Artificial Intelligence (AI) and Machine Learning (ML), including Deep Learning, Computer Vision and Remote Sensing techniques, are increasingly used across the geosciences for hazard forecasting, Earth observation, environmental monitoring, climate services and other applications that may inform operational and policy decisions. Examples such as Google Flood Hub, GraphCast, ECMWF’s AIFS, post-disaster damage assessment pipelines and geospatial foundation models show the growing potential of these approaches. However, strong performance on standard benchmarks does not automatically make an AI-derived product suitable for high-stakes decisions.
This session asks what AI-enabled geoscience needs to demonstrate before it can be relied upon in operational, policy or governance contexts. We welcome contributions that move beyond conventional performance metrics and address reliability, uncertainty, transparency, robustness and practical usability. Technical approaches may include Physics-Informed Neural Networks, uncertainty quantification methods such as Conformal Prediction or Bayesian Neural Networks, and Explainable AI approaches assessed against the information needs of end-users and decision-makers.
Relevant contributions may examine validation against rare or out-of-distribution events, uncertainty communication, robustness across locations and time periods, data quality and bias, reproducibility, provenance and traceability, and approaches to operational readiness, including Technology Readiness Level frameworks. Contributions based on systems tested or used in practice are particularly encouraged, including early warning systems, automated hazard mapping, environmental monitoring, climate adaptation tools and Earth observation applications for disaster risk reduction.
The session will also address the governance and societal implications of AI-derived scientific evidence, including standards and regulatory frameworks, human oversight, algorithmic bias, equity, public trust, stakeholder participation, accountability and the potential for AI-supported decisions to affect communities differently.
The aim is to connect geoscientists developing AI methods with those working on how scientific evidence is assessed, communicated and governed. The central question is not only whether an AI model works, but whether there is sufficient transparent, robust and actionable evidence to justify its use in real decisions.