Multicriteria parameter estimation for models of stream chemical composition

Abstract

[1] The inability to develop an accurate and precise parameter estimation method for catchment hydrochemical models has been a persistent problem. We investigate the use of multicriteria calibration techniques and the selection of the criteria as a first step in solving this problem. We applied a multicriteria search algorithm to the Alpine Hydrochemical Model (AHM) of the Emerald Lake watershed, Sequoia National Park, California. A total of 21 chemical and hydrologic criteria were available for determining model performance. Four subsets of these criteria were selected for the multicriteria analysis using three different methods. The first set used the four least correlated observations of stream chemical composition. A second set of criteria was determined by using the four species with the least correlated root mean square error criteria values (r < 0.05 for each pair). Finally, two sets were chosen using the results of a multicriteria sensitivity analysis. The most accurate and precise results were observed using criteria selected using results from a sensitivity analysis, with the correlation analyses being a poor method for selecting criteria. This set of criteria emphasizes attributes of the model structure, the observations, and our understanding of the processes influencing watershed hydrology and water quality. Also, our results gave improved estimates of several hydrologic and biogeochemical processes in addition to identifying a flaw in the current representation of mineral weathering within the AHM, as applied to the Emerald Lake watershed.

Multiobjective Automatic Parameter Calibration of a Hydrological Model

“…The two objectives were simultaneously optimized; they were not aggregated in this study. In the field of hydrological model calibration, the NSGA-II, Multiobjective Complex Evolution (MOCOM) algorithm and the Multiobjective Shuffled Complex Evolution Metropolis (MOSCEM) algorithm are widely used [8,[46][47][48][49][50][51][52][53][54][55][56][57][58][59]. A set of Pareto optimal solutions (non-dominated parameter sets) was obtained from the model.…”

Multidecadal hydrochemical response of a Sierra Nevada watershed: sensitivity to weathering rate and changes in deposition

“…Previous applications of the AHM have focused on single year or 2-year simulations (e.g. Meixner et al, 2002), and while such simulations are adequate for predicting the current episodic response of these watersheds (Wolford and , they cannot be used to address chronic and episodic acidification in the Sierra Nevada over decadal time periods. Simulations need to take into account the widely variable precipitation climate of the Sierra Nevada (Cayan et al, 1998) as well as the role of increases in atmospheric deposition and their impact on water quality through time (Cosby et al, 1985a).…”

Fit-for-purpose analysis of uncertainty using split-sampling evaluations

“…The trade-off between different fitting criteria may lead to a model bias in the calibration period, which happened in the case presented here. Such trade-offs are caused by model structural errors (Meixner et al, 2002;Gupta et al, 2003), and have to be solved by a better mathematical representation of the system. (e) Observations are typically considered as being correct.…”

“…(c) The model structure and hypothesis (the process equations) are correct and represent the real world. Model errors are considered by using multiple models in a Bayesian framework (Hoeting et al, 1999;Montanari & Brath, 2004), by regionalized sensitivity analysis (Osidele & Beck, 2001), the GLUE methodology (Beven & Binley, 1992) or with a Pareto analysis (Meixner et al, 2002;Gupta et al, 2003). (Freer et al, 2003) and a necessity to compromise as a consequence.…”

“…It is well known that hydrological models, particularly those of the rainfall-runoff process and even more so for models of water quality, are not perfect models and thus the assumption that the model is correct does not hold for the application of hydrological models (e.g. see Beven, 1993;Mroczkowski et al, 1997;Boyle et al, 2001;Meixner et al, 2002).…”