Sébastien Chognard

Dr. Sébastien Chognard

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
10 March 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
CPCarole Planque
Carole Planque
RLRichard Lucas
Richard Lucas
DCDaniel Clewley
Daniel Clewley
SCSébastien Chognard
Sébastien Chognard
Gregory GiulianiGregory Giuliani
Gregory Giuliani
CHClaire Horton
Claire Horton
The UN General Assembly (2015) emphasizes sustainable pathways to enhance resilience for people and nature, with future development driven by data and evidence. Sustainable development frameworks (e.g., the UN 2030 Agenda and the 2016 Paris Climate Agreement) highlight the importance of data and evidence in assessment and decision-making that respects national policies and priorities. Global advances in Earth observation (EO) data provision and digital solutions that increase efficiencies, timeliness, and affordability are making major contributions. However, many existing platforms rely on externally hosted cloud infrastructures and generic global classifications of environments that may not align with domestic statutory definitions, limiting national control over data governance, methodological standards, and regulatory reporting. These constraints have raised growing concerns regarding data and technological sovereignty for countries seeking authoritative, policy-ready environmental information. Using Wales (United Kingdom; UK) as an exemplar, this study showcases the design and implementation of a flexible, sovereign National Digital Infrastructure (NDI) that uses the Open Data Cube (ODC) to apply Living Earth, a novel and customizable approach for EO-focused environmental monitoring. Outputs are time series of land cover and habitat maps and change products, including post-event (e.g., fire, flood) management, which address key policy requirements and support land and water resource management (from freshwater to marine environments), while ensuring public dissemination. Major advantages include the sharing of consistent datasets across governments and partner organizations, minimizing duplication of effort, improving transparency, traceability, and reproducibility, fostering collaboration between diverse stakeholders and communities, promoting inclusivity in environmental management decision-making, and supporting sustainable outcomes.

2 March 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
PTPablo Timoner
Pablo Timoner
CPCharlotte Poussin
Charlotte Poussin
CPCarole Planque
Carole Planque
RLRichard Lucas
Richard Lucas
ZFZhengpeng Feng
Zhengpeng Feng
Gregory GiulianiGregory Giuliani
Gregory Giuliani
Geospatial embeddings (geo-embeddings) derived from large-scale Earth Observation Foundation Models offer new opportunities to increase the accuracy and temporal consistency of land cover maps in support of environmental monitoring. The research aimed to evaluate the potential of classifying environmental descriptors (EDs) from the TESSERA and AlphaEarth geo-embedding products and, from these, construct land cover maps according to the Food and Agriculture Organisation (FAO) Land Cover Classification System (LCCS). Random Forest classifiers were calibrated separately for three contrasting regions in Switzerland, Brazil and Spain using geo-embeddings values as predictors and specific land cover reference data as target variables. Model performance was assessed against stratified training and testing datasets, with hyperparameter optimization conducted through cross-validation. Robust and consistent predictive performance was observed across the three regions that were comparable when using TESSERA and AlphaEarth geo-embedding products. Accuracies exceeding 90 % were observed for the LCCS natural vegetation and water classes, but were lower for artificial surfaces (F1-score ~ 0.68-0.86) and bare soil (F1-score ~ 0.77-0.84) classes. Importantly, the cultivated vegetation class predicted by both models showed improved agreement with in situ observations compared to the aggregated Copernicus Crop Type product. Performance differences across regions were likely to be primarily influenced by reference data characteristics rather than landscape complexity. ESA WorldCover proved to be a reliable proxy reference dataset where in situ data were unavailable. Overall, geo-embeddings-based approaches provide a scalable, reproducible, and operationally viable pathway for classifying EDs with this translating into greater classifications of land cover. Once trained, the model can be implemented for other years, with this supporting numerous national to international applications including in relation to land use dynamics and impacts on carbon and biodiversity.