{"paper":{"title":"Beyond the Training Data: Confidence-Guided Mixing of Parameterizations in a Hybrid AI-Climate Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A neural network for convection predicts its own error to mix with a physics scheme, enabling stable hybrid climate simulations.","cross_cats":[],"primary_cat":"physics.ao-ph","authors_text":"Helge Heuer, Julien Savre, Manuel Schlund, Mierk Schwabe, Tom Beucler, Veronika Eyring","submitted_at":"2025-10-09T11:44:47Z","abstract_excerpt":"Persistent systematic errors in Earth system models (ESMs) arise from difficulties in representing the full diversity of subgrid, multiscale atmospheric convection and turbulence. Machine learning (ML) parameterizations trained on short high-resolution simulations show strong potential to reduce these errors. However, stable long-term atmospheric simulations with hybrid (physics + ML) ESMs remain difficult, as neural networks (NNs) trained offline often destabilize online runs. Training convection parameterizations directly on coarse-grained data is challenging, notably because scales cannot b"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The NN parameterization predicts its own error, enabling mixing with a conventional convection scheme when confidence is low, thus making the hybrid AI-physics model tunable with respect to observations and reanalysis through mixing parameters. This improves process understanding by constraining convective tendencies across column water vapor, lower-tropospheric stability, and geographical conditions... In AMIP-style setups, several hybrid configurations outperform the default convection scheme (e.g., improved precipitation statistics). With additive input noise during training, both hybrid and pure-ML schemes lead to stable simulations and remain physically consistent for at least 20 years.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The neural network trained on adjusted ClimSim data with subtracted radiative tendencies can be transferred to ICON-A despite distribution shifts, and its self-predicted confidence score reliably indicates actual inference errors in the new model to enable effective mixing.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A ClimSim-trained physics-informed NN convection parameterization is transferred to ICON-A with confidence-guided mixing to a conventional scheme, yielding stable 20-year AMIP simulations that outperform the default convection scheme in precipitation statistics.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A neural network for convection predicts its own error to mix with a physics scheme, enabling stable hybrid climate simulations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ff53f59eb0b50aab8bc14336d05a64032a0eb8912a93be07f6bdeeacbd7b1c22"},"source":{"id":"2510.08107","kind":"arxiv","version":5},"verdict":{"id":"7f94971c-5c67-4e4d-b8b6-a8cd97e9621c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T09:14:08.117630Z","strongest_claim":"The NN parameterization predicts its own error, enabling mixing with a conventional convection scheme when confidence is low, thus making the hybrid AI-physics model tunable with respect to observations and reanalysis through mixing parameters. This improves process understanding by constraining convective tendencies across column water vapor, lower-tropospheric stability, and geographical conditions... In AMIP-style setups, several hybrid configurations outperform the default convection scheme (e.g., improved precipitation statistics). With additive input noise during training, both hybrid and pure-ML schemes lead to stable simulations and remain physically consistent for at least 20 years.","one_line_summary":"A ClimSim-trained physics-informed NN convection parameterization is transferred to ICON-A with confidence-guided mixing to a conventional scheme, yielding stable 20-year AMIP simulations that outperform the default convection scheme in precipitation statistics.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The neural network trained on adjusted ClimSim data with subtracted radiative tendencies can be transferred to ICON-A despite distribution shifts, and its self-predicted confidence score reliably indicates actual inference errors in the new model to enable effective mixing.","pith_extraction_headline":"A neural network for convection predicts its own error to mix with a physics scheme, enabling stable hybrid climate simulations."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2510.08107/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}