Assignment #7

Perez-Pedini C, Limbrunner JF, Vogel RM (2005) “Optimal location of infiltration-based best management practices for storm water management,” JOURNAL OF WATER RESOURCES PLANNING AND MANAGEMENT, 131(6) pp. 441-448

Summary:

The authors of this article utilized computer modeling and genetic algorithms to attempt to determine ways to locate Best Management Practices (BMPs) within a watershed to most effectively reduce overall peak flows at the basin outlet. The BMPs are considered that cause greater infiltration at certain points within the watershed (and thus reducing the flow available for peak flow). They are meant to be an extension of traditional detention basins in the watershed. The analysis process consisted of first developing a watershed model, calibrating the model (through various calibration parameters) for a specific watershed based upon actual storms that occurred, application of a genetic algorithm to determine which areas in the watershed could implement a BMP that could best reduce the peak runoff (based upon a specified number of BMPs to implement in the system), and then further analysis on the results to determine managment strategies and statistical similarities between inputs and response.

In the model formulation, the authors utilized a system where the watershed was broken into hydrologic response units (HRUs) that are sized as 120m squares aligned as a grid throughout the watershed. Each HRU receives the runoff, allows for initial infiltration into the soil, allows for transport of the available runoff to one downstream HRU (based on the greatest elevation change), has groundwater storage, and allows for the transfer of groundwater via underground transport to adjacent HRUs. Additionally, streams are located throughout the watershed such that the runoff entering an HRU designated as a stream location is collected and moved to the outlet of the basin as a faster rate. After applying the mathematical model, they imported a watershed to Excel via ArcGIS and then calibrated the various model parameters to meet the outflow from specific storms.

The genetic algorithm programming approach was applied for the system based upon some limitations. The watershed had over 4,000 HRUs, and each potentially could have a BMP in place; however, allowing all of these to be availabile as decision variables resulted in a model that had too large of a decision space for the GA too be applied effectively, and thus, they limited the potential BMP locations to 1) the most impervious (highest Curve Number) and 2) the closest HRUs to the streams (as they have the most effect on the initial peak flow at the basin outlet). They ran the simulations by assuming a certain number of BMPs would be in place (and then determining which locations would result in the minimum outflow runoff). They eventually found that after implementing about 25 to 100 BMPs, the rate of flow loss for adding additional BMPs was decreasing.

Further analysis on the system was performed to determine some information about management practices. The central question at this stage was to determine if the watershed should be planned on a master planning level such that initial projects be completed that may not have the greatest impact on the overall outlet flow (if implementing another combination of BMPs in the future would then result in a global minimum being able to be found later on). The results showed that this was not necessary—the best solution would be to choose the most effective BMPs initially and then not worry about which locations would be best in the future. Additionally, the authors were unable to find statisical relationships between relevant parameters (such as location upstream from outlet, etc.) that would provide an optimum solution; thus, the authors demonstrated that a GA application is really necessary for this application.

Discussion:

I thought the authors did a very good job of explaining their methodology. Overall, I think they used a sound methodology to determine a very complex question. I think that the following points are relevant for discussion:
1) In formulating the model, the authors ignored the type of BMP and just assumed that having a BMP would reduce the curve number by 5. I'm not sure if this is really a good assumption as different locations could naturally result in a better potential for different BMPs to have greater or lesser effects. What if a BMP just could not be put in place in one of the "optimum" locations--particularly if only a few BMPs (25 or so) would be put in place.
2) Is the model developed physically accurate for the system? The ability of the calibration to meet the example storm appears to indicate that the model can work, but I still wonder about some of the assumptions made in the formulation process--sometimes this is just hard to follow.
3) How valid is the conclusion that the master plan does not need to be considered? From their writeup and model, it seems valid to me; however, question 1 seems to come into play. Given some of their assumptions, if they were made differently, would the respone prove different and thus nullify their results?
4) Are the parameters used to limit the potential number of sites for BMPs valid? They basically reduced the amount of potential BMP sites in half. Logically, the parameters they used seem to make sense, but you could wonder if this is necessarily appropriate.

Assignment #6

Behera, P, Papa, F., Adams, B (1999) “Optimization of Regional Storm-Water Management Systems” Journal of Water Resources Planning and Management, 125(2) pp. 107-114

Summary:
In this study, the authors developed a methodology (using dynamic programming) to optimize the design and configuration of a storm-water management system for single-storm events in land development applications, primarily with the use of detention basins that regulate flows. Overall, the objective of the model is to determine a way to minimize the cost associated with the development of the storm-water management plan while still meeting environmental constraints such as peak discharge requirements and water quality parameters. The model is applied on an entire watershed basis such that the optimal locations and configurations of many detention basins can be applied such that the overall discharge from the watershed meets the requirements. Decision variables in the model include the active volume of water contained in each detention basin, the volumetric release rate from the pond, and the depth of the pond. The objective function is applied with associated costs for construction of the basin and land area taken by the pond (which relate to the decision variables). Finally, the system constraints were applied such that the water quality/quantity parameters could be met.

The initial simulation, the modeling considered only a single catchment of the watershed with one detention basin (and meeting optimum water quality/quantity parameters). Subsequent to this operation, the authors considered overall basin configurations with multiple basins (in this case a three catchment configuration). For this application, they added quality and quantity constraints such that the total outflow from all basins would meet requirements. In this application, they utilized a dynamic programming model that would allow for them to determine the overall minimum cost by considering each basin (stage) separately in decision making.

Discussion:
Overall, I thought the approach of the authors seemed to be valid for the system. It was interesting to see a direct application for linear programming, and reading the article can help me think of other applicable situations, such as river water quality modeling for point sources (determining where to force waste generators to meet higher levels of treatment).

Assignment #5

Berry, J., Fleisher, L, Hart, W. Phillips, C. and Watson, J.-P. (2005) “Sensor Placement in Municipal Water Networks”, Journal of Water Resources Planning and Management, 131(3) pp. 237-243

Summary:

This article represents a more recent approach to similar problems that were discussed in last week's blog concerning optimal sensor placement in water distribution networks. In this case, the primary model concern is to minimize the population at risk of exposure to contamination from a single event (instantaneous deposit of contaminant at one source node in the system--which begs the question of which location to consider for attack). The model considers exposure to occur if a flow could reach a node from the distribution network without having a sensor upstream of the given flow path. The model also ignores time considerations such that the flow throughout the system has occurred for sufficient time to generate a steady state. The model also does not consider health impacts or the concentrations of exposure.

The model utilizes EPANET to determine the flow characteristics throughout the system. Additionally, it considers the potential amount of population present at each node (which varies based upon the flow situation as people are at home or in the workplace). For the attack location, the models assumed a synthetic distribution (actual attack would occur based upon, for example, terrorist desires or ease of contamination at the specific location). The actual IP model seeks to minimize exposure through looking at the product of the probability of attack at a node, the probability of attack at a node, and a parameter indicating whether the node is contaminated from an attack at any other given node (including its own) for the flow characteristics being studied. The decision variables for the system are the placement of a sensor at a given edge (pipe), and the parameter fits into the optimization line by its relation to the third part of the objective funtion ("parameter indicating whether the node is contaminated from an attack at any other given node")---the relationship is expressed as a constraint of the system.

Ultimately, the authors utilized the simulation on three set cases (two from EPANET examples and one from an actual test city).

Discussion:
Overall, I enjoyed this article. It seemed to be very well written (most of the time I had questions or concerns about their methods they detailed their logical process and reasoning). My primary concern at this point is likely their assumption that once a sensor is ahead or downstream of a node that the population at the node would no longer be at risk of attack. In actuality, there would likely be a time lag (they conveniently assumed time constraints away), so people would still be exposed. I think some of Dr. Zechman's research (per seminar last fall) has gone into attempting to address some of these issues.

Assignment #4

Lee, B. H. and Deininger, R. A. (1992) “Optimal Locations of Monitoring Stations in Water Distribution Systems”, Journal of Environmental Engineering, 118(1) pp. 4-16

Summary:
This article is meant to primarily address a potential method for determining sampling locations within distribution systems that will most efficiently ensure that the system meets water quality criteria. The need for this method was generated by SDWA requirements for sampling water within water distribution networks. As the system is made of various pipes with nodes and relevant demands (requirements for outflows at a node), many potential sampling locations are possible (at the nodes). At some nodes, water unused by the demands flows to other nodes, and it can be assumed that the water quality at nearby nodes is related. Thus, this system seeks to distinguish which nodes could be sampled such that the optimum amount of water within the system could be sampled (as much as possible) compared to the number of sampling locations (as few as possible).

To accomplish this goal, the analysts utilized linear programming with a binary system in the actual analysis; however, considerable effort went into developing their mathematical model to input into the LP program. For the mathematical model, they first analyzed the water distribution system based upon a set of demand patterns to determine which nodes related to adjacent nodes with respect to water quality. Then, a binary matrix was created such that if a sufficient percentage of the water flowing into a node would have similar water quality compared to another node, the matrix value would be 1 (this would then be completed for all nodes vs. all nodes). This could then be used to ensure that all nodes would contribute enough to the sampling such that the maximum amount of water would be tested.

Finally, the authors presented some case studies where they applied their method to various actual distribution systems, eventually trying to determine locations that could be optimized for variable demand patterns that would likely be seen in the system.

Discussion:
I initially thought the authors had found a very neat solution method to tackle a complex problem. In many ways, i still think that; however, I can also see that the model did make a couple of assumptions that may make the analysis less sure than they thought. One major thing that I would try to do in their analysis would be to determine if a means of looking at the time from the inflow into the system could be reached. In their process of trying to account for this aspect, they used a 0.5 assumption variable that seemed somewhat arbitrary. My basic question would probably wonder how their constraints worked in the model to prevent just sampling at the first location...all of the water goes through that point (and quality characteristics are met there), so without looking at quality decay measures, the most optimal point for sampling would to simply sample there (all water would then meet requirements).

Assignment #3

Garrett Hardin, "The Tragedy of the Commons," Science, 162(1968):1243-1248.

Summary:
This article is an interesting philosophical piece that was published in a scientific journal (which is somewhat interesting in itself) that talks about solving problems where technical means are simply not able to generate a result (meaning that some human thought or some other intervention is required to generate a "solution" going forward).

Principally, the paper uses the analogy of a "commons", which basically means an area of resource that is shared by a group of people (for raising livestock, hence the commons). The problem arises when the use of the resource of the commons reaches its limit (that is to say that the entire area suffers if the resource is utilized any more). The tragedy lies in that an individual that uses the additional resource receives much more benefit than loss since the problem is shared equally by all who use the commons. From this, we see the downfall of the commons because every user thus would rationally decide to get more benefit, and thereby everyone uses all of the natural resource. An extension of this problem discussed includes the likelihood that people will have as many children as possible (through various discussions).

The next important concept that the author attempts to tackle is to define effective mechanisms for regulating this effect (partially through pointing out the faults with using conscience-type means separate from penalties). The primary mechanism would be through a governmental (or mutually agreed upon body) that utilizes coercion.

Discussion:
This topic is very interesting for those interested in activities dealing with the development of natural resources. Many topics, especially like water, fall into similar categories. This discussion reminded me about some of the later stages that my dad talks about in his Environmental Management class with respect to the "Pattern of Resource Development" (which describes how a resource is used and misused (typically) throughout its lifespan from discovery to eventual total depletion.

From a philosophical standpoint, I think I disagree with the author to some extent about the establishment of means to regulate. I do generally agree that in most cases (particularly in business and water related issues) that some body is necessary that has authority and coercive abilities (take for example the EPA with water pollution). I do think that given certain situations, people can govern themselves if they use the proper mindset (one of his big assumptions in the consience portion seemed to dwell on the fact that people could be counted on to not act in a correct manner in relation to others based upon their conscience. In all, I am looking forward to class discussions on this topic.

Assignment #2

Atwood, D and S. Gorelick (1985) “HYDRAULIC-GRADIENT CONTROL FOR GROUNDWATER CONTAMINANT REMOVAL”, Journal of Hydrology, 76(1-2), pp. 85-106

Summary:
This article details a case study on groundwater containment pumping designs contaminated plumes associated with the Rocky Mountain Arsenal Superfund site in Colorado. In this case, a plume of contaminated groundwater existed underground from a pollution event, and a study was being performed to determine proper operation characteristics for a pump and treat treatment system to contain the plume (prevent its further movement), collect and treat the contaminated groundwater (via means not pertinent to the paper), and reinject water back into the aquifer to maintain hydraulic gradient needs. This paper focuses on the case study of first using physical transport modeling to simulate and determine the optimum location of the central collection well system that removes the contaminated water from the aquifer for treatment. Next, the paper utilizes a linear optimization process to determine potential well locations and operating schedules that could be used to ensure that the plume does not move off site (thereby creating the necessity to move the treatment well further downstream to collection the contaminated water) based upon the collection well location determined by the modeling from the first part.

In the linear optimization portion, the objective function seeks to minimize the amount of pumping and recharge necessary to control the aquifer movement (each set of wells and locations could be either pumped or charged based upon the program). Contraints upon the linear optimization process include 1) insuring that the model meets the containment criteria for the plume (it can't move around per the optimization's original goal) and 2) a constant removal rate from the central well (the removal of the contaminated water for treatment in the plume). Additional constraints occurred in the form of nonnegativity requirements. Considerable effort went into creating the hydraulic requirements necessary for the first constraint (relating numerically the motion of the groundwater based upon different pumping conditions). They then ran the programming to determine operation criteria over time (on sequencial and global scales). In the end, the model shows a well confined plume (the plume does not move from the site in question) that is treated in a minimal amount of time with less construction costs (fewer wells would be needed) compared to what would happen without the containment problems.

Discussion:

I found this article really interesting, especially given my background and knowledge of environmental remediation through my previous environmental classes. In CVEN 682 we actually discussed this site as an example of the CERCLA process, and thus I found the article a little more interesting. I also was able to see where I will be able to combine some of the new topics I am learning in Water Resources with my interests in environmental. As it says in the article, the solution managed to utilize linear optimization techniques with a nonlinear problem by separating the physical model from the optimization model. As we begin to do more optimizations, we should be aware of this and look for ways to make linear programming useful, even in models that are not necessarily linear (find ways to make assumptions--that's really a lot of what engineering is about). We can also see if more up to date models can account for these differences without having to separate the models (given greater computing abilities). In all, I really liked this example and case study.

Assignment #1

For this assignment, we are asked to read and review two articles. The first was given by Dr. Zechman, and the second was selected on the application of linear programming to solve a water resources problem. Each article is listed below (transition to the second article is noted by the second article citation.

Article #1:

Liebman, Jon (1976). "Some simple-minded observations on the role of optimization in public systems decision making," Interfaces 6(4) pp. 102-108. (Accessed using TAMU library website).

Summary:
This article presents an overview and a set of observations noted by the author (Liebman) in relation to utilizing simulations and optimizations in public decision making (rather than private decision making). The central question of the article seems to ask why optimization techniques work on some problems (private sector or simple public sector problems) and then why they don't work on other problems (large, complex public sector problems). The first task of the article was to present potential problems with solving public sector problems, in particular complex problems that involve tough questions (such as water resources problems). The author explains that the difficulty in using optimization techniques in water resources (and other public problems) is that often the public cannot agree on a specific set of goals that should be met, thus making it difficult to determine what parameter an analyst should optimize the problem to. Another problem with optimization is that it requires a knowledge of what will happen if a certain action is done (aka if the "science" behind the action is understood). Because of these problems, many public questions could not be effectively optimized because of the large degree of uncertainty. He essentially reasons that the analyst should thus be aware that these problems have occurred and try to find ways to limit the propigation of these problems in the solution. The author then makes 4 simple suggestions to attempt to deal with these problems, including:

1. That an analyst should realize that modeling is thinking being done publicly (i.e. helping others show that you are utilizing a thought process when arriving at decisions--and documenting the otherwise internal process outwardly)
2. Realize that models are not unique and that they can be implemented in multiple ways.
3. Understand that models are a message that others can look at and interpret when making decisions.
4. Redoing models ("Reinventing the wheel") can often be helpful in the public light when modeling, especially since other may want to include components in the model that had previously been assumed away.

Discussion:
After spending some time reading and thinking about the article (and seeing the concise summary towards then end), I gained a better appreciation for the thoughts expressed in the article (at first I had trouble following the main points). This article works well as an introduction to this course because it helps the reader (and potential future analyst) realize some of the limitations to modeling--likely some that may come up in this class. From a technical perspective, the paper is somewhat limited in that it does not necessarily give specific details in how to resolve the problems of optimization on a case-by-case basis; however, given the nature of the article, this point can be overlooked since it is meant to be of broad application. Basically, I think this article can be used to help me (in particular) to define some optimization problems better or to help me generate an output (or set of outputs) in optimization that could then be used to help make decisions. From this, I must understand that one particular solution may not necessarily be possible; however, I could work with the problem to generate information that would be useful when the time to make decisions actually comes.


Article #2:

Hall, Nigel (2001). "Linear and quadratic models of the southern Murray-Darling basin". Elsevier Science Ltd. (accessed at http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6V7X-4447S59-R&_user=952835&_coverDate=09%2F30%2F2001&_rdoc=1&_fmt=&_orig=search&_sort=d&view=c&_acct=C000049198&_version=1&_urlVersion=0&_userid=952835&md5=1838c449face56bbd66061704eac1f62)

Summary:

This article provides a summary of a simulation effort of an Australian watershed based upon both linear programming and quadratic programming techniques. The article briefly describes the watershed being modelled and the aspects that are to be modeled. The system required separate components of models for a river system, regions in Autralia, water supply, water demand, and water quality issues (salinity) and its affects. It also models various crops and seasonal changes. The main focus of the article is to account an added quadratic programming approach to a linear model. After running the simulation, the analysis is prepared such that the results of the linear programming model and quadratic programming model are compared to the actual water usage of the region for a specified control period, and the analysis showed that the quadratic models came much closer to predicting the actual water usage for the region. This calibrated model could then be used to help policy makers predict the affect of policies on the area before implementing the policies.

Discussion:
After reviewing the entire article, I realized that it was sumwhat off topic for my purposes of looking at linear programming. I expected more information to be given about the initial linear programming done on the area. Overall, I felt the article was interesting from an informational perspective, and even more impactful to someone who was highly interested in the specific Australian region as more information could be gained about the area. Also, a quadratic programming application could utilize the information presented when applying an analysis technique. Overall, I wish the article would better explain the system being analyzed (though I think the author's focus was slightly different than my focus--they were accounting for the differences in the improved modelling techniques). As a continuation, I think the research could be used to look at more years for the model and acutally used for more applications. The author does do a somewhat poor job of explaining the inputs and correlations (i.e. how the model is built), so I would also improve the article to include these aspects.