http://nova.newcastle.edu.au/vital/access/services/Feed ${session.getAttribute("locale")} 5 http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:10966 Despite the widespread use of conceptual hydrological models in environmental research and operations, they remain frequently implemented using numerically unreliable methods. This paper considers the impact of the time stepping scheme on model analysis (sensitivity analysis, parameter optimization, and Markov chain Monte Carlo-based uncertainty estimation) and prediction. It builds on the companion paper (Clark and Kavetski, 2010), which focused on numerical accuracy, fidelity, and computational efficiency. Empirical and theoretical analysis of eight distinct time stepping schemes for six different hydrological models in 13 diverse basins demonstrates several critical conclusions. (1) Unreliable time stepping schemes, in particular, fixed-step explicit methods, suffer from troublesome numerical artifacts that severely deform the objective function of the model. These deformations are not rare isolated instances but can arise in any model structure, in any catchment, and under common hydroclimatic conditions. (2) Sensitivity analysis can be severely contaminated by numerical errors, often to the extent that it becomes dominated by the sensitivity of truncation errors rather than the model equations. (3) Robust time stepping schemes generally produce “better behaved” objective functions, free of spurious local optima, and with sufficient numerical continuity to permit parameter optimization using efficient quasi Newton methods. When implemented within a multistart framework, modern Newton-type optimizers are robust even when started far from the optima and provide valuable diagnostic insights not directly available from evolutionary global optimizers. (4) Unreliable time stepping schemes lead to inconsistent and biased inferences of the model parameters and internal states. (5) Even when interactions between hydrological parameters and numerical errors provide “the right result for the wrong reason” and the calibrated model performance appears adequate, unreliable time stepping schemes make the model unnecessarily fragile in predictive mode, undermining validation assessments and operational use. Erroneous or misleading conclusions of model analysis and prediction arising from numerical artifacts in hydrological models are intolerable, especially given that robust numerics are accepted as mainstream in other areas of science and engineering. We hope that the vivid empirical findings will encourage the conceptual hydrological community to close its Pandora's box of numerical problems, paving the way for more meaningful model application and interpretation. 2012-06-25T06:20:50.757Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:10481 Catchments that do not behave in the way the hydrologist expects, expose the frailties of hydrological science, particularly its unduly simplistic treatment of input and model uncertainty. A conceptual rainfall–runoff model represents a highly simplified hypothesis of the transformation of rainfall into runoff. Sub-grid variability and mis-specification of processes introduce an irreducible model error, about which little is currently known. In addition, hydrological observation systems are far from perfect, with the principal catchment forcing (rainfall) often subject to large sampling errors. When ignored or treated simplistically, these errors develop into monsters that destroy our ability to model certain catchments. In this paper, these monsters are tackled using Bayesian Total Error Analysis, a framework that accounts for user-specified sources of error and yields quantitative insights into how prior knowledge of these uncertainties affects our ability to infer models and use them for predictive purposes. A case study involving a catchment with an apparent water balance anomaly (a hydrological monstrosity!) illustrates these concepts. It is found that, in the absence of additional information, the rainfall–runoff record is insufficient to explain this anomaly – it could be due to a large export of groundwater, systematic overestimation of catchment rainfall of the order of 40%, or a conspiracy of these factors. There is “no free lunch” in hydrology. The rainfall–runoff record on its own is insufficient to decompose the different sources of uncertainty affecting calibration, testing and prediction, and hydrological monstrosities will persist until additional independent knowledge of uncertainties is obtained. 2012-03-21T04:40:02.428Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:1276 Parameter estimation in rainfall-runoff models is affected by uncertainties in the measured input/output data (typically, rainfall and runoff, respectively), as well as model error. Despite advances in data collection and model construction, we expect input uncertainty to be particularly significant (because of the high spatial and temporal variability of precipitation) and to remain considerable in the foreseeable future. Ignoring this uncertainty compromises hydrological modeling, potentially yielding biased and misleading results. This paper develops a Bayesian total error analysis methodology for hydrological models that allows (indeed, requires) the modeler to directly and transparently incorporate, test, and refine existing understanding of all sources of data uncertainty in a specific application, including both rainfall and runoff uncertainties. The methodology employs additional (latent) variables to filter out the input corruption given the model hypothesis and the observed data. In this study, the input uncertainty is assumed to be multiplicative Gaussian and independent for each storm, but the general framework allows alternative uncertainty models. Several ways of incorporating vague prior knowledge of input corruption are discussed, contrasting Gaussian and inverse gamma assumptions; the latter method avoids degeneracies in the objective function. Although the general methodology is computationally intensive because of the additional latent variables, a range of modern numerical methods, particularly Monte Carlo analysis combined with fast Newton-type optimization methods and Hessian-based covariance analysis, can be employed to obtain practical solutions. 2012-03-12T07:33:42.738Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:8931 Conceptual rainfall-runoff models (CRRs) are widely used to simulate catchment runoff. However, assigning values to their parameters at ungauged catchments remains problematic with the prospect of large systematic errors. To counter this there have been many attempts to regionalize CRR parameters using catchment characteristics. However, they typically have low accuracy. One of the reasons for this is that rainfall errors introduce random biases in calibrated CRR parameters, thus confounding any relationship between CRR parameters and catchment characteristics. This study seeks to circumvent this problem by regionalizing flow duration curves (FDCs) and calibrating the CRR model to the regionalized FDC. The potential of the method is explored using the SIMHYD model for a case study based on 55 catchments in eastern NSW. Strong linear (R2>0.80) relationships were found between neighbouring FDC quantiles. In addition, strong relationships were found between catchment characteristics and high-flow FDC quantiles with exceedance probabilities of 0.1 or less. The SIMHYD rainfall-runoff model was calibrated to daily runoff time series and observed and regionalized FDCs. Calibration to the regionalized FDC produced results comparable to calibration to the observed FDC. However, in two out of six cases, the results were poor. This suggests calibration to regionalized FDCs has promise but more work is needed to better understand its performance and understand why it can sometimes fail badly. 2011-09-13T01:01:12.642Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:308 In order to assess performance and to improve predictions, land surface models are routinely calibrated against measurements of either latent heat or sensible heat fluxes. Generally, little regard is given to the multi-output nature of these models, resulting in a model evaluation that is inherently biased towards the calibration variable. In this paper, an assessment strategy that accounts for multiple outputs is explored and an examination of incorporating alternative sources of information to assess performance is undertaken. The benefits of such a multi-objective calibration framework are illustrated through comparison with traditional single objective calibration. Results indicate that combining different observation data streams for calibration purposes assists in producing a more robust process model and provides improved surface flux predictions. Further, the utility of using correlated, if not commensurate, sources of data, is demonstrated through analysis of a time series of surface temperature measurements. (C) 2004 Elsevier B.V. All rights reserved. 2010-04-27T05:49:05.176Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:3842 The PUB initiative aims to integrate knowledge of hydrological processes to provide the best hydrological characterization of ungauged basins. This requires the integration of models and methods to achieve those objectives. In this paper, recent modelling activities are reviewed, with the aim of demonstrating potential application to ungauged basins. First, the development and testing of process-oriented hydrological models is presented. Examples are shown of the utility of remote sensing in conditioning hydrological model parameters at the catchment scale. Subsequently, a Bayesian error-sensitive model calibration scheme (BATEA) is presented. This scheme acknowledges that rainfall errors propagate and persist in hydrological models, corrupting the parameter estimates. It is shown that BATEA offers parameter estimates unbiased by error in rainfall data. BATEA will be applied to multiple models across a range of basins using MOPEX and Australian data. As regionalization relationships will be derived through unbiased model parameter estimates, it is hoped that stronger relationships between catchment characteristics and model parameter values may be identified, permitting improved model performance in ungauged basins. Finally, multi-decadal climate variability across New South Wales is demonstrated and an ENSO-based mechanism is elucidated. Such understanding of climate/hydrology interfaces offers a greater insight into hydrological risk assessment at different temporal scales and may easily be coupled to regionalized models for ungauged basins at continental scales. 2010-04-27T05:32:55.472Z ]]> http://nova.newcastle.edu.au/vital/access/manager/Repository/uon:3330 Conceptual hydrological modelling has traditionally been plagued by calibration difficulties due to the roughness and complex shape of objective functions. These problems led to the abandonment of powerful classical analysis methods (Newton-type optimisation, derivative-based uncertainty analysis) and have motivated extensive research into nonsmooth optimisation and even new parameter estimation philosophies (e.g. GLUE). This paper shows that some of these complexities are not inherent features of hydrological models, but are numerical artefacts due to model thresholds and poorly selected time stepping schemes. We present a numerically robust methodology for implementing conceptual models, including rainfall-runoff and snow models, that ensures micro-scale smoothness of objective functions and guarantees macro-scale model stability. The methodology employs robust and unconditionally stable time integration of the models, complemented by careful threshold smoothing. A case study demonstrates the benefits of these techniques. 2010-04-27T05:00:29.752Z ]]>