Modeling and Forecasting Vibrio Parahaemolyticus Concentrations in Oysters

Vibrio parahaemolyticus (V.p) is an epidemiologically significant pathogen that thrives in coastal waters where oysters are harvested, posing high risks to human health and shellfish industry and requiring effective forecasting models for emergency preparedness and interventions. This study sought to develop forecasting models with differing lead times, which are able to predict the level of V.p in oysters in advance to mitigate the health risk to the general public and the economic loss to the shellfish industry. The Random Forest method along with 227 sampling datasets from two different geographic locations were utilized to: (1) Identify the most critical environmental predictors controlling the level of V.p in oysters, (2) Select the most important time lags for the environmental predictors as model input variables, and (3) Develop four forecasting models (RF-1Day, RF-2Day, RF-3Day, and RF-4Day) with the lead time of one to four days. The uncertainty involved in model predictions was quantified using the bootstrapping method. Results showed that V.p abundance in oysters is controlled by antecedent environmental conditions 1-11 days before. The antecedent environmental conditions can be described using time-lagged Sea Surface Temperature (SST) and salinity. The V.p abundance can well be forecasted 1 - 4 days in advance using the four models. The performance of the models decreases with increasing lead time. The RF-3Day and RF-4Day models can be employed primarily for emergency preparedness due to their relatively long lead time while the RF-1Day and RF-2Day models can be used primarily for management interventions due to their relatively high predictive performance. (C) 2020 Elsevier Ltd. All rights reserved.

Automated summary

A study developed forecasting models with different lead times using the Random Forest method to predict the level of V.p in oysters, based on data from two geographic locations. Results showed that antecedent environmental conditions significantly influence V.p abundance, and the models can effectively predict V.p abundance 1-4 days in advance, with model performance decreasing as lead time increases.