The TEFRA (Testing Eruption Forecasting through Research and Assimilation) project will establish the first open, community-scale competition in volcanic eruption forecasting, designed to rigorously produce and evaluate competing approaches to eruption prediction using modern, open-access datasets.

TEFRA is supported by the Gordon and Betty Moore Foundation.

The project includes both the infrastructure for the competition and funds for research teams to develop the models. Funds will be available for teams to do the necessary research based on a proposal call in Northern hemisphere Spring 2027. The competition and research team funding will run from approximately September 2027 through June 2029.

​

Although volcanic eruptions are among the few natural hazards that have proven potential for successful forecasts, forecasting today remains highly dependent on expert interpretation and volcano-specific experience rather than reproducible, quantitative scientific frameworks. All but a few of the hundreds of eruptions over the last decades have occurred without a quantitative forecast (e.g., a 50% chance of eruption in the next two weeks). We aim to transform this landscape by using the SZ4D network of scientists to organize an international forecasting competition in which independent teams will generate and apply statistical, physics-based, and hybrid AI/data-assimilation methods to a common suite of target volcanoes and standardized datasets. By directly comparing forecasting strategies against the same eruptions, observational constraints, and evaluation metrics, TEFRA creates comparable models that can determine where approaches succeed or fail, and why.

​

The competition structure is the scientific core of the project. Teams will develop and test competing forecasting models on a carefully selected suite of well-observed volcanic systems spanning contrasting eruption styles and forecasting challenges, including closed-vent, open-vent and hydrothermal systems, which are known to have different eruptive styles and precursors. Forecasts will be evaluated quantitatively and compared to a benchmark model on the accuracy of the eruption onset, eruption magnitude, intensity, style, vent location and duration. Annual workshops open to the broad community and coordinated benchmarking exercises will ensure iterative comparison, refinement, and cross-fertilization between teams while maintaining independent model development. The result will be a suite of operational volcano forecasting models that have been tested retrospectively and are thus ready to be put to the test of true prospective prediction.