DSD-INT 2017 ML-based Flood Forecasting Systems in Data-Scarce Regions - Nevo
1 like2,030 views
The Google Flood Forecasting Initiative aims to save lives and reduce suffering during floods by providing timely, high-resolution forecasts and actionable safety information. Floods, being the deadliest natural disaster, affect millions and cause significant economic damage, highlighting the importance of improving alert systems. The initiative combines Google's computational resources and machine learning expertise with hydrologic knowledge to enhance flood forecasting and response efforts.
DSD-INT 2017 ML-based Flood Forecasting Systems in Data-Scarce Regions - Nevo
- 2. Google Crisis Response Our Mission Save lives and prevent suffering with timely information. Help everyone make better decisions in emergencies by quickly getting them credible, actionable information; enabling them to contribute and exchange information.
- 3. SOS Alerts - The Beginning
- 4. SOS Alerts - Terror Attacks
- 5. Crisis Response - Typhoons
- 7. Haiti 2010 Japan 2011 Boston 2013 “I was at the airport sitting right outside that terminal when it was all happening. I had no clue something this serious was happening until I used Google and found out. You saved my life.” Fort Lauderdale SOS Alert recipient, Airport Shooting - 6th Jan, 2017 Fort Lauderdale 2017Paris 2015
- 8. Flooding impact ● Floods are the most common and the most deadly natural disaster on the planet. They affect 250M people, and cause 7,000 fatalities and $9.8B in economic damages annually ● Simple safety guidelines prior to and during a flood can prevent a substantial portion of flood fatalities ● Improving alerts for floods has the most life-saving potential of all interventions assessed by Google’s Crisis Response team
- 9. Flooding Alerts - Current Status ● In flooding events, we provide real time warnings, updates and safety recommendations ● Currently, the warnings are based exclusively on governmental alerts. Which is a problem.
- 10. The Google Flood Forecasting Initiative Goal: High-resolution high-accuracy flood forecasts and warnings where it matters most
- 11. The Google Flood Forecasting Initiative ● Why Google? ○ Computational resources ○ Access to global data (Elevation, user-generated data, etc.) ○ Scalability ○ Machine learning expertise Why not (just) Google? Hydrologic expertise Operational expertise Relationship with organizational consumers We’re looking for collaborators!
- 12. ● Use machine learning models to integrate information from various signals: Optical, MW, SAR and others ● Currently use Gradient Boosting Machines, but experimenting with other frameworks ● This data can (hopefully) be useful for all hydrologic models in data scarce regions, not just ours Discharge data - via Remote Estimation Sentinel-1 Sentinel-3
- 13. Hydrologic Modeling - Research Directions ● Incorporating machine learning into classic physical models ● Estimating physical model parameters in ungauged basins ● Downscaling model-generated input data
- 14. Inundation Modeling ● Google has already spent many millions collecting global high-resolution elevation data ● We are currently working on implementing standard raster-based flood inundation models based on this data ● Pilot based on CWC stream gauge data
- 15. Questions? Feel free to contact me at sellanevo@google.com