You are on page 1of 4

# 3.

3 Model validation A model validation was performed on the Steady State model to evaluate the significance of the values that were predicted. The model used a RungeKutte method with a time step of 2 months. The techniques applied to validate the model are considered a standard for evaluating whether modeled values correlate with measured values. There must be three criteria satisfied for validation of the model. The three criteria include; Residuals of the predicted concentrations must be randomly distributed, the slope of the regression line must be significantly close to 1, and the correlation coefficient must be statistically significant at the 95%confidence level. The first statistical technique applied was a linear regression analysis. Figure __ below shows a linear regression analysis. For the model to be validated the slope of the line must be significantly close to 1. Measur ed (mg/m 3) 50 58.9 43 55 19 36.1 30 3 25 Calcula ted (mg/m 3) 44.7 38.9 37.3 36.1 31.4 31.3 30.3 30.2 30.0 Chi Square ^2 0.62168 135 10.2730 363 0.86983 193 9.87419 205 4.90862 592 0.75237 246 0.00235 591 24.4924 905 0.83430

Year 9/15/19 70 9/15/19 72 9/15/19 73 9/15/19 74 9/15/19 85 9/15/19 86 9/15/19 95 9/15/19 96 9/15/19

99 9/15/20 00 9/15/20 01

373 0.88460 24.8 29.9 876 1.05128 35.5 29.9 586 Total 54.5647 ^2 847

## Figure __: Comparison of Predicted Values to Measured Values

Using a linear least squares regression to fit a trendline to the measured data, the data is shown as non-linear due to a low r2 value of -1.644. Since the slope of the line is 0.702 (below 1) the model fails one step in the validation. A slope less than 1 tells us that the model is under predicting concentrations.

## Figure ___: Comparison of Predicted Values to Measured Values

Chi Square analysis of the measured results versus the modeled results shows that the actual results differ too greatly from the calculated results for the two to be proven correlated. A sum of the chi squares results in a value of 54.56. This value isnt even less than the Chi Square value for a .1% confidence level (29.59 Chi Square value). The biggest factor in this large value is an outlier concentration of 3 mg/m3 in the measured data. Without this outlier, the Chi Square sum drops down to ~30 which is a much more reasonable number, but still larger than any reasonable confidence level.

These results indicate that the model does not do a very good job of approximating the actual concentrations in the records. A plot of the residuals was the second criteria performed for validation. For the criteria to be satisfied the residuals most be randomly distributed. As seen in Figure __ above, the residuals are not randomly distributed. Since the values are not randomly distributed the model fails the second criteria of validation. The third criteria were not performed because the model is assumed to be invalidated. The Runge-Kutte model was not used to predict concentrations when selecting different remedial options.