FORGE: tackling four decades of technical debt in an operational forecasting system

ECMWF’s Integrated Forecasting System (IFS) has been in continuous development since the 1980s. It runs operationally every few hours, drives medium-range forecasts used across Europe, and underpins climate reanalyses and emerging hybrid AI–physics models. It also carries four decades of technical debt, must adapt to GPU-accelerated and external HPC platforms, and is being progressively open-sourced. FORGE — Forecast System Regeneration — is a medium-term programme to modernise this codebase: tackling the technical debt, refactoring model and data structures, improving the supporting software stack, and easing deployment beyond ECMWF. This talk gives an overview of FORGE and uses it as a case study, focusing on two strands of possible interest to the RSE community: a layered testing strategy spanning unit tests, integration tests, and continuous integration on full forecast experiments, with a planned end-to-end clone of the operational suite as the basis for continuous deployment; and the open-sourcing of operational forecasting software, supported by clearer code ownership, review processes, automated linting, and joint development with ECMWF’s Member States. We share what we have learned so far. Persuading the organisation that technical debt was worth addressing was a significant step. Finding time for debt work amid pressure to deliver new science remains difficult, and the debt itself makes progress slower. We are also experimenting with agentic AI code generation for behaviour-preserving refactoring, using the test harness to make faster and more automated cleanup possible.