evidoria

← Back to browse

Good practice

Google Flood Hub — Bangladesh Water Development Board's AI-Powered River Flood Alert System

Bangladesh · Dhaka · See the Bangladesh profile

Since 2020, Google Research's machine-learning flood model — which forecasts river levels even in ungauged basins — has combined with Bangladesh Water Development Board data to push flood alerts; by 2021 it reached 1.5 million users with a lead time of three hours to three days.

1 million
Notifications sent (2020 pilot)
300,000
Android users reached (2020 pilot)
2.9 million
Notifications sent (13-31 Aug 2021)
1.5 million
Unique users reached (13-31 Aug 2021)
3.5 %
Click-through rate (2021)
3 hours to 3 days
Forecast lead time
80+ countries, ~460 million people
Global model coverage
Google Flood Hub — Bangladesh Water Development Board's AI-Powered River Flood Alert System

Details

Maturity
Scaling
Promoter
Bangladesh Water Development Board (BWDB)
Period
2020-present
Keywords
flood forecasting, disaster early warning, machine learning, hydrology, mobile alerts

Context

Bangladesh's Water Development Board (BWDB), the national hydrological agency, partnered with Google Research from 2020 to layer an AI/machine-learning flood-inundation model on top of BWDB's own gauge readings and forecasts, extending flood prediction into ungauged tributaries of the flood-prone Brahmaputra-Jamuna and Padma river systems. Localized warnings are pushed to residents through Google Search, Maps and Android notifications.

Objectives

To provide flood early-warning alerts with a lead time of three hours to three days, including in remote basins that lack dense sensor networks.

Activities

Google's hybrid hydrologic-hydraulic ML model, described in a March 2024 Nature paper, is combined with BWDB's real-time gauge data and 5-day forecasts. The system expanded from a 2020 pilot covering 36 sub-districts in 14 districts to 55 districts and 99 sub-districts by the 2021 monsoon.

Results

The 2020 pilot sent about 1 million notifications to roughly 300,000 Android users. Between 13-31 August 2021 alone the expanded system sent 2.9 million notifications to 1.5 million unique users, with a 3.5% click-through rate. The underlying model now operates in over 80 countries, reaching an estimated 460 million people globally (a global, not Bangladesh-specific, figure).

Conclusions

Google's own reporting acknowledges low Android penetration in the remote, flood-exposed areas the system most needs to reach, and describes ongoing work to extend alerts via SMS and IVR for feature-phone users. No independent, Bangladesh-specific accuracy evaluation of forecasts against observed flooding has been published.

Implementation

Implementation detail (cost, timeline, staffing, conditions for success) is not yet available for this practice.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful