As a critical indicator of atmospheric water vapor content, precipitable water vapor (PWV) holds significant scientific value in global water cycle studies, climate change research, and extreme weather monitoring. Current PWV observation techniques have limitations in terms of balance accuracy, spatiotemporal resolution, and data continuity. This study addresses the high data loss rate and systematic bias issues in the FY-4A PWV product by proposing a dual-scale spatiotemporal fusion model based on machine learning, achieving high spatiotemporal resolution reconstruction of FY-4A PWV over China. The constructed dual-scale reconstruction framework consists of two key stages: the first stage integrates FY-4A PWV and ERA5 data to generate PWV products at a 0.25° resolution hourly; the second stage introduces GNSS PWV, land cover type and normalized difference vegetation index to further downscale and produce a refined PWV dataset at a 4 km resolution hourly. The 10-fold cross-validation of the two-scale model had an R ² of 0.98 and a root mean square error (RMSE) of 2.06 mm. The independent tests based on radiosonde PWV had an R ² of 0.98 and an RMSE of 2.53 mm. The spatial distribution analysis shows that the reconstructed PWV data can not only accurately reflect the macro trend of regional water vapor distribution in China, but also show fine spatial heterogeneity in small areas of complex terrain. This study provides a novel method for optimizing the FY-4A PWV product and offers high-precision data support for regional climate research and extreme weather monitoring.

Large-scale and precise underwater localization in open seas can significantly enhance the operational performance of autonomous underwater vehicles (AUVs) across multiple application scenarios. While long base line systems offer accuracy, their high cost and complex deployment are limiting. Single-beacon positioning is more cost-effective but suffers from insufficient accuracy and a lack of real-time capability. To overcome these limitations, this paper proposes a novel real-time AUV localization method integrating a single acoustic beacon, dual transducers, and a strapdown inertial navigation system (SINS). The core innovation utilizes the geometric constraints derived from the acoustic distance difference between the single beacon and the two transducers on the AUV to achieve initial 2D positioning. This is enhanced to 3D by incorporating depth sensor data. The position solutions are fused with SINS data using an extended Kalman filter to ensure system stability and robustness. Comprehensive validation incorporates numerical simulations—evaluating varying depths and sound speed profile errors—and semi-physical simulations driven by South China Sea field data, using ultra-short baseline positioning as a reference. Results confirm robust convergence against sensor noise, installation biases, and severe deep-water geometric degradation. The field-data validation further demonstrates high precision, showing a positioning error of ⩽8 m at an acoustic slant range of 16 km and ⩽7 m at a range of 8 km. This method provides a cost-efficient and practical solution for high-precision, large-scale deep-sea AUV navigation.

This paper presents a novel sound velocity profile (SVP)-free positioning algorithm for ultra-short baseline (USBL) systems, addressing the critical challenge of accurate underwater localization in scenarios where real-time SVP measurements are impractical or unavailable. The proposed method formulates positioning as a joint estimation problem, simultaneously recovering the target’s three-dimensional coordinates and a path-averaged effective sound velocity parameter. This unified framework theoretically compensates for first-order ray-bending effects without requiring explicit ray tracing or prior knowledge of the local sound speed field. To address the numerical ill-conditioning intrinsic to USBL geometries, a stabilized Levenberg–Marquardt solver is adopted. Experimental validation conducted in a controlled anechoic water tank demonstrates that the algorithm achieves sub-decimeter positioning precision, with mean three-dimensional deviations from the epoch-averaged position ranging from 2.08 cm to 5.34 cm and standard deviations up to 2.84 cm. The effective sound velocity is estimated with a standard deviation within 2.60 m s − 1 , and the relative positioning repeatability error (defined as RMSE divided by slant range) remains below 0.5% across all tested configurations. The algorithm exhibits computational efficiency suitable for real-time implementation. From an engineering perspective, this SVP-free approach offers significant potential operational advantages: it eliminates the logistical burden and inaccuracies associated with continuous sound velocity profiling, and maintains computational efficiency suitable for real-time autonomous underwater vehicle navigation.