| Popis: |
Water infiltration is a driving force influencing crop growth, soil erosion, and chemical leaching processes. Knowledge of the relative precision and accuracy of infiltration models is needed for best characterization of the infiltration parameters. The two-parameter GreenAmpt and Philip, three-parameter Horton, Mezencev, Swartzendruber, and Parlange et al., and four-parameter Barry et al. infiltration models were compared for their precision and accuracy of estimated parameter confidence intervals using simulated infiltration reference data. To account for potential levels of uncertainty, three levels of measurement error were included using a Monte Carlo analysis. Reference data were generated for a clay and a sandy loam soil using an adaptive-grid finite-element code. Results show that extending the measurement period provided parameter estimates with higher confidence, a more precise estimate of that confidence, and better defined minima in the objective function. The empirical Horton model resulted in the worst fits due to model bias, which also prevented estimation of parameter uncertainty for this model. The semianalytical Swartzendruber and the physically based Parlange et al. and Barry et al. models provided the best fits. Considering all selected criteria, the Swartzendruber model was a reasonable compromise under the conditions imposed in this study. M ANY APPROACHES have been presented to solve the problem of vertical infiltration of water into a homogeneous, semi-infinite soil. Youngs (1995) reviewed the historic development of infiltration theory including the classic solutions based on the Richards equation. Kutflek and Nielsen (1994, p. 140-159) presented a comprehensive review of analytical and empirical solutions. Typically, such parameterized solutions are fitted to measured infiltration data. The optimized parameters serve as a convenient, condensed description of the infiltration process. Moreover, model parameters can be used for predictive purposes, and physically based infiltration models allow estimation of soil hydraulic properties. Prediction uncertainty, inherent in model simulations, needs to be estimated and reported. It is caused by errors due to violation of assumptions implicit in the model and by uncertainty in the model parameters. When obtaining parameters by fitting measured data, the error of the fitting model will in general be unknown, but the parameter uncertainty as caused by measurement errors can be estimated using confidence intervals. This uncertainty estimate is derived from the objective function in the vicinity of the optimized parameter set, and is statistically accurate for linear fitting models with zero model error and independent measurements of known uncertainty (Press et al., 1992, p. 663). In con |
Plný text