Application of the Most Sensitive Initial Error in Sampling Scheme Design for a Mobile Platform

Abstract

Since ocean mobile observation equipment and numerical models have achieved remarkable results, the combination of the two has become an influential topic. A numerical model provides auxiliary information for the arrangement of observation equipment. As feedback, observations help produce more accurate initial conditions when incorporated into data assimilation. However, it is still worth investigating the ways to select the most valuable observation sites within the computational domain and the ways to design the observation scheme of the mobile platform. To improve the efficiency of observation, researchers attempt to select the best observation region in the vast ocean. The approach of deploying additional observing assets in selected regions is referred to as targeted observation. By combining the features of the targeted observation and the mobile observing platform, we propose a design approach for the observation scheme. First, based on a model, we estimate the initial perturbation that causes the greatest change in the sea surface temperature in the future. Then, according to the spatial component of the perturbation, we divide the experimental regions into sensitive regions and non-sensitive regions. Observing system simulation experiments are carried out to verify that samples in sensitive regions are more helpful to improve model prediction. Afterward, considering the variation of the perturbation with time, we propose a hybrid sampling scheme design method for an underwater unmanned vehicle combining Q-learning and particle swarm optimization algorithm. Finally, the effectiveness of the hybrid algorithm is verified by comparing the sampling schemes designed in static environment. This approach provides a dynamic basis for the path planning of mobile observing platforms.