Large-Scale Ocean Environment Field Perception for Underwater Robots with Limited Measurements

Accurate perception of large-scale ocean environment fields is essential for marine monitoring, resource management, and climate research. However, direct observation is severely limited by the sparse, asynchronous, and resource-constrained measurements with mobile platforms such as underwater gliders and robots. To overcome this challenge, we propose a perception framework tailored for underwater robots operating with limited measurements. The method integrates three components: spatiotemporal decomposition to obtain compact and interpretable ocean features; multi-objective sparse sampling that jointly optimizes information gain, spatial coverage, and depth anchors; and spatiotemporal fusion to reconstruct full fields from sparse in-situ data. Experiments on a Pacific Ocean temperature dataset confirm that with only 1% sampling sparsity, the reconstructed fields achieve an average error below 0.5 ° C, while preserving thermocline variability and mesoscale structures. These results demonstrate that the proposed method enables efficient, accurate, and scalable ocean environment perception, offering significant potential for future autonomous missions with underwater robots.