Tap Artificial Intelligence to predict weather, address climate change

The Weather Company, an IBM Business, sets great by store by Artificial Intelligence (AI) which can be a game-changer in tracking/forecasting weather events in the context of climate change/global warming.

"After all, the physics of the atmosphere doesn’t change. It's being governed by the same factors today and tomorrow," explains Kevin Petty, Director, Science and Forecast Operations at The Weather Company,

Capturing weather events

What we need to recognise here is that weather in different parts of the world will change from what it has been and that could mean more severe thunderstorms or precipitation events than in the past.

"I guess we have generally done well in the past, but we need to do is to do a better job of capturing such events so that we can better inform our clients," Petty told BusinessLine in an exclusive chat in New Delhi recently.

The Weather Company is launching later this year the Global High-Resolution Atmospheric Forecasting System (GRAF) model capable of predicting a thunderstorm virtually anywhere on the planet every hour.

"We’re looking to see how best we can capture these high impact events, thunderstorms or tropical cyclones or what have you," he said in response to a question on harsh monsoon events in India in 2018 and 2019.

There’s a continual of processing of learning involved in which specific algorithms update themselves to capture more and more of these changes as they happen over a particular place/region under reference.

This aspect of continual learning addresses the question of climate change. These algorithms can be trusted to capture these events better and come out with better forecasts to meet the needs of the people.

How long does the model to learn to make it a reliable forecast? It depends on the type of event that one is chasing, says Petty.

A caveat is in order here. The AI type of model is not going to forecast what it hasn’t seen before. So, it may have a hard time capturing that outlier event and making learnings from it.

AI for forecasts

Basically, two or three years of data should help one to capture these things into the model. It can get better with time, but one can't put a number to the level of protection achievable during that period.

The goal is and should always be perfect. One does not achieve unless ones push oneself in the right direction, Petty said commenting on the research efforts at his company.

There are two areas where he specifically pays thrust on - AI for forecasting science and AI for application development. Existing models have their own strengths and weaknesses and provide only a particular answer.

But if you’re able to combine these models into one output intelligently, you can improve the production beyond any single model, Petty said.

"If we slide over to the application part - because this is where you drive power from weather data - the weather data or weather forecasts on their own have no intrinsic value," Petty pointed out.

One can as much data one wants, but if it doesn't allow him/her to make proper decisions to protect oneself, family, property or to operate business/industry, it has no value.

"But if I combine these data using AI to derive specific correlations, I can now provide decision support capability to sectors ranging from agriculture to aviation."