Laserfiche WebLink
<br />provide a discussion of the error measures as they relate to the steady-state and transient model <br />calibration. In summary, the process for model calibration involved: <br /> <br />. Phase 1: Steady-State Model Only - First a subset of the parameters was adjusted using the <br />automated model calibration facilities as part of MODFLOW2000, and subsequently the <br />calibration was adjusted via manual trial and error adjustments to minimize (2). The <br />automated model calibration was undertaken to develop a preliminary best-fit steady-state <br />model. This model was further manually refined with localized adjustments to the <br />parameters. For steady-state calibration, the weighting in (2) was a=l and all other weights <br /> <br />equal zero. Calibration of the steady-state model proceeded for over 50 model runs until the <br />pre-development SSE was minimized. <br /> <br />. Phase 2: Initial Transient Calibration - After arriving at a preliminary steady-state best-fit, <br />the model was run in transient mode to simulate aquifer response from pre-development <br />conditions through 1991. Ideally, a good unique steady-state best-fit model would require <br />minimal further adjustment of constant parameters (e.g., recharge) and fixed properties (e.g., <br />hydraulic conductivity) in transient calibration; only properties independent of the steady <br />flow equation (e.g., storage) and/or time-dependent parameters (e.g., pumping rates) would <br />be adjusted in transient calibration. Unfortunately, the initial transient simulations quickly <br />demonstrated that further parameter adjustments would be required to develop a unique <br />model that does a good job at predicting current heads and drawdown rates, as well as pre- <br />development steady-state conditions. Over 150 additional model simulations (each with a <br />unique set of model input parameters, run for both steady-state and historical transient <br />conditions) were evaluated during the transient calibration process, and Figure 3.5(a) <br />illustrates the record for each of the model error measures during this process. Figure 3.5(a) <br />was developed such that Iteration I (labeled "trans 1" in the figure) represents the weighted <br />errors (2) for the final model from the steady-state calibration process. The weights for (2) <br />were computed to yield the following weighted SSEs for Iteration I: <br /> <br />aSSE p"-d,, = 1.0 <br />pSSE HI991 = 2.0 <br />fiSSE DD".O-I991 = 3.0 <br /> <br />and m and t; were assigned a value of zero initially. Once a best-fit was obtained based on <br />this error-weighting scheme (Fig. 3.5a), the period from 1991 to 2000 was simulated as a <br />verification as described below. <br /> <br />. Phase 3: Verification and Transient Calibration Refinement At the completion of the <br />transient calibration process for simulations through 1991, a verification model run was <br />performed to assess the predictive capability of the model. The verification model consisted <br />of a transient model run from 1991 through 2000. For the verification model, simulated <br />heads in 2000 and drawdowns between 1991 and 2000 were evaluated in comparison to <br />observations. The first verification model run indicated that further parameter refinements <br /> <br />Page ]-14 <br /> <br />(3) <br /> <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />4 <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />41 <br />41 <br />41 <br />4 <br />t <br />4 <br />4 <br />41 <br />4 <br />41 <br />41 <br />41 <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />t <br />4 <br />