IMPROVED GROUNDWATER MODEL CALIBRATION AND EVALUATION USING INTERPOLATED DISTRIBUTION OF OBSERVED DATA

Lack of sufficient field data adds difficulties to the calibration and evaluation of predictive accuracy of groundwater models. This paper investigates the use of interpolation of hydraulic heads collected at limited measurement points below Indian River Lagoon to aid improve calibration and evaluation of groundwater models at two vertical-plain transects across the lagoon. Additional nodal head values were extracted from the interpolated head distributions. Groundwater models of the lagoon were developed using SEAWAT code. Calibration objective was to minimize hydraulic head residuals, and make sure that simulated head distributions compare visually well to the interpolated distributions. Calibration results were evaluated using: 1) root mean square error (RMSE), 2) Nash-Sutcliffe efficiency (NSE) index, 3) two-sided t-test, and 4) visual comparisons of measured and simulated head distributions. Statistical analysis results with and without the additional interpolation-based head values were compared. RMSE results were found to be meaningless since its magnitudes were steadily small even for poorly calibrated models. NSE was not able to capture the improvement in model calibration when comparisons were made without the additional interpolationbased head data. However, incorporating those points in the analysis, improved NSE statistical results. Similarly, the two-sided t-test improved in evaluating model calibration when used with the additional observations. Interpolation of observed data is highly recommended to obtain complete hydraulic head distributions prior to conducting statistical analyses. Interpolation not only improves statistical analyses but also provides additional calibration tool by visual comparison