Mr. Patrick Sweetland
October 22, 2002
Page 1
Gus Yates Hydrologist
1809 California Street, Berkeley, CA 94703
Tel/Fax (510) 849-4412
October 22, 2002
Mr. Patrick Sweetland, Manager
City of Daly City, Dept. of Water and Wastewater Resources
153 Lake Merced Blvd.
Daly City, CA 94015
Subject:Blueprint for a Unified and Improved Groundwater Model of the Westside Basin in San Francisco and San Mateo Counties
Dear Mr. Sweetland:
A meeting of technical experts familiar with geology, groundwater modeling, the Westside Basin and Lake Merced was held at Daly City Hall on Friday September 20, 2002. The meeting was convened by Gus Yates and John Fio pursuant to Daly City’s AB303-funded groundwater management project. In attendance were:
Gus Yates, consulting hydrologist (groundwater consultant for Daly City)
John Fio, HydroFocus, Inc. (groundwater consultant for Daly City)
Dave Van Brocklin, Luhdorff and Scalmanini Consulting Engineers (groundwater consultant for San Francisco)
John Plummer, Friends of Lake Merced
Dick Morton, co-chair, Lake Merced Task Force
Katie Pilat, hydrologist (consultant for Neighborhood Parks Council)
Erdmann Rogge, manager, San Mateo County groundwater protection program
Mondy Lariz, CalTrout, Lake Merced program manager
Other invitees who declined or were unable to attend included the three members of the SFPUC -Daly City groundwater technical review panel (Greg Bartow, SWRCB; David Evans, CSU Sacramento; and John Suen, CSU Fresno), Toni Pezzetti (CH2M HILL), Carl Hauge (DWR), Tim Colen (Lake Merced Task Force). At the last minute, David Dawdy (hydrologist for FOLM) was unable to attend, and Gus Yates met separately with him on September 27 to obtain his input.
The purpose of the meeting as stated in the AB303 scope of work was to solicit additional input on the strengths and weaknesses of the prior groundwater models (the Geo/Resource Consultants model and the CH2M-HILL model) and to create a detailed blueprint for combining the best elements of those models -- and possibly also new data and methods -- to create a reliable and broadly accepted analysis tool.
The meeting agenda consisted of a comprehensive list of hydrologic processes, data sources and simulation approaches that must be specified in the process of constructing a model. The agenda packet for the meeting included a description of how the previous models had addressed each item. The items are listed below, along with a summary of the participants= comments and recommendations.
Mr. Patrick Sweetland
October 22, 2002
Page 1
Recommended modeling software. The discussion focused on the relative merits of MODFLOW and MicroFEM, which were the model codes used for the two previous modeling efforts. The advantages and disadvantages of each are shown in Table 1. Although either model code could be successfully used as the foundation for a combined and improved model, the consensus of the group was that MODFLOW would be a better choice overall.
Table 1. Advantages and Disadvantages of MODFLOW and MicroFEM
MODFLOW / MicroFEMAdvantages / Advantages
Open source code / Finite-element mesh conforms more exactly to land use polygons
Code is free and fully supported by USGS / Finite-element mesh spacing can be decreased locally in areas of complex hydraulics
Larger pool of hydrologists can readily use it / The existing MicroFEM model is already basin-wide in extent
Lake package is integral to model
Optional water quality module built into model code
Disadvantages / Disadvantages
Decreases in row and column spacings to better simulate local complexities must extend the full width and length of the grid. / Unrealistic assumptions and features need to be changed (e.g. anisotropy of horizontal hydraulic conductivity; hydraulic decoupling of south end of Lake Merced)
Input data for the previous MicroFEM model needs to be transcribed to MODFLOW. However, the large data-development effort for the MicroFEM model is equally usable for MODFLOW. / Calibration needs improvement (eliminate obvious trends in water-level residuals in Lake Merced area; extend calibration period backward in time to include major drawdown period of basin; add calibration to short-term lake level changes)
Lake simulator is separate from model; model execution must be halted each stress period to run the lake simulator.
Water quality module not available.
Proprietary source code (computational details are a "black box")
Boundary Locations. Onshore no-flow boundaries should follow the bedrock contact used in the CH2M-HILL model with the exception of the southeast corner, near Millbrae, where the boundary should be moved south to a shallow, partially-exposed bedrock ridge near Coyote point that also appears to be a groundwater flow divide.
In the west, the Pacific Ocean offshore boundary used in the CH2M-HILL model (follows the San Andreas Fault) is fine. In the east, consider realigning the offshore SF Bay boundary to more nearly parallel the shoreline.
Number of model layers. Four layers are recommended. Layer 1 (top) corresponds to dune and Colma Formation deposits in the western part of the basin and alluvial deposits overlying the older bay muds at the eastern end. Layer 2 represents one or more confining layers that create the observed vertical head gradient in the Lake Merced area and represents the older bay muds near San Bruno. Layer 3 is the Merced Formation down to the depth of the deepest water wells. Layer 4 is additional thickness of Merced Formation below the water wells. In practice, significant additional thickness is present only in the Daly City-Lake Merced area. Dividing the Merced Formation into two layers allows the pumping stresses to be correctly concentrated in the depth interval where they occur, while allowing potential storage effects of the deeper part of the Merced Formation to still be simulated.
Base of groundwater system. The base of the simulated flow domain is the contact between the Merced Formation (or locally, the dune sands or Colma Formation) and consolidated bedrock. This is the same as the base used in the CH2M-HILL model.
Tilting of layers; folds and faults. Layer top and bottom elevations should follow the top and bottom elevations of bedrock and confining layers, where present. An exception should be made in the outcrop area of the Merced Formation between Daly City and the coast, where the Merced Formation is locally tilted to angles as steep as 76 degrees. Tilting model layers to such extreme angles would complicate assignment of vertical and horizontal hydraulic conductivity as well as recharge, and potentially cause model cells to go dry in a way that would destabilize solution of the models’ mathematical equations. Instead, model layers in that area should remain near-horizontal, and the effects of Merced Formation structure on groundwater flow should be quantified by adjusting various combinations of aquifer characteristics, faults, and/or ocean boundary conductances.
The MODFLOW fault package should be used to simulate the southeastern end of the Serra Fault (near Millbrae), where the fault has a distinct offset and may create a barrier to groundwater flow.
Model node density. Vary the node density in a manner similar to the previous models, with cell dimensions as small as 100-200 feet near Lake Merced (to facilitate simulation of subsurface flow between North, South and Impound Lakes and between the lakes and the ocean) to 1,000-1,500 feet in parts of the basin where there are few pumping stresses and no measured water-level data.
Calibration periods and time step length. Model applications may range from short-term simulations for which stress periods of 1-2 days would be appropriate, to long-term simulations for which quarterly stress periods would be more reasonable. Multiple calibration periods are therefore needed. For long-term calibration, the model will simulate water years 1959-2002 using quarterly stress periods. These parameters were picked for the following reasons:
!The period includes much of the major historical decline in water levels that occurred between the late 1940s and the 1970s. The model will need to be able to simulate long-term responses to evaluate conjunctive-use scenarios.
!Annual rainfall during that period equals average annual rainfall for the past 50 years.
!Land use (specifically, imperviousness and irrigated area) was approximately constant during this period, which simplifies estimation of recharge. The basin reached its present level of development by about the mid-1950s.
!Quarterly stress periods offer some representation of seasonality of pumping, recharge, and water levels without creating excessively large input files and long run times.
For short-term stresses, such as rapid additions of water to Lake Merced, the model should be calibrated to simulate one of the historical water-addition events using stress periods of 1-2 days over a period of several months. Short-term simulations will be needed to investigate potential ecological impacts of pulses of stormwater inflow to Lake Merced or lake drawdown during pumping for emergency use.
Simulations of intermediate duration using monthly stress periods might also be needed to evaluate ecological effects of gradually raising lake levels. This would require development of monthly recharge and pumping data sets for a period of several years.
Initial conditions. Initial water levels for the entire flow domain should be developed in an iterative process by contouring the sparse available measured water levels and then running them in the model for 1-2 stress periods to dissipate any hydraulic unreasonableness. Prior experience confirms that initial-condition errors appear to dissipate rapidly.
Aquifer characteristics: Kh, Kv, S, anisotropy. Horizontal (X-Y) anisotropy of hydraulic conductivity should not be included. Calibrated hydraulic conductivity (Kh, Kv) and long-term storativity (S) from the CH2M-HILL model can be used for initial estimates. Additional calibration may be needed. Parameter adjustment zones should be broad and reflect plausible variations in hydrogeology. A separate set of S values may be needed to correctly simulate short-term hydraulic responses. The set of S values to use in any given simulation of management alternatives will depend on the type and duration of the stress.
Distributed recharge (rainfall, irrigation, pipe leaks). The soil-moisture-budget (SMB) method used for both of the prior models is a reasonable approach. Land use throughout the basin was already mapped and characterized for the CH2M-HILL model, but recharge zone parameter values should be reviewed. Rainfall at San Francisco Airport is the best indicator of rainfall in the Lake Merced area and is thus probably the best station to use for the entire basin. Local adjustments can be made using the existing isohyetal map, updated to reflect data from the OWPCP gage.
The soil-moisture-budget model should continue to use the Blaney-Criddle method for estimating evapotranspiration, because no CIMIS stations or other suitable microclimate stations are available nearby. The USGS calibrated the SMB runoff parameters to San Francisco combined-sewer flow data. The possibility of also calibrating to measured flows in other combined sewer systems in the Westside Basin area, theVista Grande storm drain or Colma Creek should be investigated
Colma Creek. Colma Creek reportedly gains flow from groundwater discharge along part of its length, which means gaged streamflows might be usable for model calibration. The MODFLOW stream package would need to be used to accomplish this.
Inflow from bedrock uplands. This minor source of recharge should be entered into the model by allocating the estimated recharge rate among a series of hypothetical Arecharge wells@ distributed along the inland boundary of model layer 1. This approach was used in both prior models.
Groundwater extractions. CH2M-HILL completed an extensive well inventory and methodically estimated historical pumping at major wells for their 1975-1995 calibration period. Estimated pumping can be extended back to 1959 using the same methods, and measured data are available for many wells since 1995.
Seawater intrusion. For basin-wide water management purposes, it is not necessary to simulate the saltwater-freshwater interface or the density-dependence of groundwater flow near the ocean boundary. It is sufficient to assume a constant freshwater head equal to sea level at the ocean boundary cells in the model and use the direction of the boundary flux as the indicator of intrusion. These cells include all layer 1 cells located offshore. Sensitivity analysis should be done to test the effects of also including cells along the offshore edges of layers 2, 3 and 4 (except along the San Andreas Fault) as part of the ocean boundary. It is recommended that the boundary condition be represented as a general-head boundary rather than a constant-head boundary. The principal difference is that the former includes a boundary conductance term representing the distance from the edge of the model grid to the true ocean boundary and the hydraulic characteristics of the intervening aquifer materials. Consequently, it provides more flexibility for calibration and hypothesis testing than does a constant-head boundary.
Lake Merced. The MODFLOW lake package (LK2) is well-suited for simulation of Lake Merced. It allows for multiple lakes (e.g. North, South and Impound), applies the correct stage-area curves for calculating evaporation and seepage, and allows timeseries of surface inflows and outflows to be included in the lake water budget (e.g. Vista Grande diversions, local runoff, historical Harding Park Golf Course pumping, hypothetical emergency pumping, etc.).
Water quality. A regional model is not an appropriate tool for simulating contaminant plumes. It may be useful for evaluating regional water quality issues such as salt balance or nitrate loading for the whole basin, however. MT3D (the solute-transport module of MODFLOW) can be used to address those issues. It is also capable of providing flow path vectors and time-of-travel calculations, which could be very helpful for model verification and meeting regulatory requirements related to percolation or injection of reclaimed water.
Calibration. Manual and automated calibration methods each have their advantages and disadvantages, and it is recommended that both be used. Automated methods incorporate a sensitivity analysis, which was notably absent from both prior modeling efforts. Automated methods also identify the Aoptimal@ parameters for minimizing errors, but the user must often impose numerous constraints to keep the parameters within a realistic range of values. Automated methods do not consider the timing and location of errors, however, and those characteristics often indicate the cause of the error. Manual calibration considers those characteristics and can also identify wells that seem to be inconsistent with general patterns. It also invariably results in an intimate understanding of the basin and the factors that most strongly influence groundwater conditions.
Long-term calibration should start with adjustments of vertical and horizontal hydraulic conductivity, storativity and the ocean boundary conductance, making adjustments only over broad zones. Further improvement can be obtained through adjustment of recharge, bedrock inflow, and the vertical distribution of pumping. For example, excessive water-level rises in wet years can indicate a need to adjust the SMB model. The magnitude of annual pumping at selected wells should be reevaluated if these other variables fail to adequately calibrate the model. All adjustments should be within the range of uncertainty of the initial estimates.
Calibration should be done using transient simulations with the initial conditions described earlier. The groundwater system was not in steady-state at the start of the calibration period (1959), and a preliminary steady-state calibration would probably be of little value.
Measured water levels are about the only data available for calibration. These can include data from water wells (many measurements over many years), Lake Merced, monitoring wells in the Lake Merced area (good detail, but short record), as well as monitoring wells at various groundwater contamination sites around the basin (particularly useful for calibrating vertical hydraulic conductivity in layers 1 and 2). If groundwater discharge into Colma Creek is clearly recognizable in the streamflow record, that flux can also be used for calibration.
Simulated alternatives for Lake Merced. The model will only be useful if it is capable of simulating any and all water management projects and scenarios that might be proposed in the foreseeable future. All of the alternatives that have been proposed in the past 15 years for managing Lake Merced water levels were considered at the meeting, and it was found that all of them could be incorporated into the model by one of the following methods: 1) addition of water to Lake Merced by way of a revised inflow timeseries in the lake package, 2) injection or extraction wells, or 3) a change in recharge simulated using the SMB model. All such changes are straightforward. Preprocessing programs may be needed to simulate the operation of some proposed projects (e.g. a timeseries filter that simulates Vista Grande diversions into Lake Merced).