Comparisons with sonar data show that validation choices strongly change the reported accuracy of satellite depth maps. It helps show when satellite maps suit reconnaissance and when local measurements are needed for navigation.
Satellite-derived bathymetry of Arabian Gulf ports under leakage-free validation: a multi-site benchmark of empirical, machine-learning and deep-learning models against 285,000 multibeam soundings, with an outlook for UAE waters · Muhammad Saeed, Fatema Al Hajri, Mohamed H. Alhosani et al. · arXiv:2610.33609
Stable ground objects such as buildings act as reference points to check the BIOMASS radar satellite's measurements over time. A technique that detected a significant measurement anomaly helps keep the first BIOMASS mission's forest-change data reliable.
Assessing BIOMASS Radiometric Stability using Permanent Scatterers · Marco Manzoni, Francesco Salvaterra, Claudio Maria Prati et al. · arXiv:2610.37400
An open-source algorithm improves how accurately special calibration photos are used to assess the James Webb Space Telescope's mirror alignment. It gives scientists more precise optical information and may let the telescope spend less time calibrating and more time observing.
Model-Based Retrieval of Four Years of JWST/NIRCam Wavefronts · Shrishmoy Ray, Benjamin Pope, Louis Desdoigts et al. · arXiv:2610.40187
A small neural network extracts wave information from high-frequency radar signals and gives more accurate estimates of Significant Wave Height. It offers a cost-effective way to improve wide-area ocean wave data for maritime safety and climate monitoring.
Spatially-Adaptive Neural-Network Based Extraction of Significant Wave Heights from HF Radar Data · Senal Chandrasekara, Reza Shahidi · arXiv:2610.36301
Visual AI models are probed and even strong scene describers often struggle to track objects once they are out of view. It matters for remote sensing AI that aims to build a continuous, predictive understanding of a changing planet from satellite images.
Probing Spatial State Representations in Visual Foundation Models · Baiqiao Yin, Zhongang Cai, Ruisi Wang et al. · arXiv:2610.40636
Foundation models, large AI systems used for mapping and change detection, can be trained with one unified algorithm rather than multiple stages. This could simplify development and improve performance of Earth observation models that help us understand our changing planet.
Training Foundation Models with a Single Algorithm · Yuda Song, Alvin Zhang, Fahim Tajwar et al. · arXiv:2610.40564
TEMPEST is a fast computer model that predicts surface temperatures on rough, irregular asteroids and moons without atmospheres, including light bouncing and heat reflection. It could help scientists interpret space mission data and learn where water ice might hide on distant worlds.
TEMPEST - A modular Python-based thermophysical model for rough, irregularly-shaped airless bodies · D. G. Lyster, C. J. A. Howett · arXiv:2610.40406
Machine learning models accurately predicted marine sediment properties from continuous sensor measurements alone, while expert geological labels did not improve learning and could degrade it. Raw satellite data might hold more detail than human categories, potentially enabling more precise, automated understanding of Earth's features.
Expert lithological labels do not improve learning when used to predict marine sediment properties · John M. Aiken, William Gilpin, Thorsten W. Becker · arXiv:2610.36410
GPS technology at the Rubin Observatory precisely measures atmospheric water vapor, which can distort light from distant stars, allowing corrections for accurate observations. This continuous monitoring could keep astronomical observations, including supernova cosmology, as accurate as possible.
Measuring Precipitable Water Vapor through Dual-Band GPS at Rubin Observatory · Anja Weyant, W. M. Wood-Vasey, Elana K. Urbach et al. · arXiv:2610.40363
Multiple ceiling-mounted radar sensors fused data into 3D point clouds and analyzed movement patterns in real time to identify human actions without cameras. This could offer a privacy-conscious monitoring solution for assisted living as demand for discreet technology grows.
Human Activity Recognition using Ceiling-Mounted Multi-Radar Sensors · Ananya Prabhu, Aditi Site, Tarmo Lipping · arXiv:2610.39628