aichat.blog

Using Ai to Expand Global Access to Reliable Flood Forecasts

Google Research 3:00 am on May 23, 2024

The paper details a river forecast model using two LSTM neural networks applied sequentially: the first ingests historical weather data, and the second uses this combined with forecasted weather to predict future streamflow volumetric flow rates. The method leverages an asymmetric Laplacian distribution mixture density for predictions, achieving comparable accuracy to existing flood prediction systems up to 5-day advance warnings while offering enhanced capabilities for larger events and rare occurrences.

Model architecture:
Two LSTMs ingesting historical weather data (hindcast) and forecasted data (forecast).
Parameters predicted are volumetric streamflow rates using an asymmetric Laplacian distribution mixture density.
Improved accuracy compared to GloFAS with comparable reliability up to 5-day lead time for major floods and rare events.
Collaborative efforts aimed at expanding global coverage and addressing various types of flood events.

http://blog.research.google/2024/03/using-ai-to-expand-global-access-to.html

< Previous Story - Next Story >

Using Ai to Expand Global Access to Reliable Flood Forecasts

Categories