| Title |
Parameter and Modeling Uncertainty Analysis of Semi-Distributed Hydrological Model using Markov-Chain Monte Carlo Technique |
| Authors |
최정현 ( Choi Jeonghyeon ) ; 장수형 ( Jang Suhyung ) ; 김상단 ( Kim Sangdan ) |
| DOI |
https://doi.org/10.15681/KSWE.2020.36.5.373 |
| Keywords |
Semi-distributed hydrological model; Markov-Chain Monte Carlo technique; Metropolis-Hastings; algorithm; Parameter calibration; Parameter uncertainty |
| Abstract |
Hydrological models are based on a combination of parameters that describe the hydrological characteristics and processes within a watershed. For this reason, the model performance and accuracy are highly dependent on the parameters. However, model uncertainties caused by parameters with stochastic characteristics need to be considered. As a follow-up to the study conducted by Choi et al (2020), who developed a relatively simple semi-distributed hydrological model, we propose a tool to estimate the posterior distribution of model parameters using the Metropolis-Hastings algorithm, a type of Markov-Chain Monte Carlo technique, and analyze the uncertainty of model parameters and simulated stream flow. In addition, the uncertainty caused by the parameters of each version is investigated using the lumped and semi-distributed versions of the applied model to the Hapcheon Dam watershed. The results suggest that the uncertainty of the semi-distributed model parameters was relatively higher than that of the lumped model parameters because the spatial variability of input data such as geomorphological and hydrometeorological parameters was inherent to the posterior distribution of the semi-distributed model parameters. Meanwhile, no significant difference existed between the two models in terms of uncertainty of the simulation outputs. The statistical goodness of fit of the simulated stream flows against the observed stream flows showed satisfactory reliability in both the semi-distributed and the lumped models, but the seasonality of the stream flow was reproduced relatively better by the distributed model. |