Active Hackathon

Google Makes Flood Forecasting System Live In Entire India

Google has expanded its flood forecasting initiative to all over India

Google, which has been working on the flood forecasting initiative since 2018 (partnering with the Central Water Commission for India and with the Bangladesh Water Development Board), has expanded its flood forecasting system to all over India and Bangladesh. The tech giant also said that it is working on expanding its reach to countries in South Asia and South America as well. Recently, it also launched Google Flood Hub for hyper-local flood-related data (going as specific as a village) in a visual-friendly format.

A report says that Bihar is India’s most flood-prone state in India, with 76% of the population in North Bihar living under the recurring threat of flood devastation. The catastrophic damage floods cause can be greatly reduced by more accurate alert systems, real-time data analysis and better flood prediction technology. And this is where big names in tech should step in to create an impact by deploying their advanced algorithms to solve a real-world problem that destroys so many lives every year.

THE BELAMY

Sign up for your weekly dose of what's up in emerging technology.

Image: Google

A large chunk of people in India and Bangladesh still do not have internet access or even a mobile connection. To make sure that such vulnerable people are not left out, Google has teamed up with local aid organizations like the Federation of Red Cross and Red Crescent Societies, the Indian Red Cross Society (IRCS), Bangladesh Red Crescent Society (BDRCS) and Yuganter to reach out to such people. Google also said that in 2021, its operational systems expanded to cover an area with over 360 million people, and it sent a massive 115 million alerts. The notifications were sent in the local language, including Hindi, Bengali, Gujarati, Marathi, Tamil, Telugu, Kannada, Malayalam, and English.

Forecast Future River Stages 

What Google aims to do through this flood warning system is to forecast future river stages along with high-resolution flood inundation situated along the river network, says the research paper titled, “Flood forecasting with machine learning models in an operational framework.” These forecasts are then disseminated in the form of flood alerts to relevant agencies and also sent out directly to the public. Presently, the system works for deployment in large gauged rivers having stream gauge data availability and potentially complex inundation areas during floods, the research clarifies. 

Stream gauges in locations of interest are defined as target gauges, each of which has a warning stage threshold and a predefined maximal lead time. An area of interest (AOI) is defined for each target gauge that brings out the surrounding region and the gauge where the inundation model is applied.

How is the model built?

The research brings out the general structure of the flood forecasting model and breaks it into four steps. 

  • Data management

In the first stage, the real-time data of Stream gauge measurements of the water stage and Satellite-derived precipitation data from the Integrated Multi-satellitE Retrievals for GPM (IMERG, Early Run) are ingested quality controlled, corrected, and preprocessed. 

  • Stage focused modelling

Here, stage forecast models are responsible for computing forecasted river stage data at target gauges. The system has two types of models- a model based on multiple linear regression and a model based on a Long Short-Term Memory (LSTM) network. The stage forecast models are trained and validated with historical data using a cross-validation scheme.  

The LSTM model was significantly better than the Linear model and was used operationally in 2021 for 165 out of 167 target gauges.

Image: Paper titled – “Flood forecasting with machine learning models in an operational framework

  • Inundation modelling

Flood inundation is computed with the Thresholding and the Manifold models. Thresholding computes inundation extent, while the Manifold model computes both inundation extent and depth. The Manifold model provides an ML alternative to hydraulic modelling of flood inundation.

  • Alerts through three channels

The last stage in the process is sending alerts which are distributed to government authorities, emergency response agencies, and people directly. They include information about the forecasted water stage at the target gauge, the inundation map, and inundation depth if available. These alerts are sent through three getaways – Google Search, smartphones within the forecasted flood inundated area receive a push notification and Google Maps. Alerts come in the local language to benefit the impacted population.

India and Bangladesh

Google said that during 2021, the flood warning system handled 376 target gauges, covering watershed sizes of 350 to 1,500,000 squared km. Stage forecast models were applied to 167 target gauges, and others were based on external forecasts. Inundation models were applied to 233 target gauges. The 115 million notifications sent reached about 22 million people.

More Great AIM Stories

Sreejani Bhattacharyya
I am a technology journalist at AIM. What gets me excited is deep-diving into new-age technologies and analysing how they impact us for the greater good. Reach me at sreejani.bhattacharyya@analyticsindiamag.com

Our Upcoming Events

Conference, in-person (Bangalore)
Machine Learning Developers Summit (MLDS) 2023
19-20th Jan, 2023

Conference, in-person (Bangalore)
Rising 2023 | Women in Tech Conference
16-17th Mar, 2023

Conference, in-person (Bangalore)
Data Engineering Summit (DES) 2023
27-28th Apr, 2023

Conference, in-person (Bangalore)
MachineCon 2023
23rd Jun, 2023

3 Ways to Join our Community

Discord Server

Stay Connected with a larger ecosystem of data science and ML Professionals

Telegram Channel

Discover special offers, top stories, upcoming events, and more.

Subscribe to our newsletter

Get the latest updates from AIM