{"id":"W4403684139","doi":"10.1002/hyp.15288","title":"Setting expectations for hydrologic model performance with an ensemble of simple benchmarks","year":2024,"lang":"en","type":"article","venue":"Hydrological Processes","topic":"Hydrology and Watershed Management Studies","field":"Environmental Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"National Oceanic and Atmospheric Administration","keywords":"Hydrological modelling; Simple (philosophy); Computer science; Hydrology (agriculture); Environmental science; Geology; Climatology; Geotechnical engineering","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":[0.009111631,0.0009710218,0.0007338774,0.0008534339,0.0004718796,0.001395922,0.001024945,0.0009674827,0.00140025],"category_scores_gemma":[0.03042984,0.0003573476,0.0007277261,0.0007380465,0.0004806224,0.001628035,0.001000501,0.001184302,0.000257732],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001409486,"about_ca_system_score_gemma":0.0009242182,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01041424,"about_ca_topic_score_gemma":0.007007239,"domain_scores_codex":[0.9972716,0.001405904,0.0001468743,0.0004194699,0.0004513571,0.0003048064],"domain_scores_gemma":[0.9866626,0.008122572,0.0007422331,0.001497844,0.002495085,0.0004797635],"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":[0.0003744873,0.0001311691,0.01572264,0.00003201576,0.0001216448,0.00003598325,0.00003341496,0.9728645,0.0007930483,0.001133302,0.001092422,0.007665318],"study_design_scores_gemma":[0.00003264983,0.000174968,0.005043108,0.00001077762,0.00002397104,0.000009584709,0.00003548445,0.9907658,0.002415182,0.001282845,0.0001879936,0.00001771702],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9337531,0.0001305078,0.05683301,0.0002882774,0.00006487946,0.0001845813,0.001497831,0.000999222,0.006248618],"genre_scores_gemma":[0.9915337,0.00001154861,0.007354309,0.00003611469,0.000005555126,0.00007757648,0.0007444356,0.00005352555,0.0001832776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01041424,"threshold_uncertainty_score":0.04818749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01862025629561084,"score_gpt":0.2527740476381288,"score_spread":0.234153791342518,"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."}}