An integrated framework combining multi-resolution satellite data, deep learning, and process-based modeling for urban carbon accounting
Frontiers in Environmental Science · 16.03.2026 · 14
DOI: 10.3389/fenvs.2026.1787442 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Environmental Science, online erschienen am 16.03.2026. Verfasst von Faten Dhawi und 2 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Urban green spaces are increasingly recognized as actionable carbon sinks, yet their carbon stocks and greening potential remain poorly constrained in arid and rapidly urbanizing regions. We quantified campus-scale carbon storage and identified vegetation-driven opportunities for enhancing the urban carbon sink using an integrated remote sensing and ecosystem-service modelling workflow. The study was conducted at King Faisal University, Al-Ahsa, Saudi Arabia (≈100 ha; 25.334°N, 49.602°E). Sentinel-2 (10 m) and PlanetScope (3 m) imagery were used to derive NDVI-based land-cover areas, and YOLOv5 object detection provided a vegetation inventory. Carbon stocks were estimated with the InVEST Carbon Storage model parameterized by land-cover class. PlanetScope detected substantially more fragmented green space than Sentinel-2 (0.0890 vs. 0.0465 km 2 ; +91.4%), indicating resolution-driven underestimation of small vegetation patches in moderate-resolution imagery. Estimated carbon density was highest in green land (62 t C ha −1 ) and lower in built-up (28 t C ha −1 ), bare land (21 t C ha −1 ), and other/paved surfaces (15 t C ha −1 ), yielding a total standing stock of 2,655.98 tonnes C. YOLOv5 identified 2,762 vegetation objects (1,273 trees; 1,489 plants/shrubs), supporting repeatable monitoring. A benchmark greening scenario (CD_potential = 62 t C ha −1 ) indicated an additional storage potential of 3,544.02 tonnes C if low-carbon land covers were enhanced toward high-biomass vegetated conditions. These estimates represent standing carbon stocks and scenario-based additional storage potential, rather than net carbon sequestration rates (which would require temporal monitoring and accounting for management-related emissions). This framework demonstrates how multi-resolution earth observation and AI-enabled vegetation inventorying can support carbon-focused campus planning by prioritizing protection of existing vegetation and targeted greening interventions.
Zitieren
APA
Dhawi, F., Ghafoor, A., & Alkhaldi, N. A. (2026). An integrated framework combining multi-resolution satellite data, deep learning, and process-based modeling for urban carbon accounting. Frontiers in Environmental Science, 14. https://doi.org/10.3389/fenvs.2026.1787442
Vancouver
Dhawi F, Ghafoor A, Alkhaldi NA. An integrated framework combining multi-resolution satellite data, deep learning, and process-based modeling for urban carbon accounting. Frontiers in Environmental Science. 2026;14. doi:10.3389/fenvs.2026.1787442
BibTeX
@article{Dhawi2026,
author = {Dhawi, Faten and Ghafoor, Abdul and Alkhaldi, Nora A.},
title = {An integrated framework combining multi-resolution satellite data, deep learning, and process-based modeling for urban carbon accounting},
journal = {Frontiers in Environmental Science},
year = {2026},
volume = {14},
doi = {10.3389/fenvs.2026.1787442}
}
EUVIMED