEM) is suitable for modeling interactions of many cause-and-effect relationships in forest ecosystems. SEM was applied to understand overstory-understory relationships of late-seral herb species under mature forest conditions and immediately after variable-retention harvesting. In undisturbed forests, light attenuation, belowground competition and stand age were the primary drivers of late-seral herb cover. After variable-retention harvesting, microclimatic stresses were inferred to primarily affect late-seral species diversity and composition. Logging debris had little discernible effect on the change in the late-seral herb community. The explored statistical models complement conventional methods for studying the effects of variable-retention harvesting. These models address distributional issues of response data and provide further insight into the complex processes driving managed forest ecosystems. Future analyses should apply a suite of statistical models to gain different perspectives. Title: Exploration of statistical methods for synthesizing the effects of variable-retention harvesting on multiple taxa Author: Lam, Tzeng Yih Variable-retention harvesting was proposed to reduce loss of biodiversity and ecosystem processes associated with late-seral Douglas-fir (Pseudotsuga menziesii) forests in the Pacific Northwest. The Demonstration of Ecosystem Management Options experiment was established to test this hypothesis. Analysis presents various challenges to drawing statistical inferences about treatment effects. This dissertation explored novel statistical methods for understanding the response of multiple forest taxa to variable-retention harvesting. Excessive zero counts are common among terrestrial small mammal species that are captured infrequently. Zero-inflated and hurdle models are appealing tools for analyzing these data. A simulation was performed to understand the properties and robustness of these models. When true mean abundance was low, the estimated parameters from these models were highly unstable. Goodness of fit criteria could not discern among the processes generating the data. The Poisson and negative binomial Generalized Linear Models (GLMs) were fitted to four small mammal species with different rates of capture. Predictors included several variables representing vegetation structure. These models and overdispersed Poisson were then specified as Generalized Linear Mixed Models (GLMMs) to account for nesting and blocking in the experimental design. The fitted GLMs indicated that predictors were not consistent among models for the infrequently captured species. Differences in estimated coefficients between GLMs and GLMMs were noticeable. The overdispersed Poisson GLMM was suggested to be most suitable. Structural Equation Modeling (SEM) is suitable for modeling interactions of many cause-and-effect relationships in forest ecosystems. SEM was applied to understand overstory-understory relationships of late-seral herb species under mature forest conditions and immediately after variable-retention harvesting. In undisturbed forests, light attenuation, belowground competition and stand age were the primary drivers of late-seral herb cover. After variable-retention harvesting, microclimatic stresses were inferred to primarily affect late-seral species diversity and composition. Logging debris had little discernible effect on the change in the late-seral herb community. The explored statistical models complement conventional methods for studying the effects of variable-retention harvesting. These models address distributional issues of response data and provide further insight into the complex processes driving managed forest ecosystems. Future analyses should apply a suite of statistical models to gain different perspectives. Close