Google is using AI to monitor flood in India, pilot project Patna called a success


ndia accounts for nearly 20 per cent of the flood-related fatalities in the world. Nearly 107,487 people have died due to heavy rains and floods in India over a span of 64 years between 1953 and 2017. And the floods are only getting worse year after year. Is there something that can be done? While the Indian government is working on solutions that will mitigate effect of flood, Google has deployed its artificial intelligence (AI) systems and machine learning to predict floods with better accuracy, which can give people more time to prepare for it and migrate to a safer location.

Google has started Flood Forecasting Initiative in India, which aims to develop an ecosystem that predicts floods and informs people before such a natural calamity strikes so that loss to life and property can be reduced. The company piloted a program in Patna last year which was able to predict floods and the regions that it would be affected due to the natural disaster with an accuracy of over 90 per cent.

Achieving such a high accuracy is no ordinary feat, as it requires a combination of factors including a deep understanding of the topography, tons of historic data and real-time information that needs to be processed together to predict the region that would be affected by the floods. Google achieved this by combining the data from government agencies that provide on-ground information from measuring devices placed on the spot and thousands of images of the flood prone areas captured by various satellites orbiting the Earth.

The tech giant then ran hundreds of thousands of simulations on its machine learning models — such as the hydrologic model and the hydraulic model — to predict the flow of water in a particular region and create accurate flood forecasts, which in this case was near Patna.

Now, all this sounds quite simple. But in reality, the company faced a lot of challenges, both technological and regulatory while collecting data for its flood forecast model and broadcasting this information out to the public. While the company collaborated with governments and purchased and aggregated the distributed data to overcome some of the technological hurdles, to overcome the regulatory obstacles, the tech giant showed the governments results based on historic data and piloted in a small region to build the trust among other things.

“Part of what the biggest technological challenge here is to do something that is automated but is complex enough to work anywhere in the world… that is something that we are still working on,” Software Engineering Manager at Google, Sella Nevo told the India Today Tech. “On the regulatory partnership side, I think we are on the first steps of a long process to persuade governments to trust us with such an important and sensitive system.”

Once the company’s ML models made predictions about the floods, it collaborated with government agencies and local NGOs to impart this information to the people in the region where the company’s AI had predicted would be affected by the floods. Google has a separate interface for government agencies like the Central Water Works Commission in India, NGOs and international agencies like the UN and the Red Cross society, where it gives away more detailed information regarding its forecasts. In addition to this, Google also informed people using Google Maps, Google Search and Android alerts. However, these alerts contain simplified information that can be easily understood by the general public.

In low connectivity areas, however, Google used a combination of three methods to inform the residents about the possible disaster. First, the tech giant provided information to the government so that it can roll out information to the public. Second, it partnered with NGOs, in this case: SEEDS in Patna, that have dozens of workers on ground who can spread the word as quickly as possible. Lastly, the company made all its alerts publicly accessible.

“We make these alerts publicly available and allow even commercial entities to use it so thatwe are hoping that other organisations will help to fill whatever gaps are left after our efforts,” said Nevo.

Patna pilot was a success and now Google is planning to scale its operations and launch its flood monitoring initiative in many locations near the Ganges and Bhramaputra. “We have expanded around Patna and we now have fairly large area around Guwahati We now cover around six times the area we did last year,” Nevo said.

But India is not the end, Google aims to deploy this system globally and it is focusing on the countries in the South East Asia region, which will be picked based on the number of fatalities and people affected. “Our high priority countries include Bangladesh, Indonesia, the Philippines, Thailand and Vietnam. Those are the areas where we hope to launch in the future,” said Nevo.