Conference Agenda

ESA's trio of Swarm satellites, launched in 2013, and the twin satellites of Macau Science Satellite-1 (MSS-1), launched in 2023, offer an unprecedented oppportunity to study Earth's magnetic environment. Through polar orbits and global coverage, Swarm’s goal is to better understand whole-Earth processes: the geodynamo, the lithosphere, mantle conductivity and global currents in the magnetosphere and ionosphere. By contrast, through low-inclination orbits, MSS-1 maps the equatorial geomagnetic field and local-time phenomena, seeking to better understand the South Atlantic Anomaly, the equatorial lithosphere, and space-weather hazards. Although each mission is scientifically valuable in its own right, taken together they offer new opportunities for synergistic scientific exploration of our planet's magnetic environment. I will summarise how these new datasets guide research into our planet's magnetic field, and some of the latest scientific insights.

Dragon-6 STAI4CH project aims to demonstrate optical and radar EO data capabilities to provide essential information for archaeological and cultural heritage mapping and monitoring applications. Novel Artificial Intelligence (AI) and Deep learning (DL) methods are being developed to process EO data to efficiently map waterways and waterscapes of ancient Empires in China, Italy and Iraq, and detect physical transformations and land cover changes due to modern development and urbanisation that might impact the heritage assets.

During the second year of activity, STAI4CH’s team focused on enhancing the developed land cover mapping and change detection methods, which exploit semantic segmentation Neural Network (NN) based on EO self-supervised learning and change detection NN based on multi-source EO data, respectively. The workflows mainly used high resolution Copernicus Sentinel-2 surface reflectance imagery acquired in 2017-2025 over a test area encompassing a transect of the UNESCO World Heritage Site (WHS) of the Grand Canal of China, including the Jiaxing-Hangzhou section of Jiangnan Canal and the Hangzhou Xiaoshan-Shaoxing and Shangyu-Yuyao sections of Zhedong Canal. The input data selection approach adopted 6 month-long temporal sampling, to better depict the landscape during the rainy (June to August) and dry (October to February) seasons. A staged training framework method for EO data semantic segmentation was developed. Dynamic World land cover datasets were exploited for training and validation of the land classification workflow. Methodological enhancements involved the incorporation of self-supervised algorithms (MAE-style pretraining for remote sensing representation initialization), and the possibility to adaptively address class imbalance, geographic coverage bias, and weak-label noise through performance-aware guided sampling and selective supervision, towards implementation in Open Geospatial Engine (OGE). Imagery acquired by Copernicus Sentinel-1 SAR and the Chinese sustainable development science satellite SDGSAT-1 were also collected to be embedded into the methodological workflow. Under the best configuration, the proposed framework achieved 49.28% mIoU, outperforming the official Google Dynamic World product (44.77%) by 4.51 percentage points.

While algorithmic development progressed, data analysis activities over the waterscapes of the UNESCO WHS of the Ahwar marshes in Iraq were further advanced by enhancing the heritage asset mapping approach via interpretation of NDVI time-series from Landsat and Sentinel-2 imagery, Mann–Kendall tests and landscape metrics. This enabled the mapping of traditional open surface canal irrigation systems and their surrounding landscapes, which will act as a baseline knowledgebase to validate the NN-based methods, and the identification of vulnerable irrigation systems and traditional field zones experiencing vegetation loss at a higher rate than the wider landscape. Moreover, Sentinel-2 spectral signatures at training sites within the area of Ostia-Portus in central Italy were analysed with the aim to assess the potential of high resolution imagery to provide helpful insights into the identification of land with buried archaeological assets, towards the development of semi-automated spectral anomalies recognition methods using AI/DL.

The role of the Chinese and European Young Scientists continued to be pivotal for the algorithmic development and experiments in China, as well as for the archaeological mapping in Iraq and spectral anomalies detection in central Italy. Detailed results from their scientific contributions to the project will also be presented during the YS poster session.

Future work will focus on the implementation of the developed AI/DL-based methods across the whole site of the Grand Canal, the Ahwar marshes in Iraq and the waterways of the Tiber river in Italy, part of the Historic Centre of Rome UNESCO WHS. An analysis of the technical performance of the methods will be carried out, to quantify their reliability and exportability to different geographical environments, and identify pathway for further algorithmic improvement.