Geostatistical uncertainty maps for real-world efficient AUV data collection

Abstract

Autonomous Underwater Vehicle (AUV) trajectory planning for oceanographic surveys is challenging and requires comprehensive and efficient data collection for enhanced mission success. By strategically navigating and targeting high-value data points, the AUV can operate longer and gather more essential information for numerical ocean model calibration. Here, we propose a geostatistical modelling workflow with two complementary objectives. First, to jointly predict ocean temperature and spatial uncertainty maps, representing regions with limited knowledge about the ocean properties of interest, from where optimized navigation paths can be devised and updated. Second, to efficiently assimilate the collected data and update an ocean model with the new data. An autonomous oceanographic survey performed off W. Portugal illustrates the proposed modelling workflow. We use the CMEMS product of Atlantic-Iberian-Biscay-Irish-Ocean Physics Analysis and Forecast as a priori and conditioning data of the spatial predictions. During the survey, the data acquired by the AUV are assimilated and used in new geostatistical predictions for the day after the data acquisition. The results show that the proposed methodology efficiently predicts daily ocean temperature and its spatial uncertainty, allowing data assimilation from different sources (i.e., numerical models of ocean dynamics and AUV sampling). This approach enables the assimilation of AUV measurements and the model prediction to have higher value and greater reliability.