2 days ago
Forewarned is fore-armed: Adopt AI for better climate forecasting
For much of modern history, weather forecasting has battled public scepticism. In 19th century England, Admiral Robert FitzRoy's early forecasts were mocked as 'guesses dressed up as science" and umbrellas would come out only when sunshine was predicted.
Until recently, India's own meteorological service faced similar mistrust, with people joking that IMD stood for 'It Might Drizzle.' However, this is beginning to change. India now boasts of satellite infrastructure for climate and weather monitoring that ranks among the world's most advanced.
INSAT-3D, 3DR and newly launched 3DS satellites monitor the atmosphere, land and oceans. These will be complemented by NISAR, an upcoming Nasa-Isro mission, which will track ecosystem shifts, ice melt, sea levels, groundwater and natural hazards.
India also has 17 earth observation (EO) satellites, including Resourcesat, Oceansat, Cartosat and Scatsat, which track land use, agriculture, forest cover, water resources and ocean conditions. Megha-Tropiques and SARAL, part of a collaborative mission with France, aim to keep tropical weather systems, sea levels and ocean circulation under watch.
The National Information System for Climate and Environment Studies (NICES) is responsible for integrating satellite data to create long-term climate records and generate variables like snow cover, vegetation indices and glacial lake characteristics.
These efforts support the forecasting and tracking of critical climate events. Despite such capabilities, India is at the front line of the global climate crisis, grappling with flash floods, droughts, cyclones and heatwaves.
Also Read: Time for the Global South to leverage DPI for climate action
With a fragile Himalayas range and a long coastline, India's geography amplifies its exposure to climate risks.
A 2021 study by Council on Energy, Environment and Water revealed that over 80% of our population resides in districts acutely vulnerable to climate shocks, while much of the country lacks the adaptive capacity to withstand them.
India has seen average temperatures rise, with longer and more intense heatwaves. By 2065, these could last up to 25 times longer, posing severe health and agricultural threats. Meanwhile, erratic rainfall and a surge in dry spells have pushed nearly 600 million Indians into water stress.
These challenges demand that we re-assess our climate preparedness and explore technologies such as artificial intelligence to prepare better.
To achieve our net-zero target for emissions by 2070, climate risks must be assessed across a range of future economic scenarios and their associated carbon footprints. Traditional climate models offer limited simulations but machine learning can vastly enrich our modelling.
Also Read: How ISRO's partnership with NASA will boost India's space industry
Traditional climate models rely on complex equations to simulate the planet's systems. Globally, AI is now being used in three key ways to enhance climate modelling.
First, AI-based emulators like QuickClim and ACE can mimic traditional models by learning patterns of emissions, climate outcomes and their links, enabling faster and cheaper simulations.
QuickClim, developed by the Commonwealth Scientific and Industrial Research Organisation, used 15 emulators trained in 30 minutes each to make accurate temperature projections. ACE, developed by Allen Institute for AI, outperformed conventional atmospheric models on speed and precision.
Second, foundation models like ClimaX go beyond replication to uncover new patterns in climate data, offering deeper insights into global climate dynamics using observations that go back more than a century.
Also Read: Space mission Axiom 4: The universe can be a family one day
Third, hybrid models blend machine learning with physics-based approaches to improve accuracy in complex areas such as cloud and snow formation.
Projects like CLiMA (under Caltech, MIT and Nasa JPL), Nasa and the European Space Agency's Earth system 'digital twins' and the EU's DestinE initiative aim to create simulations to better predict climate events and minimize uncertainty.
Similarly, AI is transforming weather forecasting by producing accurate predictions up to 10,000 times faster than traditional models, while reducing energy and computing costs.
Trained on decades of weather data, AI models can quickly analyse complex atmospheric patterns and deliver near-real-time forecasts. Notable innovations include Google DeepMind's GraphCast, which uses historical and current data for six-hourly weather forecasts.
Huawei's Pangu-Weather delivers accurate weekly forecasts within seconds by processing all weather variables. Nvidia's Earth-2 combines AI and simulation tools, such as FourCastNet, to model extreme weather events such as hurricanes.
While these are encouraging developments, a key challenge in AI-based prediction remains AI's 'black box' problem, which makes it difficult to trust or fix AI tools.
This has drawn the scepticism of climate scientists, who still lean towards traditional models because of the complexity of Earth's systems and uncertainty over future scenarios. Although AI-based models show promise, their ability to forecast an evolving climate scenario remains uncertain.
Also Read: Space race: Is competition among Indian startups ready for lift-off?
India has made significant strides in satellite technology and climate data collection, but the threats of climate change demand a more forward-looking approach.
Indian research organizations and leading tech companies must step up to the task of improving the speed and accuracy of climate and weather forecasting. This is vital to our long-term climate resilience.
The authors are, respectively, associate director at The Convergence Foundation, and an economist.
