AI-Enhanced Flood Forecasting: Partner Support Expanded

Every year, devastating floods wreak havoc, leading to tragic losses, mass displacement, and huge financial damage. To support governments, NGOs, global organizations, and vulnerable communities in taking proactive measures, our team at Google Research has been working on cutting-edge AI solutions to deliver precise flood data and enhance climate resilience. Our groundbreaking global flood forecasting models, published in *Nature*, now offer near real-time information to assist communities and organizations worldwide.

In line with our goal to make flood forecasting information accessible globally through AI, we're excited to announce several key expansions:

• We're extending our AI-powered riverine flood forecasting model to cover 100 countries, up from 80, reaching areas where 700 million people live, a jump from 460 million.
• We've developed an enhanced flood forecasting model that matches the accuracy of our previous model at a five-day lead time, but now at seven days.
• We're making our model forecasts accessible to researchers and partners through an upcoming API and our Google Runoff Reanalysis & Reforecast (GRRR) dataset.
• We're expanding coverage for researchers and experts by introducing "virtual gauges" for data-scarce locations via an upcoming API, the GRRR dataset, and a new expert data layer on Flood Hub, featuring nearly 250,000 forecast points across 150 countries.
• We're also releasing historical datasets from our flood forecasting model to aid researchers in understanding and mitigating the impact of devastating floods.

These expansions will bring vital riverine flood information to millions of people globally and provide researchers and partners with more comprehensive data.

Expanding Flood Forecasting Coverage to Reach More Countries and More People

By expanding our AI-powered riverine flood forecasting model to over 100 countries, we're now able to serve 700 million people worldwide, a significant increase from the previous 460 million. This expansion is fueled by recent research breakthroughs that have boosted our model's accuracy. We've incorporated more labeled data, tripled the number of training locations, introduced a new model architecture for increased robustness, and integrated a new weather forecasting model into our flood predictions. You can dive deeper into these developments on our Research blog.

This means more communities will have timely access to flood information, improving preparedness and response efforts for millions worldwide, accessible through Google Search, Google Maps, and Google Flood Hub.

Making Flood Forecasting Data Available for Experts and Researchers in More Ways and More Places

We're now sharing our advanced flood forecasting model forecasts, which boast improved accuracy, reliability, and broader coverage, with researchers and partners via GRRR and an upcoming API.

Our new API pilot program, developed with leading organizations, is open for signups. This API will enable researchers and practitioners to access our hydrologic forecasts and anticipated flood status, even in data-limited regions, to accelerate flood research and response.

While our global hydrologic model can predict discharge anywhere on Earth, Flood Hub currently only shows forecasts at points with gauges for verification using historical data from reliable sensors. However, many parts of the world lack these sensors. To address this, we're adding a new data layer to Flood Hub with 250,000 forecast points from our Flood Forecasting model using "virtual gauges." This will provide additional flood forecasting data to experts and researchers in over 150 countries, including those without physical gauges, enhancing data availability for previously underserved areas.