Uncertainty Propagation of Hydrologic Modeling in Water Supply System Performance: Application of Markov Chain Monte Carlo Method

Abstract

It is imperative for cities to develop sustainable water management and planning strategies in order to best serve urban communities that are currently facing increasing population and water demand. Water resources managers are often chastened by experiencing failures attributed to natural extreme droughts and floods. However, recent changes in water management systems have been responding to these uncertain conditions. Water managers have become thoughtful about the adverse effects of uncertain extreme events on the performance of water supply systems. Natural hydrologic variability and inherent uncertainties associated with the future climate variations make the simulation and management of water supplies a greater challenge. The hydrologic simulation process is one of the main components in integrated water resources management. Hydrologic simulations incorporate uncertain input values, model parameters, and a model structure. Therefore, stochastic streamflow simulation and prediction, and consideration of uncertainty propagation on performance of water supply systems (WSSs) are essential phases for efficient management of these systems. The proposed integrated framework in this study models a WSS by taking into account the dynamic nature of the system and utilizing a Markov chain Monte Carlo (MCMC) algorithm to capture the uncertainties associated with hydrologic simulation. Hydrologic responses from the results of a rainfall-runoff model for three watersheds of Karaj, Latyan, and Lar in Tehran, Iran, as the case study are used as inputs to the reservoirs. Results confirm that uncertainties associated with the hydrologic model's parameters propagate through the simulation and lead to a wide variation in reservoir storage and WSS performance metrics such as vulnerability and reliability. For example, water storage simulation in the Karaj Reservoir can vary up to 70% compared with the observed values. This causes contradiction and conflict in the management of reservoirs and water systems and decision making. The results emphasize the importance of analyzing WSS performance under uncertain conditions to improve the simulation of natural processes and support water managers for a more efficient decision-making process.

System Dynamics Approach for Assessing the Behaviour of the Lim Reservoir System (Serbia) under Changing Climate Conditions

“…To close the science-practice gap, it is required to provide a clear understanding of the uncertainty within the climate change impact analysis processes [4]. The uncertainty associated with future climate variations and natural hydrologic variability represents a great challenge for the water resource system management [14]. The reservoir inflows are the most significant contributor of uncertainty to water resources management.…”

“…The reservoir inflows are the most significant contributor of uncertainty to water resources management. The sources of uncertainty in the reservoir inflows originate in model parameters and the model structure [14].…”

“…The simple averaging of the SDSM outputs is performed by Equations (8), (9) and (10). Finally, the application of Equations (12), (13) and (14) provides the contribution of each step uncertainty (RCM, BCM, ES) to the total uncertainty in the SDSM outputs.…”

Surrogate‐Based Bayesian Inverse Modeling of the Hydrological System: An Adaptive Approach Considering Surrogate Approximation Error

“…In this paper, our concern is MCMC, which is a Bayesian inverse method that can sample the posterior distribution of model parameters in a consistent and coherent manner (Brooks et al, 2011;Vrugt, 2016). Over the past decade, MCMC has been widely used in hydrological simulations (Cao et al, 2018;Gaume et al, 2010;Goharian et al, 2018;Kuczera et al, 2010;Vrugt & Beven, 2018). However, when implementing MCMC for inverse modeling, a large number of model evaluations are usually required, especially in high-dimensional, strongly nonlinear problems.…”

Comparing Machine Learning and Decision Making Approaches to Forecast Long Lead Monthly Rainfall: The City of Vancouver, Canada

“…Here, two different modeling approaches are developed and compared for rainfall forecasting. At the first step, MRMR (Maximum Relevance Minimum Redundancy) feature selection method picks the most effective signals as predictors [32]. Then, the two approaches are employed to build the forecast models.…”