{"id":"W4408423860","doi":"10.5194/egusphere-egu25-4576","title":"On the robustness of AI model forecast skill and initial condition uncertainty of the 2021 Pacific Northwest Heatwave","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Meteorological Phenomena and Simulations","field":"Earth and Planetary Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Robustness (evolution); Computer science; Climatology; Econometrics; Operations research; Geology; Economics; Mathematics; Biology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007241051,0.001093371,0.0006268618,0.000839218,0.0007274966,0.001577201,0.001209468,0.001602904,0.001290146],"category_scores_gemma":[0.03751967,0.0005293412,0.001113383,0.0004655074,0.001267015,0.002607443,0.001424493,0.002431436,0.0003424441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001056385,"about_ca_system_score_gemma":0.00153419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0627563,"about_ca_topic_score_gemma":0.02317525,"domain_scores_codex":[0.9985014,0.0004324219,0.0001319151,0.0004818144,0.0002457331,0.0002068516],"domain_scores_gemma":[0.9760272,0.01670138,0.001605958,0.002648699,0.002340769,0.0006759778],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005814762,0.0001415036,0.06414007,0.00009280632,0.0003046199,0.000091277,0.0001276138,0.9086838,0.001895074,0.001693238,0.002653085,0.01959538],"study_design_scores_gemma":[0.00003353545,0.0001017105,0.01376998,0.00003747135,0.00004863445,0.00002458317,0.00007750527,0.9818473,0.001989673,0.001612241,0.0004249667,0.00003227082],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9678344,0.001110643,0.02165451,0.001928572,0.0002598639,0.00005832277,0.001425932,0.000625168,0.005102724],"genre_scores_gemma":[0.9955506,0.0001194453,0.002441559,0.000165982,0.00005152421,0.00001287171,0.001317669,0.00005468532,0.0002856357],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0627563,"threshold_uncertainty_score":0.124782,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03063561936337044,"score_gpt":0.2556328234859691,"score_spread":0.2249972041225987,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}