Author(s): Durga Lal Shrestha; Nagendra Kayastha; Dimitri Solomatine
Linked Author(s): Dimitri Solomatine, Durga Lal Shrestha
Keywords: No Keywords
Abstract: Estimation of uncertainties in hydrological models due to uncertain parameters and other sources are often carried out by Monte Carlo (MC) simulation due to its flexibility and robustness. However, MC simulation requires large number of runs to establish the reliable estimate of uncertainties. In this paper we estimate the model parametric uncertainty by building a surrogate model to be used in real-time situation to replicate the time consuming MC simulation. The surrogate model is built by a machine learning method, M5 model tree, which is equivalent to a piece-wise linear regression model. The proposed approach is applied to a hydrological conceptual model. The results indicate that the proposed method is efficient and effective; thereby replicate MC simulation when it is impracticable to run numerous simulations for complex, computationally intensive hydrological models and when the forecast lead time is very short.
Year: 2009