Yossi Matias, VP Engineering & Research, and Grey Nearing, Research Scientist, from Google Research, discuss the impact of floods as the most common natural disaster, causing approximately $50 billion in global financial damages annually. The frequency of flood-related disasters has increased significantly since 2000, largely due to climate change. Nearly 1.5 billion people, representing 19% of the global population, face significant risks from severe flood events. Enhancing early warning systems to provide accurate and timely information to these populations can potentially save thousands of lives each year.
To address this issue, Google Research initiated a flood forecasting effort in 2017, aiming to develop a real-time operational flood forecasting system. Through years of research and collaboration, Google has made advancements in flood forecasting technology, leveraging machine learning (ML) to improve global-scale predictions, particularly in regions with limited flood-related data availability. The research, published in Nature, demonstrates how ML can enhance flood forecasting accuracy, extending the reliability of global nowcasts and improving forecasts in regions of Africa and Asia to levels comparable to those in Europe.
Google’s FloodHub tool provides real-time river forecasts up to seven days in advance, covering river reaches in over 80 countries. These forecasts empower individuals, communities, governments, and organizations to take proactive measures to protect vulnerable populations. The development of ML-based hydrological models has been a key factor in expanding flood forecasting coverage to include regions in India and Bangladesh, with ongoing efforts to further improve predictions globally.
The river forecast model developed by Google Research utilizes Long Short-Term Memory (LSTM) neural networks to make accurate predictions. By ingesting historical and forecasted weather data, along with static watershed attributes, the model outputs probabilistic forecasts of streamflow at different time intervals. Through rigorous training with publicly available data sources, the model has shown significant improvements over existing flood forecasting systems, such as GloFAS version 4.
Overall, Google’s efforts in flood forecasting highlight the potential of ML technologies to enhance early warning systems and mitigate the impact of natural disasters on vulnerable populations worldwide. Through ongoing research and collaboration, Google continues to refine its flood forecasting capabilities to provide reliable and timely information to those at risk of flood-related disasters.
Source link
