'Wavelet-based cross-correlation analysis and a hybrid wavelet-multivariate Bayesian model for short-term streamflow forecasting using local climatic data'
Andres M Ticlavilca (email@example.com), Utah Water Research Laboratory, Utah State University; Inga Maslova (firstname.lastname@example.org), American University; Mac McKee (email@example.com), Utah Water Research Laboratory, Utah State University
A new approach is presented for creating short-term forecasts of streamflow in a snowmelt-dominated watershed. The forecasting methodology relies on (1) wavelet-based cross-correlation analysis to study lead-lag relationships between streamflow and climatic time series data and (2) incorporation of the resulting information into a multivariate Bayesian regression model to develop short-term streamflow forecasts. The wavelet-based cross-correlation analysis is demonstrated with the use of daily local climatic data (in the form of precipitation, temperature and snow water equivalent) that are obtained from automated snowpack telemetry (SNOTEL) stations in the Bear River Watershed in Utah. Daily streamflow data is obtained from a U.S. Geological Survey (USGS) gage station located on the Logan River, near Logan, Utah. The data are decomposed into meaningful components formulated in terms of wavelet multiresolution analysis. Next, a computational intelligence modeling approach based on a multivariate Bayesian regression is used to produce daily streamflow forecasts up to seven days ahead. The results from the wavelet-based cross-correlation analysis are used to select the data to build the model. The proposed methods can incorporate important information from trends of the local climate time series into models that learn these patterns to produce improved streamflow predictions at different time scales.