Advisory Committee Chair
Rouzbeh Nazari
Advisory Committee Members
Md Golam Rabbani Fahad
Wesley C Zech
Document Type
Thesis
Date of Award
2023
Degree Name by School
Master of Civil Engineering (MCE) School of Engineering
Abstract
Globally, coastal communities face significant vulnerability to the risks and impacts of flooding. This study addresses two critical aspects of flood management: enhanced flood damage assessment and coastal flooding prediction. Traditional flood damage modeling techniques have limitations, and the availability of comprehensive datasets is limited. The first part addresses these challenges. Multivariable machine learning models are applied to enhance flood damage assessment. A comprehensive dataset is developed by combining the Alabama property dataset with the National Flood Insurance Program (NFIP) claims dataset from Hurricane Katrina. Spatial varying datasets are added to enhance the comprehensiveness of NFIP dataset and oversampling techniques are utilized to address data imbalance. Ensemble machine learning approaches are employed to develop multi-variable flood damage models, achieving satisfactory results in identifying damaged properties. In the second part, a Sequence to Sequence (Seq2Seq) learning Long Short-Term Memory (LSTM) Encoder Decoder model is proposed for coastal flood prediction. The model leverages the advantages of LSTM cells, Seq2Seq learning, while utilizing a novel learning approach based on learning from both past and forecasted data. The goal was to develop a model capable of providing a 72-hour runoff and coastal water level forecast during hurricane events. Case studies in the Lower Alabama and Lower Tombigbee watersheds validate the rainfall-runoff model, iv while coastal water level prediction models are tested using data from six stations National Oceanic and Atmospheric Administration stations. The LSTM Encoder Decoder model simulates runoff and coastal water levels for up to 72 hours with NSE and R2 > 0.7, demonstrating the ability to provide short-term and long-term predictions. The findings provide valuable insights and tools for decision-makers and stakeholders involved in flood management and mitigation efforts. By enhancing flood damage assessment and coastal flooding prediction through the application of machine learning techniques, this research aims to enhance community resiliency and support effective flood mitigation strategies.
Recommended Citation
Museru, Mujungu Lawrence, "Enhancing Community Resiliency With Big Data and Machine Learning: Property Damage and Flood Forecasting During Hurricanes" (2023). All ETDs from UAB. 412.
https://digitalcommons.library.uab.edu/etd-collection/412