A generic approach for the development of short-term predictions of E. coli and biotoxins in shellfish
Aquaculture Environment Interactions
Reason for embargo
Currently under an indefinite embargo pending publication by Inter Research. No embargo required on publication.
Microbiological contamination or elevated marine biotoxin concentrations within shellfish can result in temporary closure of shellfish aquaculture harvesting, leading to financial loss for the aquaculture business and a potential reduction in consumer confidence in shellfish products. We present a method for predicting short-term variations in shellfish concentrations of Escherichia coli (E. coli) and biotoxin (okadaic acid, its derivates dinophysistoxins and pectenotoxins). The approach is evaluated for two contrasting shellfish harvesting areas. Through a meta-data analysis and using environmental data (in situ, satellite observations and meteorological nowcasts and forecasts) key environmental drivers were identified and used to develop models to predict E. coli and biotoxin concentrations within shellfish. Models were trained and evaluated using independent datasets and the best models were identified based on the model exhibiting the lowest root mean squared error (RMSE). The best biotoxin model was able to provide one-week forecasts with an accuracy of 86%, a 0% false positive rate and a 0% false discovery rate (n = 78 observations) when used to predict the closure of shellfish beds due to biotoxin. The best E. coli models were used to predict the European hygiene classification of the shellfish beds to an accuracy of 99% (n = 107 observations) and 98% (n = 63 observations) for a bay (St Austell Bay) and an estuary (Turnaware Bar), respectively. This generic approach enables high accuracy short-term farm-specific forecasts, based on readily accessible environmental data and observations.
This work was carried out as part of the UK Biotechnology and Biological Science Research Council (BBRSC) and UK National Environmental Research Council (NERC) funded ‘ShellEye’ project (BB/M026698/1)
This is the author accepted manuscript.