{"id":"W3188169536","doi":"10.1029/2020wr029001","title":"The Abuse of Popular Performance Metrics in Hydrologic Modeling","year":2021,"lang":"en","type":"article","venue":"Water Resources Research","topic":"Hydrology and Watershed Management Studies","field":"Environmental Science","cited_by":298,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"Global Water Futures; University of Saskatchewan","keywords":"Hydrological modelling; Environmental science; Hydrology (agriculture); Computer science; Geology; Geotechnical engineering; Climatology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.002194107,0.00007488574,0.0001198081,0.0001080895,0.0004107811,0.0000275099,0.0004089621,0.00005362896,0.00008931289],"category_scores_gemma":[0.00007578354,0.00004085373,0.00003152366,0.0005434594,0.0003714284,0.00008453147,0.0007807902,0.000299569,0.0002004498],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005400832,"about_ca_system_score_gemma":0.000001943347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002204691,"about_ca_topic_score_gemma":0.0001490282,"domain_scores_codex":[0.9981713,0.0003161213,0.0002068138,0.0002537733,0.0005076693,0.0005443768],"domain_scores_gemma":[0.9995581,0.00006356499,0.00001511903,0.0003098457,0.00002005481,0.00003332219],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001671653,0.0002359104,0.6859213,0.00006482585,0.0000671281,0.0001110115,0.02146672,0.2818066,0.00512871,0.000106385,0.0004390093,0.004485276],"study_design_scores_gemma":[0.001131621,0.0003270639,0.02066166,0.00004716156,0.00002305021,0.000008369552,0.003053623,0.6861143,0.03566306,0.005460367,0.2470995,0.0004102777],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9866617,0.0004505381,0.00001816776,0.0009253046,0.00002526672,0.000114028,3.346218e-7,0.000008808397,0.0117958],"genre_scores_gemma":[0.9964637,0.000981788,0.00005259338,0.00004820879,0.00001154289,0.00002363761,0.000002004835,0.000006464109,0.002410084],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6652596,"threshold_uncertainty_score":0.3159438,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05464573691619524,"score_gpt":0.2947246862389659,"score_spread":0.2400789493227707,"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."}}