Development of an accurate and reliable hourly flood forecasting model using wavelet–bootstrap–ANN (WBANN) hybrid approach

• Published in: Journal of hydrology (Amsterdam) Vol. 394; no. 3-4; pp. 458 - 470
• Main Authors: Tiwari, Mukesh K., Chatterjee, Chandranath
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier B.V 26.11.2010; [Amsterdam; New York]: Elsevier; Elsevier
• Subjects: ANNs; Bootstrap; Decomposition; Earth sciences; Earth, ocean, space; Engineering and environment geology. Geothermics; Exact sciences and technology; Flood forecasting; Freshwater; Hydrology; Hydrology. Hydrogeology; Learning theory; Mathematical models; Natural hazards: prediction, damages, etc; Neural networks; Time series; Transformations; Wavelet; Asia; India; Wavelet; Bootstrap; Flood forecasting; ANNs; rivers; floods; models; reliability; neural networks; wavelet transformation; wavelet; prediction; confidence interval; uncertainties; monsoons; gauging
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2010.10.001

A hybrid wavelet–bootstrap–ANN (WBANN) model is developed in this study to explore the potential of wavelet and bootstrapping techniques for developing an accurate and reliable ANN model for hourly flood forecasting. The wavelet technique is used to decompose the times series data into different components which capture useful information on various resolution levels. Five years hourly water level data for monsoon season from five gauging stations in Mahanadi River basin, India are used in this study. The observed water level time series of a particular gauging station is decomposed to sub-series by discrete wavelet transformation and then appropriate sub-series are added up to develop new time series. The bootstrap resampling method is used to generate different realizations of the newly constructed datasets using discrete wavelet transformation to create a set of bootstrap samples that are finally used as input to develop WBANN model. Performance of WBANN model is also compared with three different ANN models: traditional ANNs, wavelet based ANNs (WANNs), bootstrap based ANNs (BANNs). The results showed that the hybrid models WBANN and BANN produced better results than the traditional ANN and WANN models. WBANN model simulated the peak water level better than ANN, WANN and BANN models, and in general, the overall performance of WBANN model is accurate and reliable as compared to the other three models. This study reveals that whereas wavelet decomposition improves the performance of ANN models, bootstrap resampling technique produces more consistent and stable solutions. WBANN model is also used to assess the predictive uncertainty in forms of confidence intervals (CI) to assess the predictive uncertainty for 1–10 h lead time forecasts. Results obtained indicate that WBANN forecasting model with confidence intervals can improve their reliability for flood forecasting.