Session 5: Water stress mapping with Sentinel-2

Preamble¶

• Tools: Python, QGIS (optional)

• Data: Multi-date Sentinel-2 images over Casablanca / Bouskoura forest

• Goal: identify areas of potential vegetation water stress using spectral indices and multi-date analysis

Project objective¶

The objective of this project is to detect and map vegetation water stress using Sentinel-2 multi-date imagery.

You will spectral information to compare vegetation conditions between different dates and identify areas showing possible stress.

At this stage, the focus is on:

• preparing and organizing the data

• computing relevant spectral indices

• comparing several dates

• interpreting spatial patterns

• optionally performing a simple unsupervised classification

• preparing reference data for supervised classification in the next session

Study area¶

The study area is located in Casablanca, including the Bouskoura forest.

This area is well suited for the project because it includes:

• vegetated areas

• urban areas

• bare soil

• seasonal variability in vegetation condition

Recommended dates¶

To study vegetation water stress, it is recommended to compare images acquired during different periods of the year:

• Wet season: February or March

• Transition period: May or June

• Dry season: August or September

📂 Labworks data
📥 session_5_data.zip

Suggested spectral indices¶

The following indices are recommended for this project:

1. NDVI — vegetation activity¶

NDVI helps identify vegetation vigor.

NDVI = (B8 - B4) / (B8 + B4)

• High NDVI → dense / healthy vegetation

• Low NDVI → sparse vegetation, bare soil, urban surfaces

2. NDMI (or vegetation moisture index)¶

NDMI is useful for vegetation water content and moisture stress.

NDMI = (B8 - B11) / (B8 + B11)

• High NDMI → wetter vegetation

• Low NDMI → drier vegetation / possible stress

3. Optional indices¶

If useful, you may also test:

simple use of SWIR band (B11)

difference between dates:

NDVI_date2 - NDVI_date1

NDMI_date2 - NDMI_date1

These differences can help highlight areas where vegetation condition decreases over time.

Expected workflow¶

Task 1 – Select and prepare the data¶

1. Download Sentinel-2 images for the selected dates

2. Keep the relevant bands:

• B4 — Red

• B8 — NIR

• B11 — SWIR

3. Clip the images to the study area

4. Check that the rasters are aligned and comparable

Task 2 – Compute the indices¶

Use Python or the Raster Calculator in QGIS to compute NDVI and NDMI for each date.

Expected workflow¶

Task 1 – Select and prepare the data¶

1. Download Sentinel-2 images for the selected dates

2. Keep the relevant bands:

• B4 — Red

• B8 — NIR

• B11 — SWIR

3. Clip the images to the study area

4. Check that the rasters are aligned and comparable

Task 3 – Compare dates¶

Display the computed indices for the different dates.

Use:

• color maps

• histograms

• pixel value inspection

• side-by-side comparison

Questions¶

• Which date shows the highest vegetation activity?

• Which date shows the strongest moisture deficit?

• Are all vegetated areas affected in the same way?

• Do some areas remain green or moist during the dry season?

Task 4 – Map potential stress classes¶

Create a simple classification of vegetation stress based on NDVI and/or NDMI or other selected features.

Example:

This classification can be done:

• visually

• by thresholding

• using raster expressions

Example idea¶

Use NDMI values to separate wetter and drier vegetation.
Then compare with NDVI to distinguish sparse vegetation from stressed vegetation.

Task 5 – Optional unsupervised classification¶

If time allows, you may test a simple unsupervised approach to group pixels with similar spectral behaviour.

The objective is to:

• explore patterns in the data

• identify potential vegetation stress groups

• compare results with index-based interpretation

Possible variables:

• NDVI

• NDMI

• Red band

• NIR band

• SWIR band

Example method: K-means clustering

Task 6 – Prepare training data for the next session¶

In QGIS, create a vector layer and draw polygons representing different vegetation conditions.

Suggested classes:

• healthy vegetation

• moderately stressed vegetation

• highly stressed vegetation

• non-vegetated surfaces

• water

Important recommendations¶

• Draw polygons on homogeneous areas

• Avoid boundaries between land cover types

• Create several polygons per class

• Distribute polygons across the study area

These polygons will be used as training data for supervised classification in the next session.

Expected outputs¶

Each group should produce:

1. A short description of the selected dates

2. At least one NDVI map

3. At least one NDMI map

4. A preliminary vegetation stress map

5. A vector layer containing training polygons

6. A short interpretation of the spatial and temporal patterns observed

Deliverables¶

At the end of the project preparation phase, each group should submit:

• the selected Sentinel-2 rasters

• the computed index rasters

• the QGIS project

• the training polygons

• one or two exported maps (PDF)

• a short methodological note