A Deep Learning-Based Model for Secondary Prediction on Deep-Sea Collector Plumes

In recent years, deep-sea mining has received renewed attention. However, sediment plumes will be generated by deep sea mining, which may be a huge problem. And the evolution and transport of sediment plumes generated by seafloor mining activities are essential to the assessment of environmental impacts. While lacking of studies on plume transport and dispersion, the results of numerical simulations are still not consistent with the observations very well. In this study, a deep learning model SEDSEQ is constructed using plume observations, numerical simulation results fusing with other multi-source data to make secondary predictions of deep-sea collector plumes. The optimization effect of the model was evaluated using valid datasets with various evaluation metrics. The results show that the plume concentration predictions optimized by the SEDSEQ model are significantly better than the results of numerical simulations, which is improved by 25% at least. For different stations, the results of the model optimization are also evaluated. The results of the study indicate that the SEDSEQ can successfully optimize the results of plume model, providing a new method and idea for plume simulation.