Artificial Intelligence approaches for predicting Harmful Algal Blooms (HABs)

Abstract

HABs can produce toxins, which accumulate in filter-feeding shellfish and intoxicate human consumers. The toxins are heat stable and so can't be destroyed by freezing and/or cooking. Under current regulations shellfish toxin monitoring is effectively retrospective: regulators sample, await results, and if the regulatory threshold is breached there is an investigation into the amount of shellfish harvested since the sample was taken, which might then result in a full-scale food chain product recall. By gathering high resolution field monitoring data using novel qPCR and lateral-flow (LF) techniques, we planned to refine and validate a computer model for predicting HABs caused by Dinophysis species. One of the aims of the project was to use the higher resolution data collected as part of the project to train the model towards a more accurate forecast in respect of breaches in the Dinophysis toxin threshold up to 6-8 weeks ahead. The model would then aid planning decisions for harvesting and will save costly recalls and protect human health (in this case from Diarrhetic Shellfish Poisoning - DSP). Other strands of the project consisted of use of a Novel monitoring tools, a qPCR for quantifying HAB cell abundance in seawater, and a Lateral Flow testing for quantifying Dinophysis toxins in shellfish, directly in the field. Field data from these novel methods will be validated by an accredited light microscopy technique which enables the cell densities to be quantified in water and by liquid chromatography with tandem mass spectrometry (LC-MS/MS) for validating the shellfish flesh test results from the field.