Vegetation index from satellite imagery — Task

Implementation: Vegetation index from Sentinel-2 (Recipe - QGIS).

Vegetation index from satellite imagery is the Task of turning a multispectral satellite image of a study area into a map of how green, how dense and how active its vegetation is: NDVI, or another index. It is the first quantitative step from an image to a map of the land surface, and the basis for comparing places, seasons and years. Theory: Remote sensing, especially Spectral signatures and vegetation indices.

Used in examples: Vegetation greenness at Ledreborg (Example).

Task Schema (tool-agnostic)

1. Define the study area — a polygon, with a margin.
2. Choose the image (decision point; Operator: selection of images by location, date and cloud cover) — a date that fits the question (the season, the stage of the crops), clear over the study area itself, surface reflectance (Level-2A for Sentinel-2).
3. Get the bands for the area (Operator: clip to extent) — red and near infrared for NDVI; the scene classification for a cloud mask; blue and green for a true-colour image to look at.
4. Know the values (decision point) — reflectance (0–1) or scaled integers (× 10,000); whether an offset must be removed; the NoData value.
5. Look at the image — a true-colour and a colour-infrared composite: clouds, haze, shadows, and what the area holds.
6. Calculate the index (Operator: map algebra) — NDVI = (NIR − Red) / (NIR + Red).
7. Mask what is not surface (Operator: map algebra with a condition) — cloud, cloud shadow, defective and unclassified pixels.
8. Check the result — the values lie between −1 and 1, and known surfaces have the values they should: water below 0, roads and roofs near 0, forest and crops in June 0.8–0.9.
9. Summarise if needed (Operator: zonal statistics) — the index per zone or per land-use class (Aggregating a continuous raster to zones (Task)).

Output: an index raster of the study area, masked, with a record of the image (satellite, date, product level) and the formula.

Epistemic Choices to record

• Date: NDVI follows the growing season. A June image and an August image of the same field can differ from bare soil to dense crop. Comparisons need images from the same time of year.
• Product level: surface reflectance makes dates and places comparable; top-of-atmosphere reflectance still contains haze.
• Saturation: above about 0.8 NDVI barely separates dense vegetation types; forest and a closed crop look alike. Another date, another index or a colour-infrared image may separate them.
• Mixed pixels: at 10 m, narrow and small features (roads, streams, ponds, hedges) take part of their value from their surroundings.
• Mask: which classes count as not surface; the scene classification is automatic and misses thin cloud and some shadow.
• What NDVI is: an index of greenness, not a measure of biomass, yield or biodiversity. Linking it to those needs field data.