A Robust Monthly Streamflow Forecasting Model Using a Multivariate Bayesian Regression Model Coupled with Wavelet Decomposition Approach

Presenter Information

Andres M. Ticlavilca

Location

ECC 216

Event Website

http://water.usu.edu/

Start Date

4-3-2012 3:35 PM

End Date

4-3-2012 3:40 PM

Description

This research presents a modeling approach that incorporates wavelet-based analysis techniques used in statistical signal processing and multivariate machine learning regression to forecast monthly streamflow. The streamflows from two USGS gauge stations located on the Lake Fork River and Yellowstone River in the Uinta Basin, Utah are used. The model is developed using a Multivariate Relevance Vector Machine (MVRVM) that is based on a Bayesian learning machine approach for regression. The inputs of the model utilize past information of streamflow and Pacific sea surface temperature (SST). The inputs are decomposed into meaningful components formulated in terms of wavelet multiresolution analysis (MRA). The proposed hybrid of wavelet decomposition and machine learning regression approaches captures sufficient information at meaningful temporal scales and improves the performance of the monthly streamflow forecasts in Utah. A bootstrap analysis is used to explore the robustness of the hybrid modeling approach.

This document is currently not available here.

Share

COinS
 
Apr 3rd, 3:35 PM Apr 3rd, 3:40 PM

A Robust Monthly Streamflow Forecasting Model Using a Multivariate Bayesian Regression Model Coupled with Wavelet Decomposition Approach

ECC 216

This research presents a modeling approach that incorporates wavelet-based analysis techniques used in statistical signal processing and multivariate machine learning regression to forecast monthly streamflow. The streamflows from two USGS gauge stations located on the Lake Fork River and Yellowstone River in the Uinta Basin, Utah are used. The model is developed using a Multivariate Relevance Vector Machine (MVRVM) that is based on a Bayesian learning machine approach for regression. The inputs of the model utilize past information of streamflow and Pacific sea surface temperature (SST). The inputs are decomposed into meaningful components formulated in terms of wavelet multiresolution analysis (MRA). The proposed hybrid of wavelet decomposition and machine learning regression approaches captures sufficient information at meaningful temporal scales and improves the performance of the monthly streamflow forecasts in Utah. A bootstrap analysis is used to explore the robustness of the hybrid modeling approach.

https://digitalcommons.usu.edu/runoff/2012/Posters/29