we used to go there

and play with our friends.

We used to have a lot of fun there.

Then the river came and washed everything away.

Over the last century, floods have become the most common and deadly natural disaster on the planet.

The Ghagra River takes everything.

Our village had no warning.

It would help if we were notified

but we get nothing.

Many countries currently lack effective early warning systems and alerts.

20 percent of flood fatalities occur in India.

WHEN RIVERS RISE

[radio announcer in progress] --see some of the highest rainfalls

during the monsoon season.

But this year, it has been exceptional.

- One of the most important variables during a crisis

is reliable information.

There's so much work that's already been done,

by governments, by the UN and NGOs,

and we're trying to learn from that work, build on that,

and assist them with their goals.

In India, where we're running our first pilot program,

the government has thousands of people who are measuring the water level

in rivers across India,

with what is effectively very long measuring sticks,

which are called stream gauges,

every hour.

Which allows them to know whether the river

will overflow and flood.

But it doesn't yet allow them to understand

exactly what areas are going to be affected.

What neighborhoods, or even what villages.

We received a warning from the head of the village

in the morning, just a few hours before the flood hit.

I wish I had known earlier.

- The response time is most crucial thing.

Reducing the response time always plays a very important and vital role

in the whole disaster management framework.

Advancement in technology would help us better

in spreading this message faster.

- Flood forecasting was a very exploratory project.

The big technical question was,

can we have enough information

to try to do forecasting that would be accurate enough

to make a difference.

Starting with the basic needs for getting information about what's going on,

where is it happening,

what should I be doing?

- To be able to provide a forecast in real time,

we rely on the government we're working with.

We complement their effort

by adding accurate modeling to that process.

We start by collecting thousands of satellite images

to build a digital model of the terrain.

Based on these maps,

we generate hundreds of thousands of simulations

of how the river could possibly behave.

We receive the measurements from the government,

and cross those measurements with our simulations.

We can then send those forecasts to individuals

using Search, Maps, and Android notifications.

This is an example of an alert that we can produce.

This alert is for Pedana.

This alert is actually--has over 90 percent accuracy.

The lead time is incredibly important.

- Yeah. - Even more than the specifics.

- Any information we can give, they will use.

- Want to show them the alert?

Talking to people who've directly experienced severe floods,

and understanding what they really need, is incredibly important

for us to know how we can really help.

This is an opportunity to collaborate on a global scale.

We are in a very exciting time,

to use technology to try to make a difference.

I can't imagine a greater privilege

than to do work we love, which is develop technology,

and do it in a way that could actually

help people directly in a very profound way.

Our hope for the future is to give people

a few more days of warning before a flood occurs.

And to use AI to scale this, and provide these forecasts

anywhere in the world, wherever people need them.

The Crisis Response and Research team is exploring ways to use AI to provide earlier warnings

for other natural disasters including fires and earthquake aftershocks.