{"id":"W4414947457","doi":"10.1007/s12145-025-02033-2","title":"Benchmarking coastal boundary datasets in deep learning applications","year":2025,"lang":"en","type":"article","venue":"Earth Science Informatics","topic":"Coastal and Marine Dynamics","field":"Earth and Planetary Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Moncton","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Benchmarking; Deep learning; Workflow; Metric (unit); Boundary (topology); Flexibility (engineering); Standardization","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004879341,0.001977103,0.0007861924,0.003164971,0.0008250217,0.002146122,0.002786205,0.002068098,0.00232975],"category_scores_gemma":[0.01285149,0.0003993338,0.00136823,0.003468534,0.001062156,0.002719138,0.002767571,0.001536805,0.001642876],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001439475,"about_ca_system_score_gemma":0.00107218,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01152142,"about_ca_topic_score_gemma":0.01467728,"domain_scores_codex":[0.9972016,0.000730868,0.0003814012,0.0007853008,0.0006971351,0.000203793],"domain_scores_gemma":[0.996465,0.001167473,0.0002849986,0.001003714,0.0009366194,0.000142156],"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":[0.001007887,0.0013269,0.03684524,0.001752363,0.0008472145,0.0005836519,0.0003832354,0.5170537,0.01083742,0.007060939,0.08982117,0.3324803],"study_design_scores_gemma":[0.0001576211,0.0002972842,0.01562667,0.0003615684,0.00009812025,0.0001836391,0.0005300958,0.9290444,0.02073119,0.008941414,0.02395134,0.00007668486],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6935027,0.005911663,0.1804585,0.002318213,0.001011693,0.0009692002,0.07130685,0.02795701,0.0165643],"genre_scores_gemma":[0.6340881,0.0009355582,0.1734559,0.0006772958,0.00008675928,0.0006822141,0.1859436,0.001140866,0.0029897],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01152142,"threshold_uncertainty_score":0.0258047,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00487903907673902,"score_gpt":0.2199935137247885,"score_spread":0.2151144746480494,"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."}}