Hybrid matched-block bootstrap for stochastic simulation of multiseason streamflows

A new hybrid model is proposed for stochastic simulation of multiseason streamflows, as an improvement over the hybrid moving block bootstrap (HMBB) model introduced by the authors in an earlier work. In the proposed model, periodic streamflows are partially pre-whitened using a parsimonious linear periodic autoregressive/autoregressive moving average model and residuals are extracted. The nonoverlapping within-year blocks formed from the residuals are conditionally resampled using the matched-block bootstrap (MABB) to obtain innovations, which are then post-blackened to generate synthetic replicates. The use of MABB for resampling the residuals, in lieu of moving block bootstrap (that was used in the HMBB model), has resulted in improved simulation of streamflows. The effectiveness of the proposed model in reproducing a wide variety of statistical attributes (including strong dependence structure) is demonstrated through Monte-Carlo simulations on both hypothetical and real data sets. Through application to streamflows of the Weber river, USA, the proposed model is shown to be more efficient in predicting reservoir storage capacity and in preserving storage based performance statistics and critical run characteristics compared to low-order linear periodic parametric models (Box-Jenkins type) and the HMBB model. This model offers enough flexibility and is a viable alternative to the conventional models when hydrologic sequence exhibits strong dependence and if the presence of nonlinearity and/or multimodality is suspected in the underlying hydrologic process. However, the proposed hybrid model may not be appropriate, if long-term dependence is significant in the hydrologic data. © 2006 Elsevier B.V. All rights reserved.