Reliability of Semiarid Flash Flood Modeling Using Bayesian Framework

Meteorological Applications

et al. 2017

A case study examining ensemble forecasts of semi‐arid seasonal precipitation is presented. The focus is on computing an appropriate correlation between large‐scale climate predictors and seasonal precipitation over a long‐term forecast period (1967–2009) for a semi‐arid catchment in Iran. Potential predictors of the dominant precipitation modes were identified from several large‐scale climate features using principal component analysis. Linear regression together with two nonlinear models, the adaptive neuro‐fuzzy inference system (ANFIS) and the multi‐layer perceptron, was applied to forecast seasonal ensemble precipitation time series. The analysis suggests that seasonal precipitation is statistically aligned with the predictor's variability. An ensemble forecast of spring precipitation modes showed a stronger correlation with the preceding season (winter predictors) in the ANFIS algorithm. The potential effect of climate predictors during the spring may lead to severe and longer hydrological extremes especially when they are out of phase or coincident. These results highlight that skilful prediction of semi‐arid spring precipitation may be possible using winter predictors and a nonlinear ensemble forecast model.

Section: Introductionmentioning

confidence: 99%

An ensemble forecast of semi‐arid rainfall using large‐scale climate predictors

Choubin

Malekian

Meteorological Applications

et al. 2017

“…There has been a continued debate within the hydrological community regarding to the choice of the most suitable approach for uncertainty assessment to capture different sources of error and uncertainty (Schoups and Vrugt, ; Beven et al ., ; Laloy and Vrugt, ; Pourreza‐Bilondi and Samadi, ; Pourreza‐Bilondi et al ., ). As with all methods of uncertainty analysis, our methods have their own limitations and other approaches could have been adopted, such as the use of multi‐objective algorithm (Vrugt and Robinson, ; Sadeghi‐Tabas et al ., ,b), and MCMC algorithm based on GL (generalized log‐likelihood) function (Schoups and Vrugt, ; Laloy and Vrugt, ).…”

Section: Summary and Discussionmentioning

confidence: 97%

“…In recent decades, various mathematical methods have been developed to treat uncertainty and calibrate hydrology models. Methods to represent model parameter, state, and prediction uncertainty include Bayesian approaches (Beven and Binley, ; Kuczera and Parent, ; Thiemann et al ., ; Vrugt et al ., , Abbaspour et al ., ; Tonkin and Doherty, ; Vrugt et al ., ; Moore et al ., ; Schoups and Vrugt, ; Laloy and Vrugt, ; Pourreza‐Bilondi and Samadi, ; Pourreza‐Bilondi et al ., ), set‐theoretic (Klepper et al ., ), sequential data assimilation (Madsen et al ., ; Moradkhani et al ., ; Vrugt et al ., ), stochastic optimization techniques (Duan et al ., ; Eberhart and Kennedy, ), and multi‐model averaging methods (Georgekakos et al ., 2004; Ajami et al ., ; Vrugt and Robinson, ). These methods differ in mathematical rigor, underlying assumptions about the residual error distribution, and how explicitly those assumptions are expressed in the modeling procedure.…”

Section: Introductionmentioning

confidence: 99%

Assessing Parameter Uncertainty of a Semi‐Distributed Hydrology Model for a Shallow Aquifer Dominated Environmental System

J American Water Resour Assoc

Tufford

Carbone

2017

This paper examines the performance of a semi-distributed hydrology model (i.e., Soil and Water Assessment Tool [SWAT]) using Sequential Uncertainty FItting (SUFI-2), generalized likelihood uncertainty estimation (GLUE), parameter solution (ParaSol), and particle swarm optimization (PSO). We applied SWAT to the Waccamaw watershed, a shallow aquifer dominated Coastal Plain watershed in the Southeastern United States (U.S.). The model was calibrated (2003)(2004)(2005) and validated (2006)(2007) at two U.S. Geological Survey gaging stations, using significant parameters related to surface hydrology, hydrogeology, hydraulics, and physical properties. SWAT performed best during intervals with wet and normal antecedent conditions with varying sensitivity to effluent channel shape and characteristics. In addition, the calibration of all algorithms depended mostly on Manning's n-value for the tributary channels as the surface friction resistance factor to generate runoff. SUFI-2 and PSO simulated the same relative probability distribution tails to those observed at an upstream outlet, while all methods (except ParaSol) exhibited longer tails at a downstream outlet. The ParaSol model exhibited large skewness suggesting a global search algorithm was less capable of characterizing parameter uncertainty. Our findings provide insights regarding parameter sensitivity and uncertainty as well as modeling diagnostic analysis that can improve hydrologic theory and prediction in complex watersheds. Editor's note: This paper is part of the featured series on SWAT Applications for Emerging Hydrologic and Water Quality Challenges. See the February 2017 issue for the introduction and background to the series.(KEY TERMS: SWAT; parameter uncertainty; shallow aquifer; modeling diagnostic analysis.) Samadi, S., D.L. Tufford, and G.J. Carbone, 2017. Assessing Parameter Uncertainty of a Semi-Distributed Hydrology Model for a Shallow Aquifer Dominated Environmental System.

“…DREAM (zs) used the

\overset{R}{true}

‐statistic (Gelman & Rubin, 1992) to determine convergence to the stationary posterior distribution. Readers are referred to Vrugt et al (2009), Schoups and Vrugt (2010), and Pourreza‐Bilondi et al (2017) for more discussion on DREAM (zs).…”

Section: Methodsmentioning

confidence: 99%

“…Several indices were used to quantify the goodness of sensitivity analysis as well as calibration performance for BMA: the P factor, which is the percentage of data bracketed by a 95% prediction uncertainty band (95PPU; maximum value is 100%), and R factor (or d factor), which is the average width of the uncertainty band divided by the standard deviation of the corresponding measured variable (minimum value is zero; Abbaspour et al, 2004). Theoretically, the value for the P factor ranges between 0% and 100%, while the R factor ranges between 0 and infinity (see Pourreza-Bilondi et al, 2017, for further information). Based on the requirement of the geometric structure of the prediction bounds, two different indices for assessing the average asymmetry degree of the prediction bounds with respect to the observed hydrograph are proposed.…”

Section: Parameter Sensitivity and Uncertainty Algorithmmentioning

confidence: 99%

Bayesian Model Averaging With Fixed and Flexible Priors: Theory, Concepts, and Calibration Experiments for Rainfall‐Runoff Modeling

Pourreza‐Bilondi

Wilson

et al. 2020

J Adv Model Earth Syst

This paper introduces for the first time the concept of Bayesian model averaging (BMA) with multiple prior structures, for rainfall-runoff modeling applications. The original BMA model proposed by Raftery et al. (2005, https://doi.org.10.1175/MWR2906.1) assumes that the prior probability density function (pdf) is adequately described by a mixture of Gamma and Gaussian distributions. Here we discuss the advantages of using BMA with fixed and flexible prior distributions. Uniform, Binomial, Binomial-Beta, Benchmark, and Global Empirical Bayes priors along with Informative Prior Inclusion and Combined Prior Probabilities were applied to calibrate daily streamflow records of a coastal plain watershed in the southeast United States. Various specifications for Zellner's g prior including Hyper, Fixed, and Empirical Bayes Local (EBL) g priors were also employed to account for the sensitivity of BMA and derive the conditional pdf of each constituent ensemble member. These priors were examined using the simulation results of conceptual and semidistributed rainfall-runoff models. The hydrologic simulations were first coupled with a new sensitivity analysis model and a parameter uncertainty algorithm to assess the sensitivity and uncertainty associated with each model. BMA was then used to subsequently combine the simulations of the posterior pdf of each constituent hydrological model. Analysis suggests that a BMA based on combined fixed and flexible priors provides a coherent mechanism and promising results for calculating a weighted posterior probability compared to individual model calibration. Furthermore, the probability of Uniform and Informative Prior Inclusion priors received significantly lower predictive error, whereas more uncertainty resulted from a fixed g prior (i.e., EBL). Plain Language Summary This study presents a two-step procedure that includes model calibration of a range of hydrological models using DREAM (zs) algorithm, followed by ensemble prediction of streamflow using Bayesian model averaging (BMA) with various prior structures. The hydrological modeling simulations were first coupled with a new sensitivity analysis model and a parameter uncertainty algorithm to assess the sensitivity and uncertainty associated with each hydrologic model simulation. BMA was then used to subsequently combine the simulations on the most important parts of the posterior probabilities of each constituent hydrological model. Analysis suggests a BMA with fixed and flexible priors provides a coherent mechanism and promising results for calibrating a weighted posterior probability compared to individual model calibration. The hierarchy of prior distributions used in this study increased the flexibility of BMA fitting for daily streamflow simulation and reduced the dependence of posterior and predictive uncertainty (including model probabilities) on prior assumptions of hydrological modeling simulation.