Flood forecasting
Floods are the most common and deadly natural disaster, affecting between 95 million and 240 million people annually, causing between 5,000 and 60,000 fatalities, and resulting in between $21 and $33 billion in economic damages. Much of the harm caused by floods is preventable, as early warnings can prevent up to 43% of fatalities and 35-50% of economic damages.
The World Bank estimates that early warning systems are the most cost-effective infrastructure investment in climate resilience, with a cost-benefit ratio of more than 1:9. In countries like Bangladesh, this is closer to 1:300. Recognizing the potential for life-saving impact, the United Nations has identified improved flood alerts as “low-hanging fruit” in disaster risk reduction.
The Google Flood Forecasting Initiative brings together Google Research and Crisis Response teams on Search and on Maps to provide accurate and actionable flood alerts covering all affected by floods globally. The Research team develops forecast and detection models using AI and ML, which are productionized to distribute early flood warnings via Search alerts, Maps and notifications.
MVP
In 2019, the FF Initiative launched a pilot in India’s Ganges River Basin, an area where high population density, heavy rainfall during monsoons, and terrain prone to drastic seasonal floods combined with a difficulty to share emergency information in a timely manner. The models were highly reliable, but the initial alerts were displayed off Search and in English only, and user feedback from the pilot revealed they were too text heavy, with a flood map too complex for users to understand.
2019 flood alerts
For the 2020 monsoon season, when the flood alerts were planned to scale across India and Bangladesh and move to Search pages, I designed the alerts’ new UX aimed at making complex concepts of risk predictions simple and clear to a primarily NIU (New Internet User) audience - focusing on increasing clarity and usability through design iterations that addressed language barriers, digital literacy challenges, and trust issues.
This included creating visual-forward layouts, multi-language support, and prominent location refinement to make it easy for users to navigate to the forecast for different locales. I closely collaborated with the Maps’ Crisis team on the flood maps that were displayed on both Search and Maps. We also updated the notifications to include multiple languages and a flood icon, with the goal of communicating the initial risk well enough to encourage users to click through to the full alert page.
2020 flood alerts
The design improvements greatly enhanced the reach, usability, and effectiveness of flood alerts, leading to more people being informed and taking action before floods. Following the launch, the redesigned flood alerts showed substantial growth in impressions and CTR, indicating increased reach, visibility and user engagement, as well as increase in DAUs and improved page quality metrics. The product was also featured in Google I/O and Google Design’s ‘Best of 2020’ and has received extensive press coverage.
Sundar Pichai presenting flood alerts at Google I/O ‘22
Hydrograph
Over the next few years, we continued making incremental improvements to simplify the alerts and improve their comprehension, such as eliminating the multitude of UI treatments and changing the map colors.
One key insight from UXR and user feedback was that the different types of infographics were either confusing and complex to understand or not informative enough. I evolved the alerts’ hero visual into a more informative data visualization called a hydrograph, which shows the trend of the flood over time, including a forecast when available. It can give users a better understanding of the current flood situation and how it is likely to develop in the coming hours and days.
We launched the redesign in 2024 as part of a set of updates that included increasing alert lead time from 3 to 7 days, and scaling coverage to a total of 88 countries. The implementation of the hydrograph resulted in a significant uplift in the alerts’ helpfulness metrics and an increase in DAUs.
Read more:
A big step for flood forecasts in India and Bangladesh
An Inside Look at Flood Forecasting
The Technology Behind our Recent Improvements in Flood Forecasting
Expanding our ML-based flood forecasting
Research paper: Flood forecasting with machine learning models in an operational framework
Fortune: Google bolsters its A.I.-enabled flood alerts for India and Bangladesh
The Verge: Google’s AI flood warnings now cover all of India and have expanded to Bangladesh
TechCrunch: Google's AI-powered flood alerts now cover all of India and parts of Bangladesh