Welcome to NOAH
Saltwater Upconing Prediction
Provincetown, Massachusetts

Problem Definition

Provincetown, Massachusetts, located near the Atlantic Ocean on the Cape Cod peninsula, has experienced saltwater upconing due to high groundwater extractions for community water supply. Artificial neural networks were developed to predict highly dynamic specific conductance variations in a monitoring well due to vertical displacement of the freshwater-saltwater lens under variable pumping and weather conditions. The single monitoring well is located in close proximity to a production well used to supply the community with drinking water. Figure 1 depicts the study area and geological cross section of the aquifer and well placement. The ANN models were also used to conduct a sensitivity analysis to identify important relationships between the input predictor variables (e.g. pumping rates and precipitation) and the corresponding conductivity values.

Results

ANNs were developed to predict specific conductance one month, two months, and three months ahead. In addition, a single ANN was developed to predict conductivity over an extended 46-month prediction period using monthly time steps or stress periods. Model accuracy was compared against measured/interpolated conductivity values and a linear regression model (LR), and in general, excellent predictive performance with the ANN models was achieved. For example, although the average percent change of conductance over 90-day prediction periods was 39%, the absolute mean prediction error achieved with the ANN model was only 1.1%. Figure 2 compares measured versus ANN predicted conductance values for the 3 month prediction period.

As shown in Figure 3, for the 46-month extended period simulation, where the models reinitialized themselves with their previous monthly prediction, the ANN accurately reproduced available measured/interpolated values, accurately reproducing measured conductance values measured at the end of the prediction period. In contrast, the LR model did not provide accurate predictions over the extended prediction period, significantly underestimating conductance values over the final 1.5 years. Note that conductance values were not available for comparison for all time periods, but the valleys and peaks correspond with higher and lower pumping periods (i.e. seasons).

Sensitivity analyses conducted with the ANN models demonstrate that while weather is somewhat important for predicting conductance, pumping rates and initial conductance values (i.e. measured at beginning of prediction period) are the most important for accurately predicting final conductance values. In addition, the effect of weather variables such as temperature and precipitation become more important over longer prediction periods (e.g. 90 days), which physically makes sense as their individual and combined effects (e.g. groundwater recharge) become more pronounced over longer time periods.

The ANN could be used to help optimize pumping of the wellfield. Using the ANN derived state-transition equations as part of the constraint set, formal optimization methodology can be used to maximize total pumping without violating imposed water quality constraints in the monitoring well. The higher predictive accuracy achieved with the ANN that with the traditional physical-based model would result in identification of superior management solutions that would protect the drinking water supply against water quality degradation.

A more detailed overview of this work can be found in the following journal article:

Coppola, E. C. McLane, M. Poulton, F. Szidarovszky, and R. Magelky (2005). Predicting Conductance Due To Upconing Using Neural Networks, Journal of Ground Water, 43, no 6: 827-836.