A LEAST-COST STRATEGY for EVALUATING a BROWNFIELDS REDEVELOPMENT PROJECT SUBJECT to INDOOR AIR EXPOSURE REGULATIONS Xiaomin Wang

PROCEEDINGS, TOUGH Symposium 2012 Lawrence Berkeley National Laboratory, Berkeley, California, September 17-19, 2012

A LEAST-COST STRATEGY FOR EVALUATING A BROWNFIELDS REDEVELOPMENT PROJECT SUBJECT TO INDOOR AIR EXPOSURE REGULATIONS

Xiaomin Wang1, André J.A. Unger1, and Beth L. Parker2

1Department of Earth and Environmental Sciences, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, N2L3G1, Canada 2School of Engineering, University of Guelph, 50 Stone Road East, Guelph, Ontario, N1G 2W1, Canada e-mail: [email protected]

ABSTRACT may be exposed to the air, potentially threaten public health, and have negative impacts on This paper presents a framework for alleviating ecological systems. Redevelopment of such the risk of exposure during brownfields redevel- brownfields is beneficial for the environment as opment projects by improving the prediction of well as for communities. It is an efficient, effec- contaminated indoor air concentrations, and by tive, and environmentally friendly way to searching for the optimal balance between encourage development using existing infra- increasing site characterization budgets and structures, services, and resources. Redeveloping decreasing contingency measures to cover the brownfields can also generate great economic risk exposure (and also to quantify and manage benefits if appropriate and reasonable methodol- the risk). A three-dimensional model CompFlow ogies are designed. There are strong environ- Bio is used to anticipate the intrusion pathways mental, social, and economic grounds to of TCE from a residual source zone into redevelop and utilize brownfields (US EPA, buildings. The result indicates that the geologi- 1999; NRTEE, 2003; UK Environment Agency, cal condition of the site and the heterogeneity in 2003). each stratigraphic unit have the most compelling impact on the indoor air concentration of TCE. Since the 1980s, North America and Europe We develop a methodology by which to price have made moves to foster sustainable redevel- the risk associated with the number of sampling opment of brownfields. The United States of data available from site characterization and America enacted the Liability Act Superfund to long-term monitoring, incorporating hydrogeo- reclaim 1,410 heavily contaminated sites, and logical as well as financial uncertainties into an the National Brownfields Association has insurance premium. As a result, the trade-off estimated that 600,000 brownfield sites exist between increasing site-investigation budgets across the nation (Mueller, 2005). Canada also and overall cost reductions is optimized. has paid a great deal of attention to brownfield TM INTRODUCTION redevelopment projects (aboutRemediation , 2012). Policy makers and developers have Brownfields are defined by the U.S. Environ- focused their efforts on transforming brown- mental Protection Agency (US EPA) as “aban- fields into industrial areas, commercial areas, doned, idled, or under-used industrial and and residential areas, depending on the charac- commercial facilities where expansion or teristics of the community and the site itself. At redevelopment is complicated by real or the same time, brownfield redevelopment perceived environmental contamination.” projects also harbor great risks, which can Brownfields are usually unintended by-products impede the decision-making process. Major of industrial practices of the last several decades, risks faced during brownfield redevelopment in which measures were not taken to ensure that associated with costs include possible cost industrial operations did not harm the natural overruns in cleanup operations (the technologies environment. In brownfields, there exist known for remediation of brownfields are summarized and potential contaminants, such as volatile by Reddy et al., (1999)), possible liability claims organic compounds (VOCs) or semi-volatile from accidents or contaminant exposure from organic compounds (SVOCs) which over time, past spills or spills during the cleanup, and

- 1 -

uncertainty about future community acceptance costly. What is the most cost-effective way to (Wade VanLandingham et al., 2002). All of conduct site characterization? To answer this these risks mainly arise from imperfect question, different concentration and knowledge of the complexity of brownfield permeability sampling schemes are set up. The sites. Complications include complex geological heterogeneous permeability fields are formations and physical properties of the subsur- constructed using a conditional stochastic face, the uncertainty of the source zone, the approach, generating a high-resolution aquifer number and features of existing and potential analog based on permeability measurements. contaminants, and the unreliability of detecting The geostatistical program S-GEMS (Stanford and contouring of the contaminant plume. As a geostatistical earth modeling software) is result, a better understanding of the physics of utilized for this purpose. Once we have concen- contaminated soil vapor transport and develop- tration measurements, an optimal estimator ing a methodology to evaluate the associated Kalman filter can be used to deduce a minimum risks are crucial. error estimate to improve model prediction. Zhang and Pinder (2005) applied this algorithm Soil vapor VOC intrusion pathways into build- to determine the least-cost design of groundwa- ings have been commonly studied using analyti- ter quality monitoring networks. cal models, numerical modeling, and field studies. A discrepancy between field measure- How can one optimize the trade-off between ments and analytical solutions is often seen. The increasing site-investigation budgets and overall influence of the geometry of houses on the soil cost reductions? First of all, the cost of risks, vapor distribution is often ignored. Yu et al. combining both hydrogeological and financial (2009) simulated the TCE soil vapor intrusion risks, is defined using a modified actuarial into a single house using CompFlow Bio in two approach. The risk-cost-benefit analysis is then dimensions. They examined the roles that a conducted according to the Massmann and number of uncertainties—heterogeneity, the Freeze (1987) strategy, of which the objective pressure difference between indoor and outdoor, function is the sum of the project risk, cost, and source-zone location, aperture of the crack in the benefit, to maximize profits by adjusting the foundation slab, depth of the foundation slab, number of data. The financial factors, which in capillary fringe thickness, and infiltration rate— most cases follow a Geometric Brownian might play in determining the indoor air Motion, (e.g., the discount rate, U.S. house price concentration of TCE. The largest remaining index), are estimated using financial models, and uncertainties result from the inherent complexity their importance is weighed compared with the of the geology, heterogeneity, which was hydrogeological factors (mentioned above) responsible for deflecting the groundwater contributing to uncertainties in the decision- plume upward towards the capillary fringe, making process. diffusing across the capillary fringe, and then migrating into the basement. This work intends SIMULATING THE FATE AND to use the three-dimensional CompFlow Bio TRANSPORT OF SOIL VAPOUR TCE model to simulate the fate and transport of soil vapor into multiple residential houses. We focus The numerical simulator CompFlow Bio is more on the significance of the geological applied to the soil-vapor-intrusion problem in conditions, the geometry of house locations, and this case. It is a fully coupled, multiphase, multi- the heterogeneity influencing the indoor air component model capable of simulating concentration distribution of TCE. Attenuation groundwater flow and solute transport with factors in different scenarios are calculated in variable water saturation in the subsurface. It contrast with those using the model by Johnson involves advective and diffusive transport of and Ettinger (1991). TCE in water, oil, and gas phases under scenarios with different geological and hydrogeologi- Developers want to keep project overhead as cal conditions, (e.g., heterogeneity, the capillary low as possible, but characterizing site geology fringe thickness, the existence of low permeable and probing for contaminants are necessary and

- 2 -

soil and the distance of houses offsetting the Recharge

TCE plume). Furnace Basement

Gas phase Conceptual Model flow Foundation crack

Mass transfer across The hypothesis of this model is that a brownfield Groundwater flow capillary fringe site has been remediated and residential build- NAPL source zone ings have been built, but a residual TCE source Groundwater plume exists below the water table in the subsurface. The problem is as illustrated in Figure 1. The source of TCE, which is located upstream of Figure 1. Conceptual model residential houses, can either dissolve into the ambient groundwater or volatilize into the ambient soil gas. It continues to migrate along with groundwater flow, diffuse across the capillary fringe, and is subsequently transported towards the foundation slab of a structure located below grade, due to the vacuum pressure in the basement.

The computational model domain for the base scenario is depicted in Figure 2 and Figure 3 without the clay layer. Six houses are constructed on the domain with seven monitor- Figure 2. The plan view of the model domain for ing wells in the front, and the central houses are the base scenario aligned with the source zone. The houses are depressurized, with the pressure difference from the ambient atmosphere pressure 10 Pa. The aquifer is heterogeneous. The permeability field is created stochastically with the same statistics (mean, variance and correlation lengths) as the Borden field site (see Figure 4). Five scenarios are set up for investigating the sensitivities of the different influential factors.

1) Base scenario; Figure 3. A 2D cross-sectional view of the model 2) Base scenario under recharge condition with domain for Scenario 3 with a clay layer recharge rate of 20 cm/yr;

log (k ) 3) Base scenario with a layer of clay 10 -15 2 Basement -9.5 permeability 10 m and the thickness 2 m -9 -10 k = 1 X 10 -10.5 situated beneath the houses; 2 -11 m -11.5 8 -12 4) The same as case (3) but under recharge -12.5 6 -13 conditions; -13.5 4 -14 Z 150 -14.5 5) The vadose zone is 7 m thicker than the one 2 100 in the base scenario. 0 X 50 40 Y 20 0 0 Foundation slab Clay layer -13 2 -15 2 k = 1 X10 m k = 1 X10 m Figure 4. 3D plot of the heterogeneous permeability field with a clay layer

- 3 -

Model Results Table 1. The attenuation coefficients calculated using both J&E and CompFlow Bio (CFB) models The results in Figure 5 for Scenarios 1, 2, and 5 show that the TCE plume travels longitudinally along the groundwater flow direction for some distance, and then transfers across the capillary fringe into the vadose zone. The plume is quite narrow because the transverse migration of the plume is primarily driven by the concentration gradient and heterogeneity. When the vacuum

log (X ) pressure is applied, the central houses directly (a) House No. 10 TCEg 8 8 8 -2 Height (m) Height (m) Height (m) 2 5 123 456 -3 above the groundwater plume have the highest 6 6 6 -4 4 4 4 -5 indoor air concentration of TCE. The TCE -6 2 2 2 -7 -8 concentrations in the lateral houses vary, which 0 0 0 -9 0 50 100 150 0 1020304050 0 1020304050 results from heterogeneity, i.e., the structure of Distance (m) Lateral distance (m) Lateral distance (m) (b) log10 (XTCEg ) 8 8 8 -2 Height (m) Height (m) the permeability field. If a low-permeability soil Height (m) 25 123 456 6 6 6 -3 -4 layer is located below the foundation slab, when 4 4 4 -5 -6 2 2 2 -7 it is saturated with water, it behaves as a barrier -8 0 0 0 -9 0 50 100 150 0 1020304050 0 1020304050 and prevents the soil vapor from migrating Distance (m) Lateral distance (m) Lateral distance (m) log (X ) (c) 10 TCEg upwards into the basement. The thicker vadose 15 15 15 12 45 -2 2 5 3 6 -3 zone alters the offset between the source zone -4 -5 Height (m) Height (m) Height (m) 10 10 10 -6 and center of the foundation slab, allowing more -7 -8 mass transfer of vapor TCE into the houses. The -9 5 5 5 indoor air concentration of TCE calculated for the five scenarios is illustrated in Figure 6. It 0 0 0 0 50 100 150 0 1020304050 0 1020304050 indicates that the geological clay unit and the Distance (m) Lateral distance (m) Lateral distance (m) recharge condition play important roles in vapor Figure 5. 2D cross-section plots along x- axis and y- TCE transport. axis of mole fraction of TCE in gas phase 1800 days after TCE was injected into the Additionally, the comparison of the five aquifer: (a) Scenario 1, (b) Scenario 4, different scenarios is conducted in terms of the and (c) Scenario 5 attenuation coefficient (defined as the ratio of (a) (b) (c) 100 the contaminant vapor concentration in buildings base scenario 1 2 3 10-2 base & recharge clay

to the vapor concentration at the source of TCE (ppmv) -4 f 10 clay & recharge thicker vadose zone contamination) with respect to the Johnson and -6 10

Ettinger (1991) (J&E) model. The attenuation 10-8 coefficients calculated by the CompFlow Bio are 10-10

10-12 Indoor air concentration o far smaller compared to those under the same (d) (e) (f) 100 conditions using the J&E model (see Table 1). 4 5 6 10-2 TCE (ppmv) -4 f 10

Since heterogeneity is a primary driving factor 10-6 in determining soil gas exposure pathways, the 10-8 Monte Carlo method is adopted, using permea- 10-10 10-12 Indoor air concentration o 0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000 bility as the primary uncertainty for the model Time (days) Time (days) Time (days) input, to address this issue. Figure 7 shows the Figure 6. Indoor air concentration of TCE in concentration results for different permeability logarithmic scale with transient air realizations. The differences in indoor air exchange rate limited to wind speed for the five scenarios: (a) house no. 1, (b) concentration of TCE can reach up to four house no. 2, (c) house no. 3, (d) house no. orders of magnitude in one single house. 4, (e) house no. 5, and (f) house no. 6 Note: The numbers 1–6 represent the house number from Figure 2. The dashed line represents the regulatory limit for TCE

- 4 -

(a) (b) (c) 100 data (80 K) is also considered. Every permeabil- 1 2 3 10-2 ity realization is used as a numerical model input TCE (ppmv) -4 f 10 to estimate the indoor air concentration of TCE 10-6 at six houses and five observation wells at 100- 10-8

10-10 day intervals over 2000 days. Sequentially, the

10-12 Indoor air concentration o space-time TCE concentration covariance matrix (d) (e) (f) 100 4 5 6 is constructed for the 11 locations. Concentra- 10-2

TCE (ppmv) tion samples are taken from 2, 4, and 7 locations -4 f 10

-6 (taken as 2C, 4C, and 7C) out of 11, and then 10

10-8 KF is used to update the concentration estima- 10-10 tion by minimizing the error covariance matrix.

10-12 Indoor air concentration o 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 The concentration sample in house no. 2 is Time (days) Time (days) Time (days) always included. Figure 7. Indoor air concentration of TCE in logarithmic scale with the air exchange The purpose of this method is not only to rate 0.5/hr for 50 unconditioned random examine the data-worth issue for this applica- permeability realizations: (a) house no. 1, tion, but also to use sparse data to infer whether (b) house no. 2, (c) house no. 3, (d) house one of the houses located adjacent to house no. 2 no. 4, (e) house no. 5, and (f) house no. 6 has indoor air problems subject to the condition Note: The numbers 1–6 represent the house numbers from Figure 2. The dashed that house no. 2 is monitored and has problems; line represents the regulatory limit. The adjacent houses are not monitored, but develop- red line represents the hypothetical real ers assume liability if a problem occurs. After world result filtering, the expected value and the standard deviation of the TCE concentration in adjacent DATA-WORTH ASSESSMENT houses can be calculated.

Kalman filtering (KF) is a technique to describe k(m)2 2E-10 borehole locations 1.8E-10 how a system would respond after processing 1.6E-10 1.4E-10 1.2E-10 1E-10 measurements to achieve an optimal estimation. 8E-11 6E-11 4E-11 It is, in essence, a recursive solution to a least- 2E-11 1E-11 squares problem. In this work, KF is applied to estimate and predict the indoor air concentration in multiple houses, with prior knowledge and reality statistical information on state variables obtained a conditioned permeability from the CompFlow Bio model. realization using 7 boreholes Methodology Figure 8. An example of a randomly generated permeability field realization conditioned The strategy to assess the data-worth issue here on the 7 boreholes data is to choose one of the randomly generated KF Results permeability realizations to be reality, and to extract the permeability data from 1 to 7 Combinations of different numbers of permea- boreholes in front of the houses as the sampling bility data and concentration data are made in data. The next step is to use the statistics and the order to get a good estimation of reality. The correlation of the sampling data, to apply the representative four cases (1K, 2C; 1K, 4C; 7K, kriging technique to reconstruct the permeability 4C; and 7K, 7C) are chosen for illustration in field of the entire domain, and then to employ a Figure 9. sequential Gaussian simulation to create multiple realizations (shown in Figure 8). 1, 3, 5, Increasing the number of permeability samples and 7 boreholes of data (labelled as 1K, 3K, 5K, can give a better estimation of the expected and 7K) are taken respectively for the inspec- value of the TCE concentration in comparison tion, and an extreme case with 80 boreholes of with reality; increasing the number of concen-

- 5 -

tration samples may reduce the variance in the 0.3 (a) 1K, 2C TCE concentration, so that the uncertainty in 0.2 1K, 4C prediction may be reduced. 7K, 4C 0.1 7K, 7C 1 Probability (-) 0 The probability of failure is the likelihood that 0.3 (b) 1K, 2C indoor air concentration of TCE exceeds a 0.2 1K, 4C regulatory level. It is defined as the total number 7K, 4C 0.1 7K, 7C of Monte Carlo realizations in which the failure 3 Probability (-) 0 occurs, divided by the total number of realiza- 0.3 (c) 1K, 2C tions. The probability is calculated for house no. 0.2 1K, 4C 1, 3, 4, 5, and 6 at each monitoring interval; the 7K, 4C 0.1 7K, 7C result is shown in Figure 10. 4 Probability (-) 0 1 (d) (a) (b) (c) 0.8 1K, 2C -4 1 2 3 0.6 1K, 4C 7K, 4C -8 Reality 0.4 1K, 2C 7K, 7C 0.2 -12 1K, 4C 5 7K, 4C Probability (-) 0 7K, 7C -16 0.3 (e) -20 1K, 2C 0.2 1K, 4C -24 7K, 4C Indoor air concentration (ppmv) 0.1 7K, 7C -28 6

(d) (e) (f) Probability (-) -4 0 4 5 6 0 200 400 600 800 1000 1200 1400 1600 1800 2000 -8 Time (days)

-12

-16 Figure 10. The probability of failure at each 100 days

-20 monitoring intervals for four cases: (a) house no. 1, (b) house no. 3, (c) house no. -24

Indoor air concentration (ppmv) 4, (d) house no. 5, and (e) house no. 6 -28 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 20 Time (days) Time (days) Time (days) Note: The numbers 1–6 represent the house number from Figure 2 Figure 9. Updated indoor concentration of TCE in logarithmic scale using a Kalman filter: a) house no. 1, b) house no. 2, c) house no. LEAST-COST OPTIMIZATION 3, d) house no. 4, e) house no. 5, and f) house no. 6 As mentioned above, after the brownfield site is Note: The number 1 – 6 represents the cleaned up, there is still a chance that the house number from Figure 2. The dash residential houses will be affected unexpectedly line represents the regulatory limit. The by contaminated soil vapor. If it happens, the red line represents the hypothetical real cost is essentially described as the probability of world result failure times the cost of failure, with an . additional safety loading term to compensate against the associated hydrological and financial uncertainties. The probability of failure can be obtained from the last section (above), and the cost of failure is contributed by the house price as well as the uncertainty of the financial market. The safety loading term reflects the level of risk aversion and is usually defined as the variance or standard deviation in the cost of risk. The total cost concerned for the redevelopment is the sum of the cost of risk and the cost of data. In order to determine the least cost among the numerous cases listed earlier, the objective function is herein proposed in Equation 1 and 2.

- 6 -

also means a lower safety loading reserve for contingency. The cases with 5 and 7 wells of permeability data show this pattern, while the case with only 1 well of permeability data shows some oscillations and therefore should not be used for this demonstration. On the other hand, with more data, the operational cost increases. exp exp Consequently, there must be an optimal solution in which the cost-effectiveness is achieved. In Figure 11(d), seven wells of permeability data (1) with four of concentration data gives us the and ∑ minimum cost; seven wells of permeability data (2) with seven of concentration data can be an

alternative solution. where is the present value of the sum of the basic risk capital in each future period ,; (a) (b) is the expected value of losses from 700000 700000 600000 600000 repossessing houses in each future period 1perm 500000 3perm 500000 , ; is the uncertainty in predicting 5perm 7perm future repossession expenditures due to imper- 400000 80perm 400000 300000 300000 fect knowledge of hydrogeology and financial E[H] ($) A[H] ($) market; is the number of sample locations for 200000 200000 100000 100000 soil gas concentration; is the number of wells 0 0 01234567 01234567 extracting permeability; is the unit cost for No. of concentration data No. of concentration data (c) (d) 700000 700000 each sample concentration, is the cost of taking one borehole permeability measurement, 600000 600000 500000 500000 is the inflation rate at time and is the 400000 400000 discount rate at time . data ($) f 300000 300000 Total cost ($)

Cost o 200000 200000 Suppose that every house is sold at $200,000, 100000 100000 and that the central two houses have been 0 0 01234567 01234567 repurchased since a potential soil vapor TCE No. of concentration data No. of concentration data exposure risk is considered. If the indoor air concentration of TCE in lateral houses exceeds Figure 11. The variability of the related costs with the regulatory limit, it assumes that developers different number of permeability and have the liability for repurchasing the house, and concentration data: (a) the expected value when the house is resold, it will lose 20% of its of the cost of risk, (b) the safety loading original value. For developers, the maximum term on the cost of risk, (c) the cost of data, and (d) the overall cost liability would be when all the lateral houses are repossessed and they can never sell them again, CONCLUSIONS which is $800,000. The expected value of , the safety loading term, the cost of data and the total On the basis of the above results, the following cost-risk are shown in Figure 11, with numerous conclusions may be drawn: permeability and concentration sampling  First, the geological condition of an aquifer schemes. is one of the most important factors control- ling the fate and transport of TCE from a If more permeability data are available prior to source zone below the water table into the model prediction and more concentration houses. When a low-permeability layer data are applied for data assimilation, there is a saturated with water is located beneath the lower expected value of risk, due to the fact that foundation slab, it prevents the soil gas TCE the lateral houses in reality have no risk of pathways from moving upwards. If a thicker exceeding the regulatory limit. Such a situtation vadose zone exists, it makes the mass

- 7 -

transfer across the capillary fringe more Mueller, G. R., Brownfields capital - unlocking difficult, and the soil gas is more likely to value in environmental redevelopment, diffuse away from the houses. Journal of Real Estate Portfolio Manage- ment, 11(1), 81-92, 2005.  Heterogeneity in the permeability of the aquifer causes the TCE plume to be National Round Table on the Environment and deflected upward towards the houses. The the Economy (NRTEE) Canada, Cleaning up impact of heterogeneous porosity will be the Past, Building the Future: A National investigated in future research. Brownfield Redevelopment Strategy for Canada, 2003. Available at http://nrtee-  By using the spatial and temporal sampling trnee.ca/wp- data, and dynamically updating the sampling content/uploads/2011/06/brownfield- strategy, the model result is greatly redevelopment-strategy-eng.pdf (accessed improved. The more data that the Kalman February 2012). filter utilizes for assembling the concentra- Reddy, K. R., Adams, J. A. and Richardson, C., tion correlation, the more accurate the Potential technologies for remediation of estimations are. brownfields, Practice Periodical of Hazard-  The cost of risk is defined in the same way ous, Toxic, and Radioactive Waste Manage- that actuaries calculate the insurance ment, 3(2), 19821, 1999. premium, with an additional term —a safety UK Environment Agency, Brownfield Land loading term—to compensate for hydroge- Redevelopment: Position Statement. May ological and financial uncertainties. 2003. Available at http://www.environment- Agency.gov.uk/static/documents/Research/b  An optimal solution, with the minimum cost rownfield_land_908146.pdf (accessed of a brownfield redevelopment project for February 2012). different numbers of permeability data and concentration data, can be obtained by a United States Environmental Protection Agency risk-cost-benefit analysis. (US EPA), A sustainable brownfields model framework. January 1999. Publication ACKNOWLEDGMENT Number: EPA-500-R-99-001. Funding for this work is provided by the Univer- Wade VanLandingham, H., The stormstown sity Consortium for Field-focused Groundwater group and Meyer, P. B., Public strategies for Contamination Research Program. cost-effective community brownfield redevelopment. Southeast Regional Environ- REFERENCES mental Finance Center, EPA Region 4, University of Louisville, 2002. Available at TM aboutRemediation , Brownfields redevelop- http://cepm.louisville.edu/Pubs_WPapers/pr TM ment toolbox . Available at acticeguides/PG1.pdf. http://www.aboutremediation.com/Toolbox/I Yu, S., Unger, A.J.A. and Parker, B., Simulating ntroduction.asp (accessed April 2012). the fate and transport of TCE from ground- Johnson, P.C. and Ettinger, R.A., Heuristic water to indoor air, Journal of Contaminant model for prediction the intrusion rate of Hydrology, 107(3-4): 140-161, 2009. contaminant vapors into buildings, Environ. Zhang, Y. and Pinder, G.F., Least cost design of Sci. Technol., 25: 1445-1452, 1991. groundwater quality monitoring networks, Massmann, J. and Freeze, R.A., Groundwater Water Resources Research, 41: W08412, contamination from water management sites: doi:10.1029/2005WR003936, 2005. The interaction between risk-based engineering design and regulatory policy: 1. Methodology, Water Resources Research, 23(2): 351-367, 1987.

- 8 -