The public, industry and governments have become increasingly interested in green design and engineering as approaches towards better environmental quality and sustainable development. Pavement construction is one of the largest consumers of natural resources. Recycling of pavements represents an important opportunity to save the mining and use of virgin materials, conserve energy, divert materials away from landfills, and save scarce tax dollars. How much pollution, energy, natural resources, and money could be saved by using secondary materials in road construction? What are the engineering limits of using recycled materials in roads? Can we recycle over and over again pavements that contain rubber, glass, and other secondary materials? This research will quantify the environmental and economic costs and benefits of recycling asphalt pavements, and using secondary materials for their construction. The impacts will be traced through the related life-cycles and supply chains for material and energy inputs, water consumption, hazardous and non-hazardous waste generation, toxic discharges, and greenhouse gas as well as particulate matter emissions. Life-cycle environmental and economic assessment methods will be coupled with construction process models. Stakeholders will be able to use the resulting computer tool for decision-making and scenario analysis as parameters of the pavement recycling model change over time and from region to region.

The use of an indicator that is based on global climate change effects is important to support decision-makers committed to sustainable development. Our project demonstrated the use of life-cycle assessment (LCA) as a systematic approach to analyzing the construction and upgrade, operation, maintenance, and ultimate decommissioning of electric power plants. A case study of a hydroelectric power plant (Glen Canyon) was completed, including sensitivity analysis.

The LCA performed in this research quantified emissions during different phases of the life of a power plant. The emissions of greenhouse gases (GHGs) were then characterized using the global warming potential (GWP) method. GWP is a method to compare the global climate change effects of different GHGs to that of CO2. It provides a relative assessment of impacts based on selection of specific time frames. However, because the GWP compares the potential impacts of any GHG to the potency of CO2, we decided to use a different name for our indicator that assesses the potential impact of electricity generation technologies. The global warming effect (GWE) combines LCA and GWP, and was used in comparative assessments of electricity generation technologies.

In this dissertation, I explore how life-cycle assessment (LCA) results can potentially be influenced by the location of emissions, also called regionalization. The current practice in life-cycle assessment increasingly stresses the importance of being "geographically differentiated" or "regionalized." Regionalization can be important for cases that have extended supply chains distributed across different regions or for processes situated in locations with environmental conditions and population densities far from "average". A spatially resolved model for North America, called IMPACT North America, is developed here. Geographically differentiated intake fractions and characterization factors are evaluated for the North American continent for human toxicity. Special attention is given to the modeling of intake fractions and characterization factors for particulate matter. These factors can be used to adapt damage-oriented impact assessment methods. Parameters influencing intake fractions and characterization factors are identified, such as population densities and agricultural production. The importance of modeling an "urban box" and a "remote area" is also shown. The case of fly ash concrete is studied through the lens of regionalization, and critical parameters influencing the environmental trade-offs between fly ash concrete and Portland cement concrete are identified.

Through the development of the new model, along with the different applications, various questions related to spatial life-cycle assessment can be studied, and several conclusions can be drawn about what is important to consider when performing regionalization in life-cycle assessment.

I show that particulate matter dominates damage to human health and therefore the importance of devoting more detailed attention to the character of particulate matter when modeling the fate, exposure and effect of pollutants. I also demonstrate that when addressing the issue of regionalization in inventory and impact assessment, priority should be placed on the development of the archetype-based (i.e., situation-dependent) regionalization; spatialization (geography-based regionalization) should only be used secondarily and in occasional cases when results from key processes need to be expanded.

Agriculture is one of the most influential ways that humans interact with the environment. Our food system demands approximately 30% of global energy consumption, 70% of freshwater withdrawals, and 90% of consumptive water use. Roughly half of the Earth’s habitable land area is already being used in the service of agriculture. One quarter of global greenhouse gas emissions can be attributed to food production, primarily as a consequence of land use change, livestock farming, and fertilizer use. This is to say nothing of the multitude of other impacts such as conventional air and water pollution, habitat destruction, and species extinction. Several prominent trends including population growth, the expansion of the global middle class, and urbanization threaten to further strain our already deteriorating natural systems. Estimates suggest that food production must increase 70% by 2050 in order to meet demand. Attaining this target in a sustainable manner requires the acceptance of a holistic integrated engineering approach to food, energy, and water systems. It is only through such an approach that we can arrive at optimal solutions that minimize waste streams and natural resource depletion while maximizing food output. At the core of this dissertation are three interrelated research projects addressing the production and supply of fresh produce in the United States. First, we perform an environmental assessment of four high-value crops in Ventura County, California: strawberries, lemons, celery, and avocados. We calculate life-cycle energy and greenhouse gas emissions footprints and assess the impact of switching from conventional irrigation to recycled or desalinated water. Next, we expand upon the Ventura County model to include the post-harvest processing, packaging, and transportation stages. Using oranges as a case study, we estimate the carbon footprint per kilogram of fruit delivered to wholesale market in New York City, Los Angeles, Chicago, and Atlanta, and assess the relative importance of transportation mode, transportation distance, and seasonality. Finally, we apply this cradle-to-market model at a national level to assess the environmental impact of fresh tomatoes delivered to ten of the largest cities in the United States. Using linear optimization, we compute the optimal tomato distribution scheme that minimizes greenhouse gas emissions while satisfying tomato demand. This dissertation contributes to the current body of knowledge by presenting life-cycle footprints for six high-value agricultural commodities using uniquely specific regional and temporal data. We develop a holistic cradle-to-market life cycle model that integrates growing practices, water use, and embedded energy. We then apply this model in combination with linear optimization in order to mitigate the environmental impact of a popular agricultural commodity at the national level. This research underscores the importance of crop-specific and regionally-specific data collection and carbon footprinting. The adoption of a universal framework for agricultural data reporting would greatly expand the applications and accuracy of agricultural environmental assessments. Such a framework would lay the groundwork for optimal decision-making at the nexus of food, energy, and water. It would also allow for efficiency benchmarking in agricultural production and supply, and perhaps the incorporation of a performance-based ecolabel for resource-efficient crops.

Production of fuel ethanol in the United States has increased ten-fold since 1993, largely as a result of government programs motivated by goals to improve domestic energy security, economic development, and environmental impacts. Over the next decade, the growth of and eventually the total production of second generation cellulosic biofuels is projected to exceed first generation (e.g., corn-based) biofuels, which will require continued expansion of infrastructure for producing and distributing ethanol and perhaps other biofuels. In addition to identifying potential differences in tailpipe emissions from vehicles operating with ethanol-blended or ethanol-free gasoline, environmental comparison of ethanol to petroleum fuels requires a comprehensive accounting of life-cycle environmental effects. Hundreds of published studies evaluate the life-cycle emissions from biofuels and petroleum, but the operation and maintenance of storage, handling, and distribution infrastructure and equipment for fuels and fuel feedstocks had not been adequately addressed. Little attention has been paid to estimating and minimizing emissions from these complex systems, presumably because they are believed to contribute a small fraction of total emissions for petroleum and first generation biofuels.

This research aims to quantify the environmental impacts associated with the major components of fuel distribution infrastructure, and the impacts that will be introduced by expanding the parallel infrastructure needed to accommodate more biofuels in our existing systems. First, the components used in handling, storing, and transporting feedstocks and fuels are physically characterized by typical operating throughput, utilization, and lifespan. US-specific life-cycle GHG emission and water withdrawal factors are developed for each major distribution chain activity by applying a hybrid life-cycle assessment methodology to the manufacturing, construction, maintenance and operation of each component. Emissions from activities at the end of life of equipment and infrastructure are not included, as these activities have previously been shown to contribute negligibly to life-cycle emissions. Life-cycle transportation mode GHG emission factors per tonne-kilometer (t-km) are presented for long distance pipelines (5-20 g CO2-e/t-km), ocean tankers (5-17 g/t-km), fuel-carrying barges (31 g/t-km), fuel-carrying unit trains (25 g/t-km), tanker trucks (140-180 g/t-km), and bale-transporting flatbed trucks (200 g/t-km). Life-cycle emission factors are also presented per tonne of material throughput for several types of agricultural equipment (600-19,000 g CO2-e/t handled), fuel conversion facilities (9,000-98,000 g/t), fuel storage and dispensing facilities (2,000-12,000 g/t), and the portion of passenger vehicle operations dedicated to refueling errands (2,000-200,000 g/t). The emissions intensity ranges reported for specific transportation modes are largely due to the greater energy efficiency of larger vehicles and pipelines, and the emissions intensity ranges within stationary storage and handling equipment is often due to differences in utilization of capital equipment and/or material losses during storage and handling activities. Consistent with existing literature, the contribution of non-operation stages to life-cycle GHG emissions ranges from 20% to 40% for most of the components modeled. Criteria air pollutant (NOx, PM2.5, SOx, VOC, CO) emission factors are also presented for the operation stage (e.g., tailpipe only) of each transportation mode.

In order to apply the new emission factors to policy-relevant scenarios, a projection is made for the fleet inventory of infrastructure components necessary to distribute 21 billion gallons of ethanol (the 2022 federal mandate for advanced biofuels under the Energy Independence and Security Act of 2007) derived entirely from Miscanthus grass, for comparison to the baseline petroleum system. Due to geographic, physical and chemical properties of biomass and alcohols, the distribution system for Miscanthus-based ethanol is more capital- and energy-intensive than petroleum per unit of fuel energy delivered. Assuming steady-state annual turnover, operation, and maintenance of infrastructure to supply the projected quantities of ethanol and petroleum fuels, ethanol is estimated to be approximately five times more GHG and water intensive than petroleum (i.e., GHG emissions of more than 17 g CO2-e/MJ versus 3 g/MJ, and water withdrawals of 380 L/MJ vs. 77 L/MJ of consumed fuel, neglecting feedstock production and conversion). Embodied GHG emissions from manufacturing and maintaining infrastructure, equipment, and vehicles make up less than half of these emissions, at approximately 1 g CO2-e/MJ of petroleum fuel and 8 g CO2-e/MJ of ethanol. Although petroleum fuels are projected to supply twenty times the energy content of ethanol in 2022, the annual GHG and water withdrawal footprint of petroleum's liquid fuel infrastructure and distribution system is slightly less than four times that of ethanol (i.e., 110 vs. 30 million tonnes of CO2-e and 2,500 vs. 640 billion liters of water). Opportunities to significantly reduce emissions include shifting transportation to more efficient modes, consuming products closer to producers, and converting biorefineries to produce fuel with higher energy density than ethanol. Minimizing fuel transportation distance is believed to be the most feasible and cost-effective opportunity to reduce emissions in the near term.

The transportation of biofuels away from producer regions poses environmental, health, and economic trade-offs that are herein evaluated using a simplified national distribution network model. In just the last ten years, ethanol transportation within the contiguous United States is estimated to have increased more than ten-fold in total t-km as ethanol has increasingly been transported away from Midwest producers due to air quality regulations pertaining to gasoline, renewable fuel mandates, and the 10% blending limit (i.e., the E10 blend wall). From 2004 to 2009, approximately 10 billion t-km of ethanol transportation are estimated to have taken place annually for reasons other than the E10 blend wall, leading to annual freight costs greater than $240 million and more than 300,000 tonnes of CO2-e emissions and significant emissions of criteria air pollutants from the combustion of more than 90 million liters of diesel. Although emissions from distribution activities are small when normalized to each unit of fuel, they are large in scale.

Archetypal fuel distribution routes by rail and by truck are created to evaluate the significance of mode choice and route location on the severity of public health impacts from locomotive and truck emissions, by calculating the average PM2.5 pollution intake fraction along each route. Exposure to pollution resulting from trucking is found to be approximately twice as harmful as rail (while trucking is five times more energy intensive). Transporting fuel from the Midwest to California would result in slightly lower human health impacts than transportation to New Jersey, even though California is more than 50% farther from the Midwest than most coastal Northeast states.

In summary, this dissertation integrated concepts from infrastructure management, climate and renewable fuel policy, fuel chemistry and combustion science, air pollution modeling, public health impact assessment, network optimization and geospatial analysis. In identifying and quantifying opportunities to minimize damage to the global climate and regional air quality from fuel distribution, results in this dissertation provide credence to the urgency of harmonizing policies and programs that address national and global energy and environmental goals. Under optimal future policy and economic conditions, infrastructure will be highly utilized and transportation minimized in order to reduce total economic, health, and environmental burdens associated with the entire supply and distribution chain for transportation fuels.

Both correlation and mutual information (MI) are common co-expression measures. MI has a major advantage to measure non-linear relationships. However, it is not clear how much MI adds beyond standard (robust) correlation measures or regression model based association measures. We provided a comprehensive comparison between mutual information and correlation in 8 empirical data sets and in simulations. We confirmed close relationships between MI and correlation in all data sets. The biweight midcorrelation, a robust form of correlation, outperformed MI in terms of elucidating gene pairwise relationships. Coupled with the topological overlap matrix transformation, it often led to modules superior to MI and maximal information coefficient (MIC) in terms of gene ontology enrichment. In addition, we proposed the use of polynomial or spline regression models as an alternative to MI for capturing non-linear relationships between quantitative variables. Overall, our results indicated that MI networks could be safely replaced by correlation networks for stationary co-expression data.

Sample classification, especially disease status prediction, is an important area of investigation for gene expression studies. Many machine learning methods, i.e. predictors, have been developed to tackle this problem. We proposed a novel bootstrap aggregated (bagged) GLM predictor randomGLM (RGLM) that shares superior accuracy and good interpretability. RGLM incorporates several elements of randomness and instability, such as random subspace method, optional interaction terms and forward feature selection. The prediction performances of various predictors were evaluated on hundreds of genomic data sets, the UCI machine learning benchmark data and simulations. RGLM often outperformed alternative methods including random forests and penalized regression models (ridge regression, elastic net, lasso) in both binary and continuous outcome predictions. Further, RGLM provides variable importance measures that can be used to define a ``thinned" ensemble predictor (involving few features) retaining excellent predictive accuracy.

RGLM has won the 2012 COPD Improver Challenge, in which we aimed to predict the chronic obstructive pulmonary disease (COPD) status based on gene expression data. We outlined how RGLM compared with random forest on the COPD data set, and discussed potential reasons for the superior performance of RGLM in this sub-challenge.

Personal computers (PCs) are one of the most ubiquitous and indispensable electronic devices in use within California’s homes and businesses. More PCs are estimated to be in use in California than in any other U.S. state. According to the latest published estimates from U.S. Department of Energy (DOE), 11.5 million PCs were installed in California’s homes (as of 2001) and 6.5 million PCs were installed in California’s commercial buildings (as of 2003) (U.S. DOE 2003, 2006a). Based on these DOE data and recent PC sales data, it was estimated in this research project that as of 2005 nearly 19 million PCs were installed in California homes and businesses and that this number is expected to grow significantly through 2012.

This research project focused on evaluating a set of realistic measures aimed at reducing the life-cycle energy use and GHG emissions associated with operating and maintaining California’s residential and commercial PC stock (hereafter referred to simply as “California’s PC stock”). The term “maintain” is used here to describe the ongoing process of replacing, upgrading, and discarding obsolete PCs within California’s PC stock as necessary. The project employed a life-cycle assessment (LCA) approach to characterize both direct and indirect energy use and GHG emissions. Direct energy use and GHG emissions were defined as the energy use and GHG emissions attributable to the operational electricity consumption of California’s PC stock. Indirect energy use and GHG emissions were defined as the energy use and GHG emissions attributable to activities that support ongoing PC stock maintenance, namely, the manufacture of new PCs and PC components as needed and the end-of-life treatment (i.e., waste disposal and recycling) of obsolete PCs and PC components.

In this research project, the potential reductions in life-cycle energy use and GHG emissions associated with the measures considered were projected through the year 2012. The purpose of this work was to characterize the effectiveness of these measures to inform PC-related policies for near-term energy efficiency improvements and GHG emission reductions in the State of California. Baseline projections were first established for the life-cycle energy use and GHG emissions associated with operating and maintaining California’s PC stock from 2005 to 2012. These baseline projections were made using best-available data from public sources for all key modeling parameters. Whenever possible, ranges for key modeling parameters were established to account for published data variations. Two future scenarios were considered: a low energy use and GHG emissions scenario (“low scenario”) and a high energy use and GHG emissions scenario (“high scenario”). The purpose of the two-scenario approach was to establish preliminary upper and lower bounds on the results based on the feasible data ranges identified from the literature for key modeling parameters. The potential reductions in life-cycle energy use and GHG emissions associated with the measures considered were then projected for each scenario over the same time period.

Commercial PC estimate was derived by: (1) assuming that 8.6 million PCs were installed in commercial buildings in the Pacific Census Region in 2003, based on the U.S. DOE 2003 Commercial Building Energy Use Survey (CBECS) (U.S. DOE 2006), and (2) assuming that 75% of Pacific Census Region commercial sector employment (as defined by the 2003 CBECS) and hence PC usage occurs in California based on employment data from the U.S. Census Bureau (U.S. Census Bureau 2006).