Ca-STANet: Spatiotemporal Attention Network for Chlorophyll-a Prediction With Gap-Filled Remote Sensing Data

Long-term chlorophyll-a (Chl-a) prediction has the potential to provide an early warning of red tide and support fishery management and marine ecosystem health. The existing learning-based Chl-a prediction methods mostly predict a single point or multiple points with monitoring data. However, the monitoring data are subject to sparse sampling and difficult to be measured in a large-scale and synchronous way. Moreover, the advanced learning-based models for point Chl-a prediction, such as long short-term memory (LSTM) and convolutional neural network (CNN)-LSTM, are unable to fully mine the spatiotemporal correlation of Chl-a variations. Therefore, by using the satellite remote sensing data with extensive coverage, we design a framework, namely, Ca-STANet, to simultaneously predict the Chl-a of all the locations in a large-scale area from the perspective of spatiotemporal field. Specifically, in our method, the original data are first divided into multiple subregions to capture the spatial heterogeneity of large-scale area. Then, two modules are, respectively, operated to mine the spatial correlation and long-term dependency features. Finally, the outputs from the two modules are integrated by a fusion module to fully mine the spatiotemporal correlations, which are exploited to attain the final Chl-a prediction. In this article, the proposed Ca-STANet is comprehensively evaluated and compared with the legacy methods based on the OC-CCI Chl-a 5.0 data of the Bohai Sea. The results demonstrate that the proposed Ca-STANet is highly effective for Chl-a prediction and achieves higher prediction accuracy than the baseline methods. Moreover, as the OC-CCI Chl-a 5.0 data have many missing areas, we introduce DINEOF method to fill the data gaps before using them for prediction.