Hugh Ellis

The use of multi-criteria decision-making methods in the integrated assessment of climate change: implications for IA practitioners

Abstract Integrated assessment (IA) considers interactions of physical, biological, and human systems in order to assess long-term consequences of environmental and energy policies such as limits on greenhouse gas emissions, and other strategies to negate climate change. Users of IA face the daunting task of interpreting large amounts of data and uncertainties. Multi-criteria decision-making (MCDM) methods can help users process IA data, understand policy tradeoffs, and learn how their value judgments affect decisions. We held a workshop during which climate change experts tested several MCDM methods for using IA outputs to rank hypothetical policies for abating greenhouse gas emissions. Participants also evaluated several methods for visualizing tradeoffs under both certainty and uncertainty cases. This paper explores potential roles for MCDM in IA identified during the workshop, along with implications for IA design and implementation. We summarize the workshops’ results regarding intertemporal discounting (a type of MCDM weighting judgment), visualization of impacts, how MCDM methods can help users to incorporate their background knowledge, and how MCDM can improve understanding of tradeoffs and the importance of value judgments. A key result is that the interest rates IA experts recommend for discounting future impacts depend strongly on what type of impact is being discounted, as well as upon the exact phrasing of questions used to elicit rates from the experts.

STOCHASTIC DYNAMIC PROGRAMMING MODELS FOR WATER QUALITY MANAGEMENT

This paper presents optimization models for waste load allocation from multiple point sources which include both parameter (Type II) and model (Type I) uncertainty. These optimization models employ more sophisticated water quality simulation models, for example, in the case of dissolved oxygen modeling, QUAL2E and WASP4, than is typically the norm in studies on the optimization of waste load allocation. Variability in selected input parameters to the water quality simulation models gives rise to stochastic dynamic programming approaches. Two types of reliability and feasibility attributes are highlighted, associated with the management options that are generated. Several dissolved oxygen simulation models are incorporated into the optimization procedures to explore the effects of Type I uncertainty on control decisions. Information from simultaneous consideration of multiple simulation models is aggregated in the dynamic programming framework through two regret-based formulations. By accommodating both model and parameter uncertainty in the modeling framework, trade-offs can be generated between the two so as to assess their influence on control decisions. The models are applied to a waste load allocation problem for the Schuylkill River in Pennsylvania.

A joint chance-constrained programming model with row dependence

Abstract Joint chance-constrained stochastic programming models typically require random row vector independence. A joint model is developed that incorporates not only within-constraint covariance as is usually the case, but also admits dependence between constraints, that is, row dependence. The objective function of the associated chance-constrained deterministic equivalent is a multivariate normal distribution with dimension equal to the number of chance constraints in the original problem. We discuss methods to solve this multinormal integral and evaluate its derivatives. The model is implemented in portable Fortran and applied to two 9-D test problems.

Metrics matter: conflicting air quality rankings from different indices of air pollution.

Abstract Comparisons of air quality policies involve numerous considerations such as cost, health, effects on vegetation and materials, and aesthetics. Such assessments require difficult scientific and value judgments. These difficulties can also characterize comparisons that consider only physical and chemical air quality indices. We compare ambient tropospheric ozone concentrations from a baseline scenario and seven emissions scenarios for a case study. The resulting air qualities are evaluated based upon spatial and temporal distribution of impacts, exceedances of regulatory standards, concentrations weighted by population density, and a variety of averaging times. Results reveal that even when only a single pollutant is considered, comparisons of air quality can be ambiguous. Which scenario has better air quality depends on how (e.g., choice of averaging times, absolute vs. relative changes in concentrations), where (e.g., effects in specific areas vs. effects over the entire region), and when (e.g., the percent of time for which one alternative has higher concentrations than another) the comparison is made. This indicates that general descriptors of air quality such as the annual average ozone concentration do not fully describe the complexity of air quality. Use of such averages can result in different policy rankings than consideration of the full distribution of impacts.

Stochastic programming models for air quality management

Abstract Five stochastic programming models are developed for identifying cost-effective acid rain control strategies. Four of the models are based upon chance constrained formulations—the last is a two-stage programming model with simple recourse. The chance constrained models include both traditional forms in which individual constraint reliability levels are user-prescribed constants, we well as a joint chance constrained model in which reliability levels are cast as decision variables.

An Evaluation of Multicriteria Decision-Making Methods in Integrated Assessment of Climate Policy

Those who conduct integrated assessments (IAs) are increasingly aware of the need to explicitly consider uncertainty and a range of criteria when evaluating alternative policies for preventing global warming. Multi-criteria decision-making (MCDM) methods provide a useful set of tools for understanding tradeoffs and gaining insight into policy alternatives. A difficulty facing potential MCDM users is the · multitude of different techniques, each with distinct advantages and disadvantages. Methods differ widely in terms of their ease of use and appropriateness to the issue under consideration. Most importantly, different methods can yield strikingly different rankings of alternatives. A workshop was held to expose climate change experts, IA researchers, and policy makers to a range of MCDM methods and to evaluate and compare their potential usefulness to IA. Participants applied several methods in the context of a hypothetical greenhouse gas policy decision and evaluated each method. Analysis of method results and participant feedback through questionnaires and discussion provide the basis for conclusions regarding the use of MCDM methods for climate change policy and IA analysis.

Model uncertainty and model aggregation in environmental management

This paper presents dynamic programming formulations for addressing model and parameter uncertainty in environmental management problems. To address the inherent uncertainty surrounding the mathematical modelling of a physical system, we present ways to aggregate information from multiple simulation models into a dynamic programming framework. Aggregation methods are based on minimizing the extent or frequency of standard level violation, as predicted by the simulation models. Different formulations aggregate information from the multiple simulation models through extreme value, summation, risk averse and risk seeking approaches. A second basic type of uncertainty, parameter value uncertainty, is addressed by considering selected input parameters as random variables. Monte Carlo simulations are then performed to generate one-step Markov transition matrices for use in stochastic versions of the optimization models. In developing the optimization models, two types of problem feasibility are identified: nominal or first stage feasibility and secondary feasibility. Variants on the basic multiple model methodologies highlight subtleties in the definition of feasibility in multiple model cases. The methodologies are demonstrated in a water quality management example for the Schuylkill River in Pennsylvania.

Robustness in stochastic programming models

Abstract Robustness in stochastic programs is addressed through the development of a two-stage stochastic programming model with simple recourse that includes an additional robustness objective. The model is applied to a problem in acid rain control that involves the allocation of emission reductions to meet environmental quality standards in eastern North America. Tradeoffs are generated between the original objective function of the two-stage model (that minimizes expectedconstraint violation) and the second robustness objective, which represents the gradient of the two-stageobjective with respect to selected model parameters. In this application, the parameters are the second moments of transfer coefficients that relate point source SO2 emissions to wet surfate deposition at sensitive receptor locations.

Exploring the forest instead of the trees: An innovative method for defining obesogenic and obesoprotective environments

Past research has assessed the association of single community characteristics with obesity, ignoring the spatial co-occurrence of multiple community-level risk factors. We used conditional random forests (CRF), a non-parametric machine learning approach to identify the combination of community features that are most important for the prediction of obesogenic and obesoprotective environments for children. After examining 44 community characteristics, we identified 13 features of the social, food, and physical activity environment that in combination correctly classified 67% of communities as obesoprotective or obesogenic using mean BMI-z as a surrogate. Social environment characteristics emerged as most important classifiers and might provide leverage for intervention. CRF allows consideration of the neighborhood as a system of risk factors.

Nonlinear optimization of air pollution monitoring networks: algorithmic considerations and computational results

For most of the history of environmental systems analysis, the ability to formulaic large-scale environmental management models has generally outrun the ability to solve them. We refer here for the most part to nonlinear nonconvex and nondifferentiable models. The situation, however, seems to be changing rapidly. There now exists an impressive variety of general purpose nonlinear solvers and the computing hardware to make solution of such models practical. In that context then, we present a mathematical programming application involving the optimization of an air pollution monitoring network. The monitoring network model is nonlinear, nonconvex and nondifferentiable and requires in each objective function evaluation, the execution of a model to simulate pollutant concentrations in space; a routine to perform optimal interpolation (kriging) and interpolation error estimation; and adaptive quadrature procedures to integrate interpolation errors over space. The Hooke and Jeeves Discrete Step optimization alg...