DEVELOPMENT OF A MULTI-OBJECTIVE OPTIMIZATION FRAMEWORK FOR IMPLEMENTING LOW IMPACT DEVELOPMENT SCENARIOS IN AN URBANIZING WATERSHED
Open Access
- Author:
- Zhang, Guoshun
- Graduate Program:
- Agricultural and Biological Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 23, 2009
- Committee Members:
- James Michael Hamlett, Dissertation Advisor/Co-Advisor
James Michael Hamlett, Committee Chair/Co-Chair
Albert R Jarrett, Committee Member
James Samuel Shortle, Committee Member
Hangsheng Lin, Committee Member
Lawrence Albert Fennessey, Committee Member - Keywords:
- Genetic algorithm
Stormwater
Low impact development
Optimization - Abstract:
- Stormwater management, like many other real-life problems, is always characterized with competing and sometimes conflicting objectives. Decision-makers have a great need to identify and choose the most cost-effective measures when managing stormwater runoff, and this has become particularly important in today’s economic downturn. The last decade has witnessed a transformation of stormwater management concepts from the traditional centralized approach to the low impact development (LID) approach. The LID approach, while potentially having hydrologic benefits over the traditional approach and quickly gaining popularity among the stormwater community, lags behind in consistent hydrologic assessment methods due to its relatively recent emergence. The lack of methods to quantify the LID scenario hydrologic benefits, in turn, prevents the development of optimization algorithms that can help identify cost-effective LID implementation scenarios at the watershed level. In this study a generic LID scenario optimization framework was developed to provide a consistent approach for identifying cost-effective LID implementation alternatives in post-development watersheds. The developed LID scenario optimization framework consisted of three major components: a U.S. Environment Protection Agency (USEPA) stormwater management model (SWMM) for hydrologic simulations, a representation scheme for simulating LID scenarios in SWMM, and an integration algorithm that links the LID-SWMM representation to the optimizer of ε-non-dominated sorting genetic algorithm II (ε-NSGAII). The objective of the optimization framework was to minimize the total cost and the total runoff volume from the post-development watershed, while using the pre-development peak runoff rate as a constraint. The optimization framework was then tested for applicability for the Fox Hollow Watershed (FHW) in central Pennsylvania. The U. S. Environmental Protection Agency (USEPA) Stormwater Management Model (SWMM) was used for simulating hydrologic runoff conditions from the pre-development, post-development, and LID scenarios of the case study watershed. The SWMM model was parameterized for the FHW and then calibrated and verified against two monitoring stations within the watershed. Representation schemes for the integrated management practices (IMPs) of green roof, bio-retention, and porous pavement were developed within SWMM, using existing components of flow divider, storage unit, weir, and orifice. The representation schemes for bio-retention and porous pavement were tested against the long-term monitoring data from the University of New Hampshire Stormwater Center (UNHSC). The results showed that the average peak flow reduction percentages of the SWMM representation schemes were within 10% of the UNHSC observed peak flow reduction percentages for bio-retention and porous pavement. The case study watershed of FHW in this study is a small watershed in central Pennsylvania. Averaged verification results of the SWMM model in FHW showed that the calibrated model over-predicted peak flow by 33% at Station #1 and under-predicted peak flow by 19% at Station #2. Post-development scenarios were assumed for the Mitchell Tract and Flower Garden areas within the watershed. The IMPs of green roof, bio-retention, and porous pavement were used to create an assumed LID scenario in the post-development watershed. Runoff conditions from the LID scenario were compared to those from the pre-development watershed. The results indicated that the assumed LID scenario was capable of maintaining the pre-development peak flow rate from the watershed for 1-, 2-, 10-, and 100-year 24-hour design events. The LID optimization framework was created by building the SWMM model as a subroutine into the optimizer of ε-NSGAII. The optimization framework was capable of searching through various IMP combinations and identifying the tradeoff front between total runoff volume and the total LID scenario cost. The pre-development peak flow rate was built into the optimization framework as a constraint to be in accordance with local stormwater ordinances. When implemented to the case study watershed of FHW, the developed LID optimization framework was ran for 10,000 scenario evaluations for 1-, 2-, 10-, and 100-year 24-hour design events. A total number of 27, 20, 23, and 23 near-optimal LID scenarios were identified for the four design events, respectively. The developed LID optimization framework could help stormwater associates to evaluate, compare, and optimize various LID scenarios. The LID representation scheme in the optimization framework accounted for physical processes such as infiltration, percolation, ponding, and underdrains in a LID scenario. The SWMM based representation schemes for bio-retention and porous pavement were compared to the long-term monitored data at the University of New Hampshire Stormwater Center (UNHSC), and both representations had a difference of less than 10% as compared to the observed average peak flow reduction percentages. The distributed SWMM model took into account the spatial-variability of the LID implementations (and thus the timing of the convoluted hydrographs) within a watershed. The ε-NSGAII optimizer could efficiently search through potential LID scenario designs and identify the tradeoff between total LID scenario cost and total runoff volume. The generic structure of the developed optimization framework also allowed for the accommodation of other stormwater control targets.