WSP09428

• File Number: 8449.926
• Description: South Metro Water Supply Study
• State: CO
• Basin: South Platte
• Date: 2/1/2004
• Author: SMWSS Board
• Title: South Metro Water Supply Study
• Water Supply Pro - Doc Type: Publication

-

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
,
.
.
.
.
.
It
.
It
.
It
.
.
.
.
.
,
,
It
It
It
It
It
It
-
,
,
a

This certainly impacted our ability to develop a unique steady-state model, and ultimately led to
the need to perform simultaneous steady-state and transient calibration as described below.

3.2.2.2 Model Error Measures - Model fit to each of the targets was evaluated as the sum of
the squared errors (SSE) between the model predictions and the observations:
N""
SSE, = L(Y; -yy
J""]

(I)

where i is an index to represent the five separate calibration targets (pre-development heads,
1991 heads, 2000 heads, drawdown from pre-development conditions (1880) to 1991, and
drawdown between 1991 and 2000), and Yj and >J' are the modeled and observed values of the
target at location j.

As a final measure of how well a particular model (as characterized by a unique set of model
input parameters) fit the observations, a weighted sum of the above calibration target SSEs was
developed.
SSG.,al = aSSEpn-dev + fES~1I991 + mSSE,nooo + ifJSSE,;D188i>-1991 + /f)SE,;D1 99 1-2000 (2)

where a, P, llJ, 1jJ, and q are the weights assigned to each of the model error components. In the
initial transient calibration, the SSE for the 2000 heads was not considered to allow them to be
used subsequently for verification runs of the calibrated model. In model calibration parlance,
equation (2) is typically referred to as the model calibration objective function. For every model
run, the objective function (or anyone of its individual error components) can be computed, and
minimized (2) to identify the best model for predicting future aquifer conditions.

3.2.2.3 Predevelopment I Steady-State Calibration - Throughout the steady-state calibration
phase, and in the early stages of the transient calibration phase, steady-state head targets (pristine
heads) provided by the SEO as part of the SB-74 model package were used. These pristine head
values were derived from water level contour maps developed from the Hydrological Atlas series
by Robson and colleagues, and the SEO modelers interpolated from these maps head values for
every cell in the model domain. (Reference 3.2)These inferred heads at every cell were the
targets used in the steady-state calibration phase.

3.2.2.4 Transient Calibration - The 1991 and 2000 heads were taken from the annual Denver
Basin water level reports published by the Colorado SEO. Based on discussions with the ground
water consultants of the South Metro water providers, the rate of drawdown between 1991 and
2000 in their vicinity is one of the most important model performance measures to use for
ensuring that the model provides a robust and defensible tool for predicting bedrock aquifer
response to projected future pumping. The underlying rationale is that if the model can
accurately predict current heads and drawdown rates, then it will provide the best possible
estimate for when the aquifers transition from confined to unconfined conditions (arguably one
of the most pressing questions for the water providers). For this reason, in the transient
calibration process the recent (1991) and current (2000) heads and the 1991 to 2000 drawdown
rates were given the highest weights in (2).

3.2.2.5 Model Calibration Chronology - Section 3.2.1 identified model parameters adjusted
during the calibration process, and the four preceding subsections (3.2.2.1 through 3.2.2.4)

Page 3-t3