{"id":"W4410536703","doi":"10.1175/mwr-d-24-0189.1","title":"The Effect of Ensemble Size on the Mean Squared Error and Spread–Error Relationship","year":2025,"lang":"en","type":"article","venue":"Monthly Weather Review","topic":"Neural Networks and Applications","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Environment and Climate Change Canada","funders":"","keywords":"Mean squared error; Statistics; Mathematics; Ensemble average; Climatology; Geology","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007777005,0.0001179083,0.0002039024,0.0000121869,0.0002908097,0.0000548497,0.0005625466,0.00002718723,0.000005520305],"category_scores_gemma":[0.00019326,0.00005453851,0.00009425767,0.0003865731,0.0000603759,0.00005327335,0.0001062021,0.0001336808,0.00001667726],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001120228,"about_ca_system_score_gemma":0.00001339457,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001160374,"about_ca_topic_score_gemma":0.00002291768,"domain_scores_codex":[0.9989763,0.0003057843,0.0002329691,0.0002176653,0.0001303568,0.0001369427],"domain_scores_gemma":[0.9964715,0.002554229,0.0001032778,0.0008159081,0.00002657204,0.0000285204],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002559294,0.00006530854,0.002926219,0.0009981189,0.0000869015,0.00000255926,0.000312723,0.0001504105,0.000194814,0.5860368,0.06151329,0.3476872],"study_design_scores_gemma":[0.0008958706,0.0006586094,0.02929284,0.009172404,0.0003316329,0.000006115014,0.00003524264,0.01699352,0.002009296,0.03845469,0.9015964,0.000553328],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.07320973,0.582342,0.01152584,0.2502376,0.0006304937,0.009162344,0.00001657352,0.0003634366,0.07251193],"genre_scores_gemma":[0.9916438,0.003556366,0.0003246483,0.002004696,0.00002310396,0.0003392098,7.962258e-7,0.000009279956,0.002098025],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9184341,"threshold_uncertainty_score":0.2236703,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02103318626119842,"score_gpt":0.2857139700948661,"score_spread":0.2646807838336677,"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."}}