How is AI in India developing the future of weather prediction?

  • AI in India is reshaping weather forecasting, tackling extreme weather with advanced predictions.
  • Global meteorology shifts to AI, enhancing forecast accuracy, exemplified by China’s Fengwu model.
  • AI in weather forecasting boosts disaster preparedness and resource management, both in India and globally.

India is experimenting with AI to develop climate models, enhancing weather predictions. This is a crucial step as the country faces increasing torrential rains, floods, and droughts, according to a senior weather official.

In recent years, global warming has intensified weather system conflicts, leading to rising numbers of extreme weather events in India. The independent Centre for Science and Environment reports that these events have caused nearly 3,000 fatalities in 2023 alone.

Globally, weather agencies are increasingly focusing on AI for its potential to reduce costs and enhance prediction efficiency. The Met Office in Britain has said that AI could ‘revolutionize’ weather forecasting, and a recent Google-funded model has demonstrated superior performance to traditional methods.

The critical role of accurate forecasts in India, and the use of AI

Accurate weather forecasts are vital for India, home to 1.4 billion people, many of whom live in poverty. Being the world’s second-largest producer of critical crops such as rice, wheat, and sugar, the stakes are particularly high.

The India Meteorological Department (IMD) currently relies on mathematical models run on supercomputers for its forecasts. The integration of AI with an expanded observation network is expected to produce higher-quality forecasts at a reduced cost.

K.S. Hosalikar, the head of climate research and services at IMD, told Reuters that the department is developing AI-based climate models and advisories intending to enhance forecasting accuracy.

The IMD has already implemented AI to generate public alerts for heatwaves and malaria-related diseases. Hosalikar mentioned plans to expand weather observatories to the village level, enabling more detailed data collection for improved forecasting.

Recently, the Indian government expressed its intention to blend AI with traditional forecasting models and is establishing a center dedicated to exploring this approach through workshops and conferences.

Saurabh Rathore, an assistant professor at the Indian Institute of Technology-Delhi, highlighted the cost-effectiveness of AI models. Unlike traditional methods, these models don’t require the expense of running supercomputers and can be operated from high-quality desktop computers.

However, experts emphasize the need for better data to fully leverage AI’s capabilities. Parthasarathi Mukhopadhyay, a climate scientist at the Indian Institute of Tropical Meteorology, explained that spatially and temporally high-resolution data is essential for AI models to effectively enhance the location-specific accuracy of existing forecasts.

International perspectives: AI in global weather forecasting

Outside of India, China has also announced its utilization of AI to refine weather forecasting, particularly for weather disasters exacerbated by heatwaves, heavy rains, and typhoons. The Shanghai Artificial Intelligence Laboratory, in collaboration with China’s National Meteorological Center and Shanghai Meteorological Service, employed its AI-driven Fengwu meteorological model to enhance typhoon predictions.

The accuracy of Fengwu in forecasting Typhoon Doksuri, China’s strongest typhoon of the summer, was notably superior to that of both the European Centre for Medium-Range Weather Forecasts (ECMWF) and the US National Centers for Environmental Prediction (NCEP). Similarly, for Typhoon Khanun, Fengwu’s predictions outperformed those of ECMWF and NCEP.

Developed by various Chinese institutions and published in April, the Fengwu model uses multimodal and multitask deep learning technology. It delivers high-resolution atmospheric forecasts over ten days and operates efficiently on a single graphics processing unit. This model can generate ten-day global weather forecasts in just 30 seconds, significantly improving on traditional models that rely on supercomputers.

While AI weather prediction shows considerable promise, researchers acknowledge the need for further improvements. Their goal is to achieve district-level accuracy and eventually extend these predictions to the street level.

The AI model can do 10 days’ weather forecast in just 10 seconds. (Source – X).

AI models like Fengwu and Pangu Weather, another model developed in China, are enhancing the efficiency of weather forecasting. Although predicting weather remains inherently challenging, these AI models are expected to complement traditional physical models. They provide valuable insights for various sectors and Earth science research, supporting initiatives in carbon neutrality, disaster prevention, and energy security.

Despite the success of traditional numerical weather prediction, its progress is hampered by the slow growth of computing power and the complexities of physical models. AI forecasting methods, which have lower computational costs, are seen as a solution to these challenges.

Pangu Weather, developed by Huawei Cloud and recognized in a publication by Nature, employs a 3D neural network and a hierarchical temporal aggregation strategy to process complex meteorological data. Its accuracy surpasses that of some European and American meteorological centers, and the model is accessible online to users worldwide.

The Fengwu and Pangu models have demonstrated their capabilities in recent typhoon predictions. China’s Central Meteorological Observatory plans to continue integrating AI into weather forecasting, particularly for typhoon monitoring. This will involve collaboration with various universities and research institutions to enhance global weather prediction and services.

Private sector engagement in AI weather services

Companies in other Asian countries, such as Thailand and Vietnam, increasingly use AI to protect clients from weather-related disasters. Following a flash flood in 2021 that caused significant damage, an electronics factory in Thailand’s Bangpoo industrial park enrolled in a pilot forecast service offered by Weathernews, a Japanese weather firm. Started early in 2023, this service provides real-time, hyperlocal forecasts, predicting weather changes within three hours, significantly improving the Thai Meteorological Department’s daily regional forecasts.

Weathernews’ AI system collects and analyzes data to provide these precise forecasts, alerting clients to potential squalls and floods. This lets businesses take preventative measures such as erecting barriers or relocating equipment. Additionally, the company is collaborating with local authorities to install radar systems in Thailand, intending to match the forecast accuracy achieved in Japan.

Weathernews fully launched its AI-based forecasting service in Thailand in March, and in Vietnam in June, becoming the first in Asia to set up its own equipment and offer such a service. Primarily serving logistics providers, the company aims to expand its client base to 500, including electronics and auto manufacturers, to increase its annual revenue in Thailand and Vietnam to 3 billion yen (approximately US$22.6 million).

Chihito Kusabiraki, the company’s president, aims to boost overseas revenue to 70% to 80%, up from the current 40%. This expansion is crucial as Asian countries, characterized by dense populations and slow advances in disaster management, are particularly susceptible to natural disasters. According to the Asian Development Bank, developing Asia accounted for 76% of disaster victims and 25% of the global monetary damage from natural disasters between 2010 and 2020, highlighting the critical need for advanced forecasting services in these regions.

In 2021, weather and water-related hazards resulted in damages amounting to US$35.6 billion in Asia. Notably, the Philippines was ranked as the most disaster-prone country in the WorldRiskIndex, underscoring the region’s vulnerability to such events.

Startups are tapping into weather-related opportunities across Asia as well. California-based Atmo is developing an early warning system for flash floods and cyclones in Indonesia and is currently in discussions with neighboring countries. Meanwhile, Tokyo-based Spectee is utilizing social media data for disaster mapping in the Philippines and is planning to establish a local unit to extend the services already provided to 700 clients in Japan.

Since 2018, China’s Ninecosmos has advised companies, including China’s COSCO Shipping, on optimal shipping routes based on weather conditions. Additionally, it has developed an air pollution forecasting service, further expanding its range of meteorological solutions.

However, the success of these weather-related services in Asia hinges on developing affordable, locally tailored solutions. Products designed for advanced economies may prove too costly for emerging Asian markets, emphasizing the need for region-specific adaptations.