Experimenting with predicting street flooding during the 2017 Subsurface Hackathon in Houston
The basic idea
- Hydrologic models and historical discharge predict flooding in floodplains very well
- Hydrologic models do not predict local street flooding and ponding
- Clogged storm drains, etc
- Detailed topography can help predict localized flooding
- e.g. Is the area very flat?
- Are we in a local low point (even one that drains)?
So, let's train a classifier on features extracted from LIDAR-based DEMs
- Local slope, relief, etc in a moving window.
What do we train it on?
- UFlood/floodmap.io database dump: https://www.dropbox.com/sh/5757a3ujflzdwxo/AAAFD97LMXCRe0YW1HMJDvQ-a?dl=0
- Crowdsourced street flooding information during Hurricane Harvey
- Gives us detailed info about which streets flooded, even when houses didn't