Environmental prediction of enterococci analytical states across sampled coastal locations and laboratory batches in the United States
Więcej
Ukryj
1
Professional School of Environmental Engineering, Faculty of Renewable Natural Resources, National Agrarian University of La Selva, Carretera Central km 1.21, Tingo Maria, Huanuco 10131, Peru
2
Faculty of Renewable Natural Resources, Universidad Nacional Agraria de la Selva, Carretera Central km 1.21, Tingo María 10131, Huánuco, Peru
Autor do korespondencji
Alberto Franco Cerna Cueva
Professional School of Environmental Engineering, Faculty of Renewable Natural Resources, National Agrarian University of La Selva, Carretera Central km 1.21, Tingo Maria, Huanuco 10131, Peru
SŁOWA KLUCZOWE
DZIEDZINY
STRESZCZENIE
Quantitative polymerase chain reaction measurements of enterococci often include non-detects and numerical estimates below the lower limit of quantification; collapsing these analytically distinct records into a single quantitative outcome can obscure censoring. This study evaluated whether concurrent environmental conditions predict enterococci analytical states and whether predictive performance transfers to unseen sampling locations and laboratory batches. Record-level field and laboratory data from the United States National Coastal Condition Assessment 2020 included 897 eligible marine events and a secondary contrast of 420 Great Lakes events. Penalized logistic regression was evaluated using repeated five-fold cross-validation grouped first by stable location and then by complete analytical batch, with all preprocessing estimated within training folds. A secondary multinomial model retained non-detect, below-quantification estimate, and unflagged numerical states. In marine events, the environmental model reduced log-loss relative to a prevalence-only comparator by 7.58% under location-held-out validation and 6.94% under batch-held-out validation; corresponding areas under the receiver operating characteristic curve were 0.699 and 0.692, and 95% UNIQUE_ID-cluster bootstrap intervals for absolute improvement excluded zero in both schemes. The three-state model retained relative log-loss reductions of 7.21% and 5.11%, respectively. A prespecified boosting model did not improve batch-held-out performance and was not retained. Performance varied among regions and did not transfer robustly to Great Lakes batches. Concurrent environmental conditions therefore provided moderate predictive information about enterococci analytical states in sampled marine events, but performance remained insufficient for operational use and depended on analytical censoring, laboratory batch, and geographic domain.