Silva, Edson (2023-11-30). Prediction of Harmful Algae Blooms Impacting Shellfish Farms in Norway. PhD thesis, University of Bergen, Norway. https://bora.uib.no/bora-xmlui/handle/11250/3104786
Summary: Harmful algae blooms (HABs) cause severe damage to the ecosystem and human health, and have significant economic impacts on shellfish farms. HAB prediction models have become increasingly popular because they can help stakeholder to take mitigation actions and reduce economic loss. Few studies have attempted to predict toxic algae species related to shellfish contamination because the time extent of data is limited and modeling the environmental response of specific taxa is complex. However, toxic algae monitoring programs have now been running for several years and have produced large datasets of toxic algae. Combined with long-time series observations by satellites and model reanalysis, we can now calibrate prediction models for toxic algae affecting shellfish farms. This thesis calibrates machine learning models to predict toxic algae impacting shellfish farms in Norwegian coastal waters for the first time. It is conducted by combining toxic algae data from the Norwegian Food Safety Authority with satellite observations of Chla concentration, Suspended Particulate Matter (SPM), Sea Surface Temperature (SST), Photosynthetically Active Radiation (PAR), and wind speed, as well as model reanalysis data of Mixed Layer Depth (MLD) and Sea Surface Salinity (SSS). Paper I demonstrates that the blooms phenology has a strong interannual variability in the North, Norwegian, and Barents Seas, which is related to the variability of the environmental ocean and atmospheric factors (SST, MLD, SPM, and winds). It implies that these variables are potential predictors for blooms in the region. Paper II exhibit that a Support Vector Machine (SVM) model can predict the presence probability of eight toxic algae on the Norwegian coast using SST, PAR, SSS, and MLD. The models can also predict the probability of harmful levels for Alexandrium spp., Alexandrium tamarense, Dinophysis acuta, and Azadinium spinosum. It can produce a climatological overview of the HABs along the Norwegian coast and provide monitoring and prediction applications. Paper III extends the SVM application to the prediction of D. acuminata abundance in a sub-seasonal range (7 -28 days) when fed with the current and past D. acuminata abundance, SST, PAR, and wind speed. The sub-seasonal forecast model is developed for the Lyngenfjord in northern Norway as a proof of concept. The probability estimates in Paper II and the sub-seasonal forecast of D. acuminata abundance in Paper III are two complementary approaches. The first is employable in the entire coast even where algae monitoring is unavailable, while the latter requires tuning to specific aquaculture farms and can achieve refined prediction. Since the SVM models are fed with data commonly available worldwide, they are portable to other regions where data from harmful algae monitoring programs are available.
Link to publication. You are most welcome to contact us or the corresponding author(s) directly, if you have questions.