The KAIROS Project has made significant progress in developing AI-based meteorological (MET) hazard forecasts for aviation. The project consists of two primary AI-driven forecasting solutions:
Solution 1: AI Convective Forecast – AI-based models predicting convective weather with high spatial and temporal resolution.
Solution 2: AI MET Applications – Expansion of AI methodologies for forecasting other aviation-related hazards, including low visibility, turbulence, high-altitude ice crystals (HAIC), and airborne hazards like SO2 and dust.
Solution 1: AI Convective Forecast
Three AI-driven convective models have been developed to meet different operational needs:
- Regional Forecast Model: Resolution: 0.25-degree Updates: Hourly, Forecast Horizon: Up to 48 hours
- National Forecast Model: Resolution: 0.125-degree Updates: Hourly, Forecast Horizon: Up to 48 hours
- Local Nowcast Model: Resolution: 1 km, Updates: Every 10 minutes, Forecast Horizon: Up to 3 hours
Solution 2: AI MET Applications
- Low visibility: The AI model for low visibility has been fully developed and is currently undergoing validation. It integrates observational data from METAR/TAF reports and satellite-derived fog detection to predict fog formation and dissipation patterns at airports.
- Turbulence: For turbulence forecasting, early models were initially constrained by limited historical datasets, but a major breakthrough occurred when the project secured access to real-time global turbulence reports through IATA’s Turbulence Aware initiative. This new data stream allows for continuous refinement of the AI model, improving its accuracy and applicability to airline operations.
- High Altitude Ice Crystals: Forecasting high-altitude ice crystals (HAIC) presents a more complex challenge due to the lack of overlapping observational datasets. The project has explored multiple sources, including Cloudnet lidar data, Rolls-Royce engine sensor data, and a satellite-based Ice Index developed by BIRA. However, inconsistencies between these sources have made validation difficult. Moving forward, KAIROS will integrate newly available real-time HAIC data to enhance model reliability.
- SO2/Dust: For SO2 and dust forecasting, the AI models currently rely on satellite-derived observations but are transitioning to live satellite data ingestion, enabling near-instant detection of volcanic ash and dust storms that can affect flight paths.
As the project moves forward, the next steps will focus on finalizing the validation of AI models through real-world trials with aviation end-users, refining forecast accuracy, and enhancing model interpretability for operational decision-making. The project will also quantify the operational benefits of AI-based forecasting, assessing its potential to reduce flight delays, optimize routing, and enhance airspace efficiency.
