UVM Theses and Dissertations
Format:
Print
Author:
Porter, Andrea J.
Dept./Program:
Civil and Environmental Engineering
Year:
2007
Degree:
PhD
Abstract:
Informatics, the automated use of data, can be applied to the field of environmental engineering in order to focus on the application of advanced computing and information technologies. Environmental engineering informatics offers an alternative approach to traditional modeling. Advances in informatics have resulted in the development of a number of powerful problem-solving paradigms that can be used, either separately or synergistically, to address environmental engineering problems not easily solved using process modeling or empirical methods. The applicability of informatics for a wide range of environmental problems is expected based upon similarities across the environmental discipline.
The nature of many environmental problems necessitates insight into the dynamic relationships of systems while simultaneously accounting for missing, uncertain, and spatially-linked data sources. These common features can be described as: (1) complexity and ill-defined relationships between the key components; (2) fragmented, unavailable, incomplete, or incompatible data sources; (3) ambiguity, uncertainty, or fuzziness in data sets and relationships; and (4) geographic or spatial nature of the data needed to solve the problem. These four challenges and the implications for addressing a case study on plant-assisted bioremediation, a type of phytoremediation, were explored.
This study developed the framework for a novel decision support system for plantassisted bioremediation design. Additionally, two components of the system were constructed and tested using data available in the literature. First, geographic information systems (GIs) technology was combined with fuzzy logic to construct a phytoremediation plant selection tool. The focus was on one particular phytoremediation mechanism, plant-assisted bioremediation of polycyclic aromatic hydrocarbon (PAH) soil contamination. Many plant species show potential for PAH remediation, but factors such as growth requirements, climate, and soil conditions need to be considered.
Given the spatial nature of the data involved, GIs was chosen as the basis for the plant selection tool. Eight candidate plant species were selected and their growth requirements were represented using fuzzy membership functions to describe the parameters' uncertainty. Vermont and South Carolina were selected as case studies, and calculations were performed to determine the suitability ratings for each species based on the plants' growth requirements with respect to climate and soil attributes.
Second, case-based reasoning (CBR) was implemented on laboratory data sets to predict the ratio of current contaminant concentration to initial concentration (CICo). CBR is an informatics method based on solving new problems by recalling solutions to similar problems and adapting accordingly. The integral feature of a CBR methodology is a database of cases that serves as the machine's memory, but the methods by which a solution works need not be explicitly known. Because of this, CBR is particularly appealing for environmental problems with mechanisms that are not well understood, such as phytoremediation. The optimized CBR system achieved good predictive accuracy (r² = 0.94). Additionally, although predictive accuracy decreased slightly compared to that for the training set, the CBR system made good predictions for CICo even when the system was presented with cases that had not been used for the optimization and training (r² = 0.89).
The nature of many environmental problems necessitates insight into the dynamic relationships of systems while simultaneously accounting for missing, uncertain, and spatially-linked data sources. These common features can be described as: (1) complexity and ill-defined relationships between the key components; (2) fragmented, unavailable, incomplete, or incompatible data sources; (3) ambiguity, uncertainty, or fuzziness in data sets and relationships; and (4) geographic or spatial nature of the data needed to solve the problem. These four challenges and the implications for addressing a case study on plant-assisted bioremediation, a type of phytoremediation, were explored.
This study developed the framework for a novel decision support system for plantassisted bioremediation design. Additionally, two components of the system were constructed and tested using data available in the literature. First, geographic information systems (GIs) technology was combined with fuzzy logic to construct a phytoremediation plant selection tool. The focus was on one particular phytoremediation mechanism, plant-assisted bioremediation of polycyclic aromatic hydrocarbon (PAH) soil contamination. Many plant species show potential for PAH remediation, but factors such as growth requirements, climate, and soil conditions need to be considered.
Given the spatial nature of the data involved, GIs was chosen as the basis for the plant selection tool. Eight candidate plant species were selected and their growth requirements were represented using fuzzy membership functions to describe the parameters' uncertainty. Vermont and South Carolina were selected as case studies, and calculations were performed to determine the suitability ratings for each species based on the plants' growth requirements with respect to climate and soil attributes.
Second, case-based reasoning (CBR) was implemented on laboratory data sets to predict the ratio of current contaminant concentration to initial concentration (CICo). CBR is an informatics method based on solving new problems by recalling solutions to similar problems and adapting accordingly. The integral feature of a CBR methodology is a database of cases that serves as the machine's memory, but the methods by which a solution works need not be explicitly known. Because of this, CBR is particularly appealing for environmental problems with mechanisms that are not well understood, such as phytoremediation. The optimized CBR system achieved good predictive accuracy (r² = 0.94). Additionally, although predictive accuracy decreased slightly compared to that for the training set, the CBR system made good predictions for CICo even when the system was presented with cases that had not been used for the optimization and training (r² = 0.89).