Sentinel-2

class: center, middle, inverse, title-slide

.title[
# Sentinel-2
]
.subtitle[
## Multispectral Imaging for Earth Observation
]
.author[
### Shiyao Zhang
]
.institute[
### CASA, UCL
]
.date[
### 2026.2
]

---

# What is Sentinel-2?

.pull-left[
**Sentinel-2** is a twin-satellite constellation (2A & 2B) operated by the **European Space Agency (ESA)** under the Copernicus Earth Observation Programme.

Launched in **2015** (2A) and **2017** (2B), it provides **free, open-access** multispectral imagery with a global revisit time of **5 days** — making it the most widely used open Earth observation dataset for land monitoring (Drusch et al., 2012).

.footnote[ESA, 2023; Drusch et al., 2012]
]
.pull-right[
<img src="https://www.esa.int/var/esa/storage/images/esa_multimedia/images/2024/09/seville_from_copernicus_sentinel-2c/26314246-1-eng-GB/Seville_from_Copernicus_Sentinel-2C_article.jpg" style="width:100%; max-height:320px; object-fit:cover; border-radius:4px;">
Fig 1. True-colour Sentinel-2C image of Seville, Spain (Sep 2024). The Guadalquivir River and surrounding land use are visible at 10 m resolution. © ESA/Copernicus
]
---
# Technical Specifications

Sentinel-2 carries the **MultiSpectral Instrument (MSI)**: a push-broom imager covering 13 spectral bands across visible, NIR, and SWIR wavelengths.

| Band | Name | Wavelength (nm) | Resolution |
|------|------|-----------------|------------|
| B2 | Blue | 490 | **10 m** |
| B3 | Green | 560 | **10 m** |
| B4 | Red | 665 | **10 m** |
| B8 | NIR | 842 | **10 m** |
| B5-B7, B8A | Red-edge | 705-865 | 20 m |
| B11, B12 | SWIR | 1610-2190 | 20 m |
| B1, B9, B10 | Aerosol / Water vapour | 443-1380 | 60 m |

.footnote[ESA Sentinel-2 User Handbook, 2021]
---
# How Sentinel-2 Works

.pull-left[
## Data Acquisition
- **Push-broom scanning**: images the full 290 km swath simultaneously
- Acquisitions at **10:30 local solar time** to minimise shadow
- **L1C** - Top-of-atmosphere reflectance
- **L2A** - Surface reflectance (atmospherically corrected)

## Data Access
- Copernicus Data Space Ecosystem
- Google Earth Engine: `COPERNICUS/S2_SR_HARMONIZED`
- AWS Open Data Registry
]
.pull-right[
<img src="https://www.esa.int/var/esa/storage/images/esa_multimedia/images/2024/09/france_spain_strip_from_sentinel-2c/26314340-1-eng-GB/France_Spain_strip_from_Sentinel-2C_article.jpg" style="width:100%; max-height:320px; object-fit:cover; border-radius:4px;">
Fig 2. A single Sentinel-2C pass covering a 290 km strip from Camargue (France) to south of Barcelona (Spain). This wide swath is what makes the 5-day global revisit possible — the entire acquisition took just seconds. © ESA/Copernicus
]
---
# Application 1 - Urban Informal Settlement Mapping

.pull-left[
**Wurm et al. (2019)** used Sentinel-2 10 m imagery with **SVM classifiers** to map informal settlement typologies in three African cities with **>85% overall accuracy**.

The **red-edge bands** proved critical for distinguishing corrugated metal roofing from formal structures - impossible at Landsat's 30 m resolution.

- UN-Habitat: 1.1 billion people in informal settlements
- Enables near-real-time monitoring at **zero data cost**

.footnote[Wurm et al., 2019, *ISPRS J. Photogramm.*]
]
.pull-right[
## Workflow
```
Sentinel-2 L2A imagery
        |
Band ratios + GLCM texture
(NDVI, NDBI, entropy)
        |
SVM classifier (ground truth)
        |
Settlement typology map
        |
Multi-year change detection
```
]
---
# Application 2 - National Crop Type Mapping

.pull-left[
**Inglada et al. (2017)** used dense Sentinel-2 **time-series** with a **Random Forest** classifier to map 9 crop types across France at **>90% accuracy**.

The **5-day revisit** captures phenological trajectories (sowing to canopy closure to senescence) distinguishing winter wheat from rapeseed.

## Broader Impact
- EU **Common Agricultural Policy** subsidy verification
- FAO & WFP early food security warnings
- Basis of ESA's **World Cereal** global crop product

.footnote[Inglada et al., 2017, *Remote Sensing*]
]
.pull-right[
Why Temporal Density Matters

| Feature | Landsat | Sentinel-2 |
|---------|---------|------------|
| Revisit | 16 days | **5 days** |
| Free? | Yes | Yes |
| Red-edge | No | **Yes** |
| Swath | 185 km | **290 km** |

Inglada et al. (2017) found that temporal density — not spatial resolution — was the primary determinant of crop classification accuracy.
]
---
# Application 3 - Flood Mapping & Disaster Response

.pull-left[
**Twele et al. (2020)** produced flood extent maps from Sentinel-2 **SWIR + NDWI** change detection within **48 hours** of Cyclone Idai's landfall in Mozambique (March 2019).

Combined with Sentinel-1 SAR, this enabled **building-level damage assessment** guiding IFRC and UNOCHA field operations.

## Copernicus EMS
Activated in **300+ disaster events** since 2012, using Sentinel-2 as its primary optical source.

.footnote[Twele et al., 2020; Copernicus EMS, 2024]
]
.pull-right[
<img src="https://www.esa.int/var/esa/storage/images/esa_multimedia/images/2024/09/wildfire_in_california_captured_by_sentinel-2c/26314789-5-eng-GB/Wildfire_in_California_captured_by_Sentinel-2C_article.jpg" style="width:100%; max-height:320px; object-fit:cover; border-radius:4px;">
Fig 3. SWIR false-colour composite of the 2024 Airport Fire, California. Active fire shows as bright orange (Band 12 SWIR + Band 8A NIR); the burn scar appears dark red. This is the same spectral technique applied to flood and disaster mapping — SWIR bands reveal what true colour cannot. © ESA/Copernicus
]
---
# Reflection

.pull-left[
## What I Have Learnt

Before this I mostly thought of satellite data as something expensive and hard to access. Sentinel-2 changed that assumption pretty quickly - the fact that you can pull 10 m imagery globally for free in GEE is still a bit surprising to me.

The **red-edge bands** were something I had genuinely not heard of before. I kept seeing them mentioned in the papers and had to look them up - turns out they matter a lot more than the standard RGB+NIR setup I was used to from Landsat examples in lectures.

I also hadn't realised how much the **5-day revisit** changes what's actually possible. A lot of the crop and flood work just wouldn't work with 16-day gaps.
]
.pull-right[
## How I Might Use This

For my own work I'm most interested in urban applications. The LST + deprivation index idea came up when I was reading about the flood mapping paper - if you can detect surface water change that fast, tracking heat stress over a summer should be feasible too.

I want to try pulling a Sentinel-2 time series for London in GEE at some point, even just to get comfortable with the data format before doing anything more serious with it.

One thing I'm still unsure about: how much does atmospheric correction actually matter in practice for urban indices like NDBI? The papers seem to assume L2A but don't always explain the difference clearly.
]
---
# References

Copernicus EMS (2024). *CEMS Rapid Mapping activations*. https://emergency.copernicus.eu

Drusch, M. et al. (2012). Sentinel-2: ESA's optical high-resolution mission for GMES. *Remote Sensing of Environment*, 120, 25-36.

ESA (2021). *Sentinel-2 User Handbook*. European Space Agency.

Inglada, J. et al. (2017). Operational high resolution land cover map production using satellite image time series. *Remote Sensing*, 9(1), 95.

Twele, A. et al. (2020). Sentinel-1 and Sentinel-2 based near real-time flood extent mapping. *Int. J. Remote Sensing*, 42, 1-24.

Wurm, M. et al. (2019). Semantic segmentation of slums using transfer learning. *ISPRS J. Photogramm.*, 150, 59-69.