A prototype coupled modeling approach for predicting harmful algal blooms: A case study in Chile

Predicting harmful algal blooms (HABs) remains a major challenge for coastal management and aquaculture. This study compares three forecasting approaches developed under the Monitoring of Algal Blooms in Chile (MACH) project: a particle dispersion model, an LSTM neural network, and an empirical dynamic model (EDM) to evaluate their ability to forecast bloom events. Consequently, we applied the EDM to forecast two Pseudo-nitzschia species groups using data collected from Metri, Quellón, and Melinka in southern Chile. The results showed that the genus Ceratium and Leptocylindrus were commonly associated with both Pseudo-nitzschia species groups, and the best prediction by causal species was obtained for the P. seriata group, with a correlation coefficient of 0.733 ( P < 0.0001) between observed and predicted values. This case study demonstrated that species interactions can be used to predict specific HAB species; however, the prediction performance may vary depending on location and species. This study provides one of the first applications of EDM for HAB forecasting using causal species in a real-world monitoring context, demonstrating the potential of hybrid modeling frameworks to improve early warning systems and mitigate aquaculture losses. • Predicting the dynamics of harmful algal blooms (HABs) is invaluable to ensure the protection of fish and shellfish farms as well as humans. • HAB prediction models can be categorized into empirical-statistical, mechanistic, and physical models. • Species interaction can be used to predict a specific HAB species however the prediction results could vary according to location and species. • Coupling several models may be used to improve forecasting and predictive performance, robustness, and reliability.