Improved Wavelet Modeling Framework for Hydrologic Time Series Forecasting

Abstract

The combination of wavelet analysis with black-box models presently is a prevalent approach to conduct hydrologic time series forecasting, but the results are impacted by wavelet decomposition of series, and uncertainty cannot be evaluated. In this paper, the method for discrete wavelet decomposition of series was developed, and an improved wavelet modeling framework, WMF for short, was proposed for hydrologic time series forecasting. It is to first separate different deterministic components and remove noise in original series by discrete wavelet decomposition; then, forecast the former and quantitatively describe noise's random characters; at last, add them up and obtain the final forecasting result. Forecasting of deterministic components is to obtain deterministic forecasting results, and noise analysis is to estimate uncertainty. Results of four hydrologic cases indicate the better performance of the proposed WMF compared with those black-box models without series decomposition. Because of having reliable hydrologic basis, showing high effectiveness in accuracy, eligible rate and forecasting period, and being capable of uncertainty evaluation, the proposed WMF can improve the results of hydrologic time series forecasting.