Informative path planning in complex marine environments: Gap imputation and variance minimization in satellite data

Satellite remote sensing plays a critical role in operational oceanography by providing Near Real Time (NRT) measurements of essential climate variables (ECVs), such as Sea Surface Temperature (SST) and Salinity with extensive spatial coverage compared with in-situ measurements. However, these parameters are obtained through passive measurements with satellite data that often suffer from attenuation due to atmospheric effects, resulting in missing data in satellite recordings. To overcome this issue, we propose a novel approach towards intelligent autonomous measurement acquisition that outlines the potential use of Unmanned Surface Vehicles (USVs). The aim is to measure at locations with missing data by combining information-driven exploration and energy-efficient path planning. Information-driven exploration refers to identifying potential measurement points with high informativeness. We use differential entropy as information metric based on a Gaussian Process Regression (GPR) model that uncovers the variances in the satellite data. Subsequently, we plan a path of an USV by extending the Rapidly-exploring Random Tree Star (RRT^*) algorithm by introducing an energy-dependent cost function considering Sea Surface Currents (SSC) to accurately capture the complex marine environment. By this, we introduce a modified information metric trading out informativeness and energy consumption in the path planning. The results demonstrate that informative and energy-efficient path planning can significantly reduce the USV’s energy consumption, with potential savings of up to 88%.