Multi-step forecasting of chlorophyll-a concentration in coastal waters through Wavelet Dense Attention Transformer...

Harmful algal blooms (HABs) is a severe ecological disaster, and chlorophyll-a (Chla) is an important indicator reflecting the occurrence of HABs. The Wavelet Dense Attention Transformer model (WDAT) is proposed in this paper for multi-step ahead forecasting of Chla with high accuracy. Time series decoupling through Wavelet Decomposition is applied first to decompose the original Chla time series into approximation and detail components. Then, the Dense Attention Transformer model is designed to forecast these components separately. Subsequently, the prediction results are reconstructed through inverse wavelet transformation to achieve multi-step forecasting. Experiments are conducted in the coastal areas of Wenzhou, China, and Pago Pago, American Samoa. By comparing with other models, the results show that WDAT can accurately predict the complex dynamics of Chla.