WRIR 00-4094

You can **DOWNLOAD THIS REPORT (1,485 KB)
** in **
Portable Document Format** (PDF)

The **
Adobe PDF Reader** program is available for free from Adobe.

### Yobbi, D.K., 2000, Application of Nonlinear Least-Squares Regression to
Ground-Water Flow Modeling, West-Central Florida: Water-Resources Investigations
Report 00-4094, 58 p.

ABSTRACT:

A nonlinear least-squares regression technique for estimation of ground-water
flow model parameters was applied to an existing model of the regional aquifer
system underlying west-central Florida. The regression technique minimizes
the differences between measured and simulated water levels. Regression
statistics, including parameter sensitivities and correlations, were calculated
for reported parameter values in the existing model. Optimal parameter values
for selected hydrologic variables of interest are estimated by nonlinear
regression. Optimal estimates of parameter values are about 140 times greater
than and about 0.01 times less than reported values. Independently estimating
all parameters by nonlinear regression was impossible, given the existing
zonation structure and number of observations, because of parameter
insensitivity and correlation. Although the model yields parameter values
similar to those estimated by other methods and reproduces the measured water
levels reasonably accurately, a simpler parameter structure should be
considered. Some possible ways of improving model calibration are to:
(1) modify the defined parameter-zonation structure by omitting and/or
combining parameters to be estimated; (2) carefully eliminate observation data
based on evidence that they are likely to be biased; (3) collect additional
water-level data; (4) assign values to insensitive parameters, and (5) estimate
the most sensitive parameters first, then, using the optimized values for these
parameters, estimate the entire data set.