Spatial Calibration of Vertical Hydraulic Conductivity below an Estuary

Abstract

The finite-difference model MODFLOW was used to model the flow of meteoric groundwater discharge (MGWD) into the Indian River Lagoon (IRL). The model was calibrated by four methods: (1) estimation of the root-mean-squared error (RMSE); (2) estimation of the Nash-Sutcliffe efficiency (NSE) index; (3) testing the null hypothesis by the two-sided test; and (4) visual comparisons. The statistics for the first three methods were obtained by comparing model-predicted and measured freshwater hydraulic head nodal values at 14 measurement points. Visual comparisons were made by comparing model-predicted and measured freshwater hydraulic head equipotential lines below the entire IRL over a domain approximately 2 km wide. Model validation confirmed that calibration by visual comparison was very accurate. The annual MGWD values obtained by the statistical calibration techniques were compared with the visual calibration results. Acceptable calibration with RMSE values ranging from 0.18–0.31 m underpredicted or overpredicted the annual MGWD by 28–213%. The annual MGWD values were off by 105% even when the two-sided test did not reject the null hypothesis. Acceptable calibration by the NSE statistic yielded excellent results at one transect but underestimated the MGWD by 39% at the other transect. Negative NSE index values always correlated with poor calibrations, although positive values did not always correlate with good calibration. Sensitivity analyses showed that the vertical hydraulic conductivity, Kv, was the most important factor governing MGWD into the IRL because it was orders of magnitude smaller than the lateral hydraulic conductivity. The predominant Kv value was 0.015 m/day, but in some regions Kv was only 0.00015 m/day, indicating that clogging can become much more significant near estuarine beds compared to river beds, perhaps due to negligible currents near estuarine beds. Results also showed that use of a single anisotropy ratio value to represent the entire domain could lead to significant errors.