Towards greater accuracy in individual-tree mortality regression

Background mortality is an essential component of any forest growth and yield model. Forecasts of mortality contribute largely to the variability and accuracy of model predictions at the tree, stand and forest level. In the present study, I implement and evaluate state-of-the-art techniques to increase the accuracy of individual tree mortality models, similar to those used in many of the current variants of the Forest Vegetation Simulator, using data from North Idaho and Montana. The first technique addresses methods to correct for bias induced by measurement error typically present in competition variables. The second implements survival regression and evaluates its performance against the traditional logistic regression approach. I selected the regression calibration (RC) algorithm as a good candidate for addressing the measurement error problem. Two logistic regression models for each species were fitted, one ignoring the measurement error, which is the “naïve” approach, and the other applying RC. The models fitted with RC outperformed the naïve models in terms of discrimination when the competition variable was found to be statistically significant. The effect of RC was more obvious where measurement error variance was large and for more shade-intolerant species. The process of model fitting and variable selection revealed that past emphasis on DBH as a predictor variable for mortality, while producing models with strong metrics of fit, may make models less generalizable. The evaluation of the error variance estimator developed by Stage and Wykoff (1998), and core to the implementation of RC, in different spatial patterns and diameter distributions, revealed that the Stage and Wykoff estimate notably overestimated the true variance in all simulated stands, but those that are clustered. Results show a systematic bias even when all the assumptions made by the authors are guaranteed. I argue that this is the result of the Poisson-based estimate ignoring the overlapping area of potential plots around a tree. Effects, especially in the application phase, of the variance estimate justify suggested future efforts of improving the accuracy of the variance estimate. The second technique implemented and evaluated is a survival regression model that accounts for the time dependent nature of variables, such as diameter and competition variables, and the interval-censored nature of data collected from remeasured plots. The performance of the model is compared with the traditional logistic regression model as a tool to predict individual tree mortality. Validation of both approaches shows that the survival regression approach discriminates better between dead and alive trees for all species. In conclusion, I showed that the proposed techniques do increase the accuracy of individual tree mortality models, and are a promising first step towards the next generation of background mortality models. I have also identified the next steps to undertake in order to advance mortality models further.

Bayesian change-point analysis in linear regression model with scale mixtures of normal distributions

In this thesis, we consider Bayesian inference on the detection of variance change-point models with scale mixtures of normal (for short SMN) distributions. This class of distributions is symmetric and thick-tailed and includes as special cases: Gaussian, Student-t, contaminated normal, and slash distributions. The proposed models provide greater flexibility to analyze a lot of practical data, which often show heavy-tail and may not satisfy the normal assumption. As to the Bayesian analysis, we specify some prior distributions for the unknown parameters in the variance change-point models with the SMN distributions. Due to the complexity of the joint posterior distribution, we propose an efficient Gibbs-type with Metropolis- Hastings sampling algorithm for posterior Bayesian inference. Thereafter, following the idea of [1], we consider the problems of the single and multiple change-point detections. The performance of the proposed procedures is illustrated and analyzed by simulation studies. A real application to the closing price data of U.S. stock market has been analyzed for illustrative purposes.

SAMPLING BIAS IN EVALUATING THE PROBABILITY OF SEISMICALLY INDUCED SOIL LIQUEFACTION WITH SPT & CPT CASE HISTORIES

Several deterministic and probabilistic methods are used to evaluate the probability of seismically induced liquefaction of a soil. The probabilistic models usually possess some uncertainty in that model and uncertainties in the parameters used to develop that model. These model uncertainties vary from one statistical model to another. Most of the model uncertainties are epistemic, and can be addressed through appropriate knowledge of the statistical model. One such epistemic model uncertainty in evaluating liquefaction potential using a probabilistic model such as logistic regression is sampling bias. Sampling bias is the difference between the class distribution in the sample used for developing the statistical model and the true population distribution of liquefaction and non-liquefaction instances. Recent studies have shown that sampling bias can significantly affect the predicted probability using a statistical model. To address this epistemic uncertainty, a new approach was developed for evaluating the probability of seismically-induced soil liquefaction, in which a logistic regression model in combination with Hosmer-Lemeshow statistic was used. This approach was used to estimate the population (true) distribution of liquefaction to non-liquefaction instances of standard penetration test (SPT) and cone penetration test (CPT) based most updated case histories. Apart from this, other model uncertainties such as distribution of explanatory variables and significance of explanatory variables were also addressed using KS test and Wald statistic respectively. Moreover, based on estimated population distribution, logistic regression equations were proposed to calculate the probability of liquefaction for both SPT and CPT based case history. Additionally, the proposed probability curves were compared with existing probability curves based on SPT and CPT case histories.

Use of hydroclimatic forecasts for improved water management in central Texas

Accurate seasonal to interannual streamflow forecasts based on climate information are critical for optimal management and operation of water resources systems. Considering most water supply systems are multipurpose, operating these systems to meet increasing demand under the growing stresses of climate variability and climate change, population and economic growth, and environmental concerns could be very challenging. This study was to investigate improvement in water resources systems management through the use of seasonal climate forecasts. Hydrological persistence (streamflow and precipitation) and large-scale recurrent oceanic-atmospheric patterns such as the El Niño/Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), North Atlantic Oscillation (NAO), the Atlantic Multidecadal Oscillation (AMO), the Pacific North American (PNA), and customized sea surface temperature (SST) indices were investigated for their potential to improve streamflow forecast accuracy and increase forecast lead-time in a river basin in central Texas. First, an ordinal polytomous logistic regression approach is proposed as a means of incorporating multiple predictor variables into a probabilistic forecast model. Forecast performance is assessed through a cross-validation procedure, using distributions-oriented metrics, and implications for decision making are discussed. Results indicate that, of the predictors evaluated, only hydrologic persistence and Pacific Ocean sea surface temperature patterns associated with ENSO and PDO provide forecasts which are statistically better than climatology. Secondly, a class of data mining techniques, known as tree-structured models, is investigated to address the nonlinear dynamics of climate teleconnections and screen promising probabilistic streamflow forecast models for river-reservoir systems. Results show that the tree-structured models can effectively capture the nonlinear features hidden in the data. Skill scores of probabilistic forecasts generated by both classification trees and logistic regression trees indicate that seasonal inflows throughout the system can be predicted with sufficient accuracy to improve water management, especially in the winter and spring seasons in central Texas. Lastly, a simplified two-stage stochastic economic-optimization model was proposed to investigate improvement in water use efficiency and the potential value of using seasonal forecasts, under the assumption of optimal decision making under uncertainty. Model results demonstrate that incorporating the probabilistic inflow forecasts into the optimization model can provide a significant improvement in seasonal water contract benefits over climatology, with lower average deficits (increased reliability) for a given average contract amount, or improved mean contract benefits for a given level of reliability compared to climatology. The results also illustrate the trade-off between the expected contract amount and reliability, i.e., larger contracts can be signed at greater risk.

PROBABILISTIC MODELING OF RAINFALL INDUCED LANDSLIDE HAZARD ASSESSMENT IN SAN JUAN LA LAGUNA, SOLOLÁ, GUATEMALA

The municipality of San Juan La Laguna, Guatemala is home to approximately 5,200 people and located on the western side of the Lake Atitlán caldera. Steep slopes surround all but the eastern side of San Juan. The Lake Atitlán watershed is susceptible to many natural hazards, but most predictable are the landslides that can occur annually with each rainy season, especially during high-intensity events. Hurricane Stan hit Guatemala in October 2005; the resulting flooding and landslides devastated the Atitlán region. Locations of landslide and non-landslide points were obtained from field observations and orthophotos taken following Hurricane Stan. This study used data from multiple attributes, at every landslide and non-landslide point, and applied different multivariate analyses to optimize a model for landslides prediction during high-intensity precipitation events like Hurricane Stan. The attributes considered in this study are: geology, geomorphology, distance to faults and streams, land use, slope, aspect, curvature, plan curvature, profile curvature and topographic wetness index. The attributes were pre-evaluated for their ability to predict landslides using four different attribute evaluators, all available in the open source data mining software Weka: filtered subset, information gain, gain ratio and chi-squared. Three multivariate algorithms (decision tree J48, logistic regression and BayesNet) were optimized for landslide prediction using different attributes. The following statistical parameters were used to evaluate model accuracy: precision, recall, F measure and area under the receiver operating characteristic (ROC) curve. The algorithm BayesNet yielded the most accurate model and was used to build a probability map of landslide initiation points. The probability map developed in this study was also compared to the results of a bivariate landslide susceptibility analysis conducted for the watershed, encompassing Lake Atitlán and San Juan. Landslides from Tropical Storm Agatha 2010 were used to independently validate this study’s multivariate model and the bivariate model. The ultimate aim of this study is to share the methodology and results with municipal contacts from the author's time as a U.S. Peace Corps volunteer, to facilitate more effective future landslide hazard planning and mitigation.