Parameter optimisation and uncertainty assessment for large-scale streamflow simulation with the LISFLOOD model

Abstract

Summary This work addresses the calibration of the distributed rainfall-runoff model LIS-FLOOD and, in particular, the realistic quantification of parameter uncertainty and its effect on the prediction of river discharges for large European catchments. LISFLOOD is driven by meteorological input data and simulates river discharge in large drainage basins as a function of spatial information on topography, soils and land cover. Even though LISFLOOD is physically based to a certain extent, some processes are only represented in a lumped conceptual way. As a result, some parameters lack physical basis and cannot be directly inferred from quantities that can be measured. In the current LISFLOOD version five parameters need to be determined by calibration. We employ the Shuffled Complex Evolution Metropolis (SCEM-UA) global optimisation algorithm to automatically calibrate the model against daily discharge observations. The resulting posterior parameter distribution reflects the uncertainty about the model parameters after taking into account the discharge observations, and forms the basis for making probabilistic flow predictions. To overcome the computational burden the optimisation has been implemented using parallel computing. As an illustrative example, we demonstrate the methodology for the Meuse catchment upstream of Borgharen, covering approximately 21,000 km 2 . Results demonstrate the capabilities of the SCEM-UA algorithm to efficiently evolve to the target posterior distribution and to identify, except for the lower groundwater zone time constant, the LISFLOOD calibration parameters using daily discharge observations. It should be noted that the posterior parameter distributions are based on the assumption of independent errors. Analysis of the residuals revealed there is a strong evidence to reject this assumption. This will impact on the parameter posterior distributions and also lead to underestimation of the prediction limits. ª KEYWORDS

Bayesian calibration of terrestrial ecosystem models: a study of advanced Markov chain Monte Carlo methods

“…10c indicates that the errors are significantly correlated at a lag of 4, which violates the independence assumption. This violation has been reported in several time-series data models, such as the TEM in Ricciuto et al (2008), the rainfall-runoff model in Feyen et al (2007), and the groundwater reactive transport model in Lu et al (2013). The correlated errors are likely to be observed in models where systematic model errors exist like the DALEC model in this study.…”

Model-aided quantification of dissolved carbon and nitrogen release after windthrow disturbance in an Austrian karst system

“…For that reason we have chosen the Shuffled Complex Evolution Metropolis (SCEM) algorithm (Vrugt et al, 2003), which proved to be capable of exploring high-dimensional optimization problems (Fenicia et al, 2014;Feyen et al, 2007;Vrugt et al, 2006). As performance measure we used the Kling-Gupta efficiency (KGE; Gupta et al, 2009).…”

Streamflow Modelling: A Primer on Applications, Approaches and Challenges

“…Another common approach is to analyze the statistical properties of past forecast errors in order to determine error bounds around a single prediction (e.g., Montanari and Brath, 2004;Feyen et al, 2008;Weerts et al, 2011). By the same token, empirical hydrological modelling methods may directly generate confidence bounds, an obvious example being the standard error of a linear regression model.…”