Identification and uncertainty analysis of soil water retention models using lysimeter data

Abstract

The water retention characteristic is one of the fundamental properties that control water movement in unsaturated soils. Because of the complexity and heterogeneity of soil properties, variability and uncertainty of this property has long been a key issue in understanding the relationship between water content and matric pressure. Although considerable progress has been achieved in understanding spatial and local variability, discussions have primarily been based on deterministic analyses. Misleading conclusions can result since the derived variability cannot often be discriminated from uncertainties in model parameters, structure, and observed data. Using a combined deterministic and stochastic analysis, this paper addresses the variability of the retention characteristic in association with the uncertainty of water retention models. Calibrated against two sets of lysimeter data observed over 2 years, the analysis yields insights into the physical changes in the system, and model performance is significantly improved when these effects are explicitly included. sessment of model structures and uncertainties becomes essential in model identification. Although considerable research has been devoted to the development of various water retention models, estimation of spatial variability, derivation of model parameters from soil texture [McCuen et al., 1981], and interpretation of the physical meaning of the empirical parameters, issues of model uncertainty and identifiability have not been simultaneously and properly discussed. Consequently, misleading conclusions can easily result, in particular, when the model is applied to a range of states that is not covered by calibration data. For instance, when the model is applied deterministically to a soil condition for which it is not calibrated, unidentifiable parameters may become sensitive. Model structural uncertainty can be misinterpreted as spatial variability in a heterogeneous geological medium when large-scale heterogeneity is derived from sample scale data [Desbarats, 1995]. Milly [1987] showed that multiple minima exist in the objective functions used to estimate optimally the parameters of water retention models, a clear indication of lack of parameter identifiability. Thus uncertainty can pose both practical and more fundamental problems in the application of water retention models. Currently, calibration of water retention models is mostly carried out with data from drainage experiments, either through laboratory analysis of field samples, for which a relatively wide range of data is available [Ferrand and Celia, 1992], or in situ by simultaneous measurement of soil water tension and volumetric water content [Hopmans, 1987]. Although model parameters can be better identified from laboratory data, the results are often quite different from measurements at a field site [Wierenga et al., 1991; Eching et al., 1994] because of the differences in the conditions under which the model is calibrated (in the laboratory) and the model is applied (in the field). Howeve...

Uncertainty in water resource model parameters used for climate change impact assessment

“…Bultot et al, 1988Bultot et al, , 1992Wilby et al, 1994;Arnell and Reynard, 1996;Sefton and Boorman, 1997;Jones, 1999, 2002;Arnell, 2003;Diaz-Nieto and Wilby, 2004). However, this family of models has well-known limitations with respect to parameter identifiability, parameter stability, predictive uncertainty and equifinality of different model structures and parameter sets (Wheater et al, 1986;Beven and Binley, 1992;Beven, 1993;Chen and Wheater, 1999;Beven and Freer, 2001;Wagener et al, 2001;Niel et al, 2003;Sincock et al, 2003).…”

Application of an Uncertainty Analysis Approach to Strategic Environmental Assessment for Urban Planning

“…Monte Carlo approaches have been widely used in environment modeling − . The Monte Carlo methods provide a computable way to approach the reality by generating a large number of samples according to given probability distribution.…”