An approach of multi-element fusion method for harmful algal blooms prediction

The harmful algal blooms (HABs) are an issue of concern for water management worldwide. Effective strategies for monitoring and predicting of HAB spatio-temporal variability in waterbodies are more essential. To promote the monitoring and predicting of HABs, we proposed a multi-element fusion prediction (MEFP) method for cyanobacteria bloom. Considering the impact of surrounding factors for HAB occurrence, the proposed MEFP fuses multiple exogenous factors to enhance the prediction accuracy in different environments. Specifically, MEFP adopts a dual-sides network that parallelly captures the potential outbreak patterns on the numerous input features. The restricted Boltzmann machine is utilized to optimize the processing of parameter initialization. Subsequently, the attention mechanism is introduced in the post-network stage to establish the contextual relationship between the current and historical temporal information. The experimental results on the real-world dataset demonstrate the proposed MEFP model outperforms other benchmark methods.