Here's how the fastest supercomputer in India helps predict the weather

In computers, the term FLOP stands for Floating Point Operations Per Second, telling you exactly how many operations a processor is capable of performing each second. A teraflop means one million million FLOPs, while a petaflop represents one thousand million million FLOPs. Remember this as you read further.

Around 2007-08, the Indian Institute of Tropical Meteorology (IITM) was conducting weather research using a 7 teraflop computer. In 2018, it uses the power of India's fastest supercomputer, named Pratyush, and its whopping 6.8 petaflops of power. That's 6800000000000000 FLOPs, in case you were wondering.

It's mind-boggling math, Isn’t it?

So much power

Contrary to what many believe, Pratyush is composed of two different High Performance Computing (HPC) units. The name Pratyush is given to the unit at IITM, while there's another unit (called Mihir), at the National Centre for Medium Range Weather Forecasting (NCMRWA) in Noida, Uttar Pradesh. Together, the two HPCs, contribute 4 petaflops and 2.8 petaflops respectively, bringing the total power up to 6.8 petaflops.

This is the fourth fastest supercomputer in the world in the weather and climate domain.

And all of this power is used to predict the weather as accurately as possible.

Why weather?

Weather in India is no small talk. The inclement weather events are fast mutating the economic misfortunes of the country. The Indian monsoon is a critical driver for the country's farm-dependent $2 trillion economy, as at least half the farmlands are heavily dependent on rain.

With the recent spate of unpredictable weather events, climate change stakes are becoming real for India. According to an economic survey published in 2017, these weather events are costing India anywhere between $9-10 billion annually and climate change is projected to impact agricultural productivity with increasing severity from 2020 to the end of the century.

And to make matters worse, the Indian monsoon is one of the most complex climate systems of the world. The geographic location of India make the system more difficult to simulate and predict.

Pratyush with its super computational capacity, though, has been able to make it significantly accurate. Its vast computing ability helps map regions of the country at a 12km resolution for weather forecasts.

Why do you need a supercomputer to predict weather?

A few years ago, scientists at IITM had to wait for about six hours for calculations to be completed. With their brand new computer, this takes four hours. "Weather and climate forecasting are computationally intensive," says Dr. Suryachandra A. Rao, Project Director at Pratyush.

You're probably wondering why two hours matter, right? Well, it likely doesn't. Not without context anyway. The supercomputer allows scientists to predict the weather significantly more accurately and in lesser time, which is what matters.

Weather forecasting is done using equations, Rao explained. These equations are solved at different imaginary points on the globe. Earlier, the IITM used to predict data using points placed at every 200 square kilometre, which was later brought down to 100 sq. km. With the help of Pratyush, they have now brought the distance down to 12 square kilometres.

These points are akin to pixels on a camera sensor, represented by megapixels. Higher megapixel count (theoretically) allows better quality images. Similarly, closer the dots are placed, more accurate the weather forecast can be. And just like pixels are to cameras, this is just one aspect of weather forecasting.

Furthermore, the Earth's atmosphere consists of different layers, 64 to be precise, and the equations have to be run for each of these layers. So, the equations aren't just performed at each dot, but at different layers on these dots. The dots are spread over the entire globe, meaning the actual number of combinations is unimaginable.

In fact, Rao says there are 40 levels of the ocean that aren't considered at the moment. To take those into account, the computational power required will be even more, and weather prediction would be more accurate too.

See why all that power is required? No? There's more.

Scientists don't just run the calculations once. That would be most inefficient. The calculations are run 20 to 30 times, changing variables like temperature, humidity etc. This is done to reduce uncertainty and keep predictions as accurate as possible.

But the supercomputer is just a kernel of the weather forecasting process. Reams of data that is sputtered out is then matched with the human computational genius, when chief scientist Rao and his team rack their heads to interpret and compare this data.

A rare scene where big data and humans do not bristle, but align to reach a common goal.

With a supercomputer worth a whopping Rs. 450 crores (approx. $65 million) brought on and tasked to do weather forecasting, weather conversations should bump itself up from their ‘small talk’ status.