AI Falls Short on Extreme Weather Forecasting

Syllabus: GS1/ Geography, GS3/ Science and Technology

Context

• A recent study found that while Artificial Intelligence (AI)-based weather forecasting models have achieved remarkable accuracy and speed, they continue to underperform in predicting record-breaking extreme weather events.

Traditional Model of Weather Prediction

• Traditional weather forecasting uses numerical weather prediction (NWP) models. 
• The model simulates atmospheric processes using equations of fluid dynamics and thermodynamics. 
• They simulate atmospheric circulation, heat transfer, moisture movement, cloud formation, precipitation processes, and ocean-atmosphere interactions.

Weather forecasting in India

• India, at present, depends on satellite data and computer models for weather prediction. The Indian Meteorological Department (IMD) uses the INSAT series of satellites and supercomputers.
• In India three satellites, INSAT-3D, INSAT-3DR and INSAT-3DS are used mainly for meteorological observations. 
• Forecasters use satellite data around cloud motion, cloud top temperature, and water vapor content that help in rainfall estimation, weather forecasting, and tracking cyclones.

Prediction of Weather with AI Models

• Unlike traditional weather models that rely on the laws of physics, AI-based models begin with data. 
• These models use machine learning algorithms to identify patterns and learn relationships between input variables—such as temperature, humidity, wind speed—and resulting weather events like cyclones or heavy rainfall. 
• The India Meteorological Department has established a dedicated functional group on the “Application of AI/ML in Weather and Climate.” 
  • Currently, AI models being used experimentally include; Pangu-Weather, GraphCast and FourCastNet.
• IMD has also employed AI-based Advanced Dvorak Technique (AiDT) for estimating cyclone intensity.

Advantages of AI Models in Weather Forecasting

• Ability to Use Big Data: AI models can process massive datasets from satellites, radars, weather stations, and even social media, allowing them to detect subtle signals and trends.
• Handling of Nonlinear Systems: AI models have the potential to uncover hidden patterns and nonlinear cause-effect relationships among Earth system variables that physics-based models may overlook.
• Adaptability to Local Conditions: AI allows for region-specific models that account for local geographical, topographical, and climatic factors, improving forecast relevance.
• Real-time Forecasting: AI is capable of rapid “nowcasting” — forecasting weather within the next few hours — which is crucial for disaster preparedness and urban planning.

Why Do AI Models Struggle with Extreme Events?

• Dependence on Historical Data: AI models are trained using historical weather records and they identify statistical patterns and relationships within past observations.
• The Extrapolation Problem: AI models excel at predicting events that fall within the range of previously observed weather conditions (interpolation).
  • They struggle to forecast unprecedented or record-breaking events that lie outside their training data (extrapolation). 
  • As a result, extreme heatwaves, cold waves, or wind storms are often underestimated.
• Lack of Physical Understanding: AI models learn correlations between variables rather than the underlying physical laws governing the atmosphere. This limits their ability to anticipate novel weather scenarios.
• Climate Change: AI models trained on today’s climate data may become less effective in a warmer future, as the atmospheric system continues to evolve due to climate change.
• Black Box Nature of AI Models: AI systems, particularly deep learning models, operate as “black boxes”, meaning their decision-making processes are opaque.
  • This hinders trust and interpretability, especially among non-experts and operational meteorologists.

Initiatives taken to improve the efficiency

• Mission Mausam: It was launched to upgrade the capabilities of India’s weather department in forecasting, modelling, and dissemination. The objectives of the mission are;
  • Develop Cutting Edge Weather Surveillance Technologies & Systems
  • Implement Next-generation radars, and satellites with advanced instrument payloads
  • Develop improved earth system models, and data-driven methods (use of AI/ML).
• The ‘National Monsoon Mission’ was set out in 2012 to move the nation over to a system that relies more on real-time, on-the-ground data gathering.
• Doppler radars: The IMD is also increasingly using Doppler radars to improve efficiency in predictions. The number of Doppler radars has increased from 15 in 2013 to 50 in 2026. 
  • Doppler radars are used to predict rainfall in the immediate vicinity, making predictions more timely and accurate.
• The Ministry of Agriculture & Farmers Welfare have initiated the weather information network and data system (WINDS) under which more than 200,000 ground stations will be installed, to generate long-term, hyper-local weather data. 

Source: TH