J. Simunek et Mt. Vangenuchten, ESTIMATING UNSATURATED SOIL HYDRAULIC-PROPERTIES FROM TENSION DISC INFILTROMETER DATA BY NUMERICAL INVERSION, Water resources research, 32(9), 1996, pp. 2683-2696
Tension disc infiltrometers are becoming increasingly popular devices
for in situ measurement of the unsaturated hydraulic properties of soi
l. Tension infiltration data are generally used to evaluate the parame
ters K-s and alpha in Gardner's exponential model of the unsaturated h
ydraulic conductivity. Either two measurements using different disc di
ameters or measurements with a single disc but using multiple pressure
heads are then used. In this paper we describe a parameter estimation
procedure which combines the Levenberg-Marquardt nonlinear parameter
optimization method involving weighted least squares, with a quasi-thr
ee-dimensional numerical model which solves the variably saturated flo
w equation. By numerical inversion of Richards' equation the unknown p
arameters in van Genuchten's model of the unsaturated soil-hydraulic p
roperties are estimated from observed cumulative infiltration data dur
ing transient water flow. Additional measurements of the pressure head
or water content, as well as a penalty function for constraining the
unknown parameters to remain in some feasible region (Bayesian estimat
ion), can be optionally included into the parameter estimation procedu
re. The problem of optimal sampling design, that is, selecting the bes
t points in space and time for making measurements, is addressed by st
udying the sensitivity of the objective function to changes in the opt
imized hydraulic parameters. We calculate objective functions based on
available cumulative infiltration, pressure head, and water content m
easurements and also on several combinations of these data. The behavi
or of the objective function in three-dimensional parameter space is e
valuated by means of a series of two-dimensional response surfaces. Th
e utility of the parameter estimation procedure is demonstrated using
numerically generated data. The sensitivity of the procedure to differ
ent initial estimates of the model parameters is also discussed.