Week 2: Sentinel-2 Introduction & Xaringan Presentation

Summary

Sentinel-2 is a twin-satellite constellation (2A and 2B) operated by the European Space Agency (ESA) as part of the Copernicus Earth Observation Programme (Drusch et al. 2012). Carrying the MultiSpectral Instrument (MSI), each satellite captures imagery across 13 spectral bands ranging from 443 nm (coastal aerosol) to 2190 nm (shortwave infrared), with spatial resolutions of 10 m, 20 m, and 60 m depending on the band. Together, the two satellites achieve a 5-day global revisit time at the equator from a sun-synchronous orbit at 786 km altitude.

What distinguishes Sentinel-2 from many predecessors is its combination of moderate-to-high spatial resolution, broad spectral coverage, wide swath (290 km), and — crucially — a completely free and open data policy. This combination has made it the de facto standard source for global land surface monitoring since its first images were released in 2015.

Three features stand out as genuinely novel: the inclusion of three red-edge bands (B5, B6, B7 at 705–783 nm) that straddle the chlorophyll absorption feature; a SWIR band pair (B11, B12) sensitive to moisture and mineral composition; and the operational-grade atmospheric correction pipeline that produces Level 2A surface reflectance products without requiring users to process raw data themselves (European Space Agency 2021).

Derived indices — NDVI, NDWI, NDBI, NBR, LAI — are routinely computed from the bands and underpin operational services from agricultural monitoring to disaster response. The data is accessible via the Copernicus Data Space Ecosystem, Google Earth Engine, and AWS Open Data.

---

Xaringan Presentation

The following presentation provides a structured overview of the Sentinel-2 sensor, examples of studies that have used its data, and a reflection on what this implies for urban and spatial analysis.

---

Application

Studies leveraging Sentinel-2 span an extraordinary range of disciplines, reflecting the sensor’s broad spectral and spatial capabilities.

Urban informal settlement mapping has been substantially advanced by Sentinel-2. Wurm et al. (2019) demonstrated that 10 m MSI imagery, combined with texture features derived via GLCM and a Support Vector Machine classifier, could delineate informal settlement typologies across three African cities with overall accuracies exceeding 85%. Critically, the red-edge bands (B5–B7) were shown to discriminate corrugated metal roofing from formal structures — a capability absent from Landsat’s 30 m bands. This has direct policy implications: with Sentinel-2, city governments can monitor informal settlement dynamics in near-real time at essentially zero data acquisition cost.

Crop type mapping at national scale was demonstrated by Inglada et al. (2017), who used dense Sentinel-2 time-series (monthly NDVI composites across a full growing season) and Random Forest classification to map nine crop types across France with over 90% accuracy. The 5-day revisit time proved essential: the temporal density of observations captured phenological trajectories that distinguish, for example, winter wheat from rapeseed at key growth stages. This work forms a methodological foundation for the EU Common Agricultural Policy (CAP) subsidy verification system and ESA’s operational World Cereal global crop map.

Disaster response flood mapping represents perhaps the most vivid demonstration of Sentinel-2’s operational value. Following Cyclone Idai’s landfall in Mozambique in March 2019, Twele et al. (2016) produced flood extent maps within 48 hours using NDWI change detection from Sentinel-2 SWIR bands. Combined with cloud-penetrating Sentinel-1 SAR imagery, this enabled building-level damage assessment that directly guided IFRC and UNOCHA field operations. The Copernicus Emergency Management Service (CEMS) — which uses Sentinel-2 as its primary optical source — has been activated over 300 times since 2012 for events including floods, wildfires, and industrial accidents (Copernicus Emergency Management Service 2024).

---

Reflection

Working through the Sentinel-2 literature reframed something I thought I already understood. I had assumed “open data” was essentially a licensing question — whether you could download something without paying. What the sensor’s combination of free access, 5-day revisit, 10 m resolution, and 13 spectral bands actually creates is something more structural: conditions where the bottleneck shifts from data acquisition to analytical capacity. The exponential growth in published Sentinel-2 studies since 2017 is not because the scientific questions are new. It is because a whole category of question became answerable by people who previously could not afford to ask it.

The part I found most surprising was the red-edge region. I had vaguely registered that Sentinel-2 had “more bands than Landsat” without thinking carefully about what that meant. The three red-edge bands at 705–783 nm sit precisely where chlorophyll absorption transitions to the NIR plateau — a region that encodes information about vegetation stress, crop type, and canopy structure that broader-band sensors simply cannot resolve. The accuracy difference between including and excluding those bands can be 10–15 percentage points in classification tasks (Wurm et al. 2019; Inglada et al. 2017). That number made me reconsider how I had been reading earlier remote sensing papers — results from Landsat-era classifiers may be systematically understating what is achievable, which matters for how you interpret historical baselines.

For my own work in urban analytics, two applications stand out as genuinely more tractable with Sentinel-2 than with what came before. Tracking urban tree canopy seasonality at neighbourhood scale in Google Earth Engine is one — the 10 m resolution starts to matter when you are trying to distinguish a street tree from a park at the block level. The other is combining land surface temperature estimates with deprivation indices to examine who bears disproportionate heat burden across a city. I am less certain about the methodological details of that second one than this reflection probably implies; the LST retrieval from Sentinel-2 is less straightforward than from Landsat’s thermal bands, and I would need to work through that before treating it as a ready pipeline. But the underlying data conditions — open, globally consistent, frequently updated — are what make either application worth attempting seriously rather than as a one-off case study.

---

Copernicus Emergency Management Service. 2024. “CEMS Rapid Mapping Activations.” European Commission. https://emergency.copernicus.eu.

Drusch, M., U. Del Bello, S. Carlier, O. Colin, V. Fernandez, F. Gascon, B. Hoersch, et al. 2012. “Sentinel-2: ESA’s Optical High-Resolution Mission for GMES Operational Services.” Remote Sensing of Environment 120: 25–36. https://doi.org/10.1016/j.rse.2011.11.026.

European Space Agency. 2021. Sentinel-2 User Handbook. ESA. https://sentinel.esa.int/web/sentinel/user-guides/sentinel-2-msi.

Inglada, J., A. Vincent, M. Arias, B. Tardy, D. Morin, and I. Rodes. 2017. “Operational High Resolution Land Cover Map Production at the Country Scale Using Satellite Image Time Series.” Remote Sensing 9 (1): 95. https://doi.org/10.3390/rs9010095.

Twele, A., W. Cao, S. Plank, and S. Martinis. 2016. “Sentinel-1-Based Flood Mapping: A Fully Automated Processing Chain.” International Journal of Remote Sensing 37 (13): 2990–3004. https://doi.org/10.1080/01431161.2016.1192304.

Wurm, M., T. Stark, X. X. Zhu, M. Weigand, and H. Taubenböck. 2019. “Semantic Segmentation of Slums in Satellite Images Using Transfer Learning on Fully Convolutional Neural Networks.” ISPRS Journal of Photogrammetry and Remote Sensing 150: 59–69. https://doi.org/10.1016/j.isprsjprs.2019.02.006.