Search

Summary A priori parameterization approaches that improve our ability to provide reliable hydrologic predictions in ungauged and poorly gauged basins, as well as in basins undergoing change are currently receiving considerable attention. However, such methods are typically based on local-scale process understanding and simplifying assumptions and an increasing body of evidence suggests that hydrologic models that utilize parameters estimated via such approaches may not always perform well. This paper proposes a Maximum Likelihood multi-criteria penalty function strategy for evaluating a priori parameter estimation approaches. We demonstrate the method by examining the extent to which a priori parameter estimates specified for the Hydrology Laboratory’s Research Distributed Hydrologic Model (via a set of pedotransfer functions) are consistent with the optimal model parameters required to simulate the dynamic input–output response of the Blue River basin. Our results indicated that whereas simulations using the a priori parameter estimates give consistently positive flow bias, unconstrained optimization to the response data results in parameter values that are very different from the a priori parameter set. Moreover, although unconstrained optimization performed best (as measured by the calibration criteria), poor hydrograph simulation performance was evident when evaluated in terms of multiple performance statistics not used in the calibration. On the other hand, the multi-criteria compromise solutions provided improved input–output performance in terms of measures not used in calibration, with generally more consistent behavior across calibration and evaluation years, while maintaining physically realistic a priori values for most of the model parameter estimates; adjustments were found to be necessary for only a few key model parameters. [ABSTRACT FROM AUTHOR]

Summary: This paper explores the use of a semi-automated multiple-criteria calibration approach for estimating the parameters of the spatially distributed HL-DHM model to the Blue River basin, Oklahoma. The study was performed in the context of Phase 2 of the DMIP project organized by the Hydrology Lab of the NWS. To deal with the problem of ill conditioning, we employ a regularization approach that constrains the search space using information contained in a priori estimates of the spatially distributed parameter fields developed from soils and other geo-spatial datasets. Unlike the commonly used spatial-multiplier method, our more general approach allows the parameters to depart non-uniformly (to some degree) from the a priori spatial pattern. The approach reduces the number of unknowns to be estimated using historical input–output data from 860 to 35. Two commonly used summary statistics of the model residuals, MSE and MSEL, are used to optimize fitting of the model to both the peaks and the recession periods of the time series data. A signature measure approach is used to select parameter sets that are close to Pareto-optimal in terms of MSE and MSEL, but which provide more consistent representation of the hydrologic behavior of the watershed as summarized by measures derived from the flow duration curve. While the results support the methods used in this analysis and show considerable improvement over the a priori parameter estimates, we find that the basin has some peculiar behaviors (including time non-stationarity) that the HL-DHM model as implemented is not set up to reproduce. [ABSTRACT FROM AUTHOR]

Distributed hydrological models have the potential to provide improved streamflow forecasts along the entire channel network, while also simulating the spatial dynamics of evapotranspiration, soil moisture content, water quality, soil erosion, and land use change impacts. However, they are perceived as being difficult to parameterize and evaluate, thus translating into significant predictive uncertainty in the model results. Although a priori parameter estimates derived from observable watershed characteristics can help to minimize obstacles to model implementation, there exists a need for powerful automated parameter estimation strategies that incorporate diagnostic information regarding the causes of poor model performance. This paper investigates a diagnostic approach to model evaluation that exploits hydrological context and theory to aid in the detection and resolution of watershed model inadequacies, through consideration of three of the four major behavioral functions of any watershed system; overall water balance, vertical redistribution, and temporal redistribution (spatial redistribution was not addressed). Instead of using classical statistical measures (such as mean squared error), we use multiple hydrologically relevant 'signature measures' to quantify the performance of the model at the watershed outlet in ways that correspond to the functions mentioned above and therefore help to guide model improvements in a meaningful way. We apply the approach to the Hydrology Laboratory Distributed Hydrologic Model (HL-DHM) of the National Weather Service and show that diagnostic evaluation has the potential to provide a powerful and intuitive basis for deriving consistent estimates of the parameters of watershed models. [ABSTRACT FROM AUTHOR]

Summary: The mean squared error (MSE) and the related normalization, the Nash–Sutcliffe efficiency (NSE), are the two criteria most widely used for calibration and evaluation of hydrological models with observed data. Here, we present a diagnostically interesting decomposition of NSE (and hence MSE), which facilitates analysis of the relative importance of its different components in the context of hydrological modelling, and show how model calibration problems can arise due to interactions among these components. The analysis is illustrated by calibrating a simple conceptual precipitation-runoff model to daily data for a number of Austrian basins having a broad range of hydro-meteorological characteristics. Evaluation of the results clearly demonstrates the problems that can be associated with any calibration based on the NSE (or MSE) criterion. While we propose and test an alternative criterion that can help to reduce model calibration problems, the primary purpose of this study is not to present an improved measure of model performance. Instead, we seek to show that there are systematic problems inherent with any optimization based on formulations related to the MSE. The analysis and results have implications to the manner in which we calibrate and evaluate environmental models; we discuss these and suggest possible ways forward that may move us towards an improved and diagnostically meaningful approach to model performance evaluation and identification. [Copyright &y& Elsevier]