Monthly water balance modeling: Probabilistic, possibilistic and hybrid methods for model combination and ensemble simulation

• Published in: Journal of hydrology (Amsterdam) 2014-04, Vol.511, p.675-691
• Main Authors: Nasseri, M., Zahraie, B., Ajami, N.K., Solomatine, D.P.
• Format: Article
• Language: English
• Subjects: Bayesian Model Averaging; Clustering; Fuzzy clustering regression; Model combination; Ordinary Kriging (OK); Uncertainty
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2014.01.065

•Three different have been proposed for combination and ensemble simulations.•Ordinary Kriging (OK) has been firstly used for these targets.•The other methods are based on Bootstrap and Fuzzy concept.•Three different criterion have been use to evaluate these methods.•Based on them, the proposed methods have been advantages versus MLP and BMA methods. Multi-model (ensemble, or committee) techniques have shown to be an effective way to improve hydrological prediction performance and provide uncertainty information. This paper presents two novel multi-model ensemble techniques, one probabilistic, Modified Bootstrap Ensemble Model (MBEM), and one possibilistic, FUzzy C-means Ensemble based on data Pattern (FUCEP). The paper also explores utilization of the Ordinary Kriging (OK) method as a multi-model combination scheme for hydrological simulation/prediction. These techniques are compared against Bayesian Model Averaging (BMA) and Weighted Average (WA) methods to demonstrate their effectiveness. The mentioned techniques are applied to the three monthly water balance models used to generate stream flow simulations for two mountainous basins in the South-West of Iran. For both basins, the results demonstrate that MBEM and FUCEP generate more skillful and reliable probabilistic predictions, outperforming all the other techniques. We have also found that OK did not demonstrate any improved skill as a simple combination method over WA scheme for neither of the basins.