Model fusion for improving hypoxia forecasts in Corpus Christi Bay, TX, USA: A study of boosting and historical scenario modeling

Chinta, Indu

Permalink

https://hdl.handle.net/2142/18625 Copy

Description

Title

Model fusion for improving hypoxia forecasts in Corpus Christi Bay, TX, USA: A study of boosting and historical scenario modeling

Author(s)

Chinta, Indu

Issue Date

2011-01-21T22:52:15Z

Director of Research (if dissertation) or Advisor (if thesis)

Minsker, Barbara S.

Department of Study

Civil & Environmental Eng

Discipline

Environ Engr in Civil Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Hypoxia
• forecasting
• model fusion
• boosting
• historical scenario modeling
• Corpus Christi Bay

Abstract

This study aims to create more accurate and efficient near-real-time forecasts of hypoxia that will give researchers advance notice for manual sampling during hypoxic events. Hypoxic or dead zones, which occur when dissolved oxygen levels in water drop below 2 mg/L, are prevalent worldwide. An example of such an hypoxic zone forms intermittently in Corpus Christi Bay (CC Bay), Texas, a USEPA-recognized estuary of national significance. Hypoxia in CC Bay is caused by inflow of hypersaline waters that enter from adjacent bays and estuaries, natural fluctuations in oxygen levels due to the oxygen production-consumption cycle of the aquatic flora and fauna, seasonal fluctuations, and discharges from several wastewater treatment plants. The hypoxia forecasting method tested in this work involves a suite of data-driven model fusion techniques such as historical scenario modeling and boosting both a k-nearest neighbor (KNN) algorithm and the historical scenario model. Existing data-driven k-nearest neighbor and physics-based valve models are used as the basis for the model fusion. The historical scenario model combines the k-nearest neighbor algorithm with the valve model to predict the probability of hypoxia twenty-four hours ahead. Boosting involves training the model repeatedly on subsets of the training dataset. The results of the fused models are compared with those of the individual models to test the effectiveness of model fusion in predicting the estuarine conditions. The results showed that the valve model, which has been hitherto computing oxygen profiles, can be extended to forecast probabilities of hypoxia when combined with the k-nearest neighbor algorithm to form the historical scenario model. The findings also show that boosting significantly enhances the performance of the k-nearest neighbor algorithm and the historical scenario model, although further testing on more extensive continuous datasets is needed to verify the findings in other locations. The results show promise for model fusion to be effective for real-time forecasting in hypoxia-affected water bodies.

Graduation Semester

2010-12

Permalink

http://hdl.handle.net/2142/18625

Copyright and License Information

Copyright 2010 Indu Chinta

Owning Collections