{"id":"W2060054786","doi":"10.1175/jcli-d-13-00457.1","title":"Surface Temperature Probability Distributions in the NARCCAP Hindcast Experiment: Evaluation Methodology, Metrics, and Results","year":2014,"lang":"en","type":"article","venue":"Journal of Climate","topic":"Climate variability and models","field":"Environmental Science","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Hindcast; Climatology; Environmental science; Skewness; Climate model; Climate change; Percentile; Meteorology; Statistics; Geography; Mathematics; Geology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01859007,0.000954121,0.0006518441,0.001426653,0.0005599273,0.001219575,0.001129503,0.001132398,0.0008839816],"category_scores_gemma":[0.0242056,0.0002867119,0.0008982893,0.0008971646,0.0009126111,0.001469353,0.001312288,0.001010117,0.0001548361],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001735283,"about_ca_system_score_gemma":0.0009802404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01152737,"about_ca_topic_score_gemma":0.01086775,"domain_scores_codex":[0.9952121,0.00251593,0.0003110685,0.0007438763,0.00100273,0.0002142811],"domain_scores_gemma":[0.9815408,0.01085898,0.001534563,0.002512484,0.003012315,0.000540892],"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":[0.003860652,0.002963563,0.2248369,0.0003183007,0.0008071357,0.0002254689,0.0003862969,0.6528324,0.01592732,0.005926021,0.003863247,0.0880528],"study_design_scores_gemma":[0.0002456017,0.002558667,0.05890096,0.00002596525,0.00008323227,0.00006348684,0.0002708768,0.9213756,0.01310008,0.001726728,0.001573772,0.00007504343],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9393432,0.0001027664,0.05383938,0.000107662,0.00004842097,0.0009714989,0.003248163,0.0006248223,0.001714205],"genre_scores_gemma":[0.9393325,0.00003670649,0.0546078,0.0000657878,0.00002120686,0.000947634,0.004370958,0.00008908733,0.0005283779],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01859007,"threshold_uncertainty_score":0.09831482,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09548134239871503,"score_gpt":0.3552115101846428,"score_spread":0.2597301677859278,"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."}}