ABSTRACT

Areas contributing recharge to four well fields in two study sites in southern Rhode Island were delineated on the basis of steady-state groundwater-flow models representing average hydrologic conditions. The wells are screened in sand and gravel deposits in wetland and coastal settings. The groundwater-flow models were calibrated by inverse modeling using nonlinear regression. Summary statistics from nonlinear regression were used to evaluate the uncertainty associated with the predicted areas contributing recharge to the well fields.

In South Kingstown, two United Water Rhode Island well fields are in Mink Brook watershed and near Worden Pond and extensive wetlands. Wetland deposits of peat near the well fields generally range in thickness from 5 to 8 feet. Analysis of water-level drawdowns in a piezometer screened beneath the peat during a 20-day pumping period indicated vertical leakage and a vertical hydraulic conductivity for the peat of roughly 0.01 ft/d. The simulated area contributing recharge for average withdrawals of 2,138 gallons per minute during 2003–07 extended to groundwater divides in mostly till and morainal deposits, and it encompassed 2.30 square miles. Most of a sand and gravel mining operation between the well fields was in the simulated contributing area. For the maximum pumping capacity (5,100 gallons per minute), the simulated area contributing recharge expanded to 5.54 square miles. The well fields intercepted most of the precipitation recharge in Mink Brook watershed and in an adjacent small watershed, and simulated streams ceased to flow. The simulated contributing area to the well fields included an area beneath Worden Pond and a remote, isolated area in upland till on the opposite side of Worden Pond from the well fields. About 12 percent of the pumped water was derived from Worden Pond.

In Charlestown, the Central Beach Fire District and the East Beach Water Association well fields are on a small (0.85 square mile) peninsula in a coastal setting. The wells are screened in a coarse-grained, ice-proximal part of a morphosequence with saturated thicknesses generally less than 30 feet on the peninsula. The simulated area contributing recharge for the average withdrawal (16 gallons per minute) during 2003–07 was 0.018 square mile. The contributing area extended southwestward from the well fields to a simulated groundwater mound; it underlay part of a small nearby wetland, and it included isolated areas on the side of the wetland opposite the well fields. For the maximum pumping rate (230 gallons per minute), the simulated area contributing recharge (0.26 square mile) expanded in all directions; it included a till area on the peninsula, and it underlay part of a nearby pond. Because the well fields are screened in a thin aquifer, simulated groundwater traveltimes from recharge locations to the discharging wells were short: 94 percent of the traveltimes were 10 years or less, and the median traveltime was 1.3 years.

Model-prediction uncertainty was evaluated using a Monte Carlo analysis; the parameter variance–covariance matrix from nonlinear regression was used to create parameter sets for the analysis. Important parameters for model prediction that could not be estimated by nonlinear regression were incorporated into the variance–covariance matrix. For the South Kingstown study site, observations provided enough information to constrain the uncertainty of these parameters within realistic ranges, but for the Charlestown study site, prior information on parameters was required. Thus, the uncertainty analysis for the South Kingstown study site was an outcome of calibrating the model to available observations, but the Charlestown study site was also dependent on information provided by the modeler. A water budget and model-fit statistical criteria were used to assess parameter sets so that prediction uncertainty was not overestimated. For the scenarios using maximum pumping rates at both study sites where the well fields intercepted most of the precipitation recharge that would otherwise have discharged to nearby small streams, results from the probabilistic contributing area indicated that, generally, areas closer to the well fields with shorter traveltimes are associated with higher probabilities and are more likely to coincide with the deterministic contributing area than are areas farther from the well fields with longer traveltimes and associated with lower probabilities. For both the maximum pumping rates and for the average pumping rate at the South Kingstown study site, the deterministic contributing areas generally corresponded to areas associated with high probabilities (greater than 50 percent). For the average pumping rate in the South Kingstown study site, areas associated with low probabilities were not only distant from the well fields but were located where simulated streams in the calibrated model intercepted precipitation recharge, thus indicating that this recharge may instead go directly to a well. That part of the sand and gravel mining operation between the well fields that was not in the deterministic contributing area was in the probabilistic contributing area, including some areas associated with high probabilities. For the maximum pumping rate in the South Kingstown study site, some areas on the opposite side of Worden Pond from the well fields that were not in the deterministic contributing area were in the probabilistic contributing area, but mostly low probabilities were associated with such areas. For the average pumping rate in the Charlestown study site, areas associated with high probabilities were limited to the well-field side of a nearby wetland; the deterministic contributing area on this side of the wetland coincided with this area of high probabilities. For both pumping rates, areas associated with low probabilities extended through the middle of the peninsula toward the mainland; for the maximum pumping rate, the low probability areas included small, isolated areas on the mainland.

First posted on July 29, 2010