Authors: Sooyeon Yi*, University of California - Berkeley, Yang Ju, University of California - Berkeley, Yiyi He, University of California - Berkeley
Topics: Water Resources and Hydrology
Keywords: inland floods, statistical and machine learning, climate change
Session Type: Paper
Start / End Time: 5:20 PM / 7:00 PM
Room: Bayside C, Sheraton, 4th Floor
Presentation File: No File Uploaded
The state of California is likely to see more flooding induced by climate change. Generating spatially explicit maps of flooding susceptibility are essential to help decision makers and stakeholders to identify assets at risk and determine adaptation options. The previous studies used process-based models to generate flood maps which are computationally expensive and difficult to cover the entire state over multiple climate scenarios and planning horizons. This research shows how to conduct a long-term, large-scale, multi-scenario flood mapping under climate change using machine learning models with less computational cost. We trained the models using historical flooding observations as the dependent variable, and rainfall, runoff, topography, soil type, and stream density, etc. as the independent variables. The historical flooding observations were from satellite-derived monthly global surface water product, and rainfall and runoff were from General Circulation Models. We will then select the model that best fits the historical data through a cross-validation process. Finally, we will use the model with rainfall and runoff projections to identify flooding susceptibility between 2000 to 2100. This process will be iterated through multiple planning horizons and climate change scenarios to understand changes and uncertainties in flooding susceptibility. The results will help stakeholders and decision makers to derive flood mitigation strategies.