Search for a command to run...
Underwater environments pose significant challenges for navigation and mapping, in large part due to the attenuation of electromagnetic radiation in water. This limits high-resolution sensing to short ranges, so that large-scale mapping generally requires acoustic systems with lower resolutions than vision or lidar. Fully-submerged platforms lack continuous GNSS, with similar acoustic-based technologies being costly and time consuming. This leads to heavy reliance on dead reckoning for underwater navigation, which in turn is prone to cumulative errors and drift. In confined overhead environments like natural caves, the difficulties inherent in underwater sensing, perception, and control and the limited mitigating strategies available have traditionally restricted mapping to human divers. This paper details the use of the SUNFISH<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">®</sup> AUV in large-scale, georeferenced 3D underwater mapping within the overhead environment of the Floridan Aquifer System (FAS). The SUNFISH AUV is a compact, person-portable (55 kg) vehicle with 6-DOF control, enabling maneuverability in complex, obstaclefilled environments. It integrates standard dead-reckoning and mapping sensors and specialized payloads, including a continuous nitrate sensor. The AUV supports a spectrum of autonomy, from servo-level remote control to autonomous exploration and decision-making. To produce accurate georeferenced 3D maps, we employ a three-stage augmentation to basic dead reckoning: (1) RTK GNSS and inertial initialization at the surface for a georeferenced start; (2) SLAM updates based on multibeam data to create self-consistent maps and bound dead-reckoned drift; and (3) post-processing with underwater control points from GNSS surface fixes and diver-deployed magnetic beacon radiolocation fixes to counter remaining errors. These strategies enhance georeferencing accuracy to decameter-level horizontal error or better over kilometer scales. In 2024 and 2025, the SUNFISH AUV successfully mapped two natural cave environments in the FAS. At Jackson Blue Spring, it produced a 3D volumetric map of 420 m of cave conduit with nitrate, salinity, and temperature data at 1 m resolution. At Lafayette Blue Springs, 5.1 km of a branching network was surveyed, collecting extensive water quality data and achieving a 7.5 m georeferenced horizontal <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1 \sigma$</tex> error at 50 cm resolution or better. The project demonstrated the AUV's capability to collect georeferenced 3D maps and high-resolution physical, hydrological, and chemical data, which is invaluable for understanding karst aquifer dynamics and informing water quality management in these environments.