Land Use / Land Cover (LULC) Mapping, Explained
A Land Use / Land Cover (LULC) map answers a question that underpins nearly every environmental study: what is actually on the ground, and where? "Land cover" describes the physical surface — water, forest, cropland, bare soil, built-up land — while "land use" describes how people employ it. Turning a raw satellite image into a clean, labelled map of these classes is one of the most widely used workflows in remote sensing, and it is the foundation for measuring deforestation, urban sprawl, wetland loss, and agricultural change.
How classification works
The key insight is that different surfaces reflect sunlight differently across the electromagnetic spectrum. Healthy vegetation reflects strongly in the near-infrared; water absorbs it almost completely; concrete and bare soil have their own distinct signatures across the visible and shortwave-infrared bands. Each pixel in a satellite image therefore carries a small spectral "fingerprint." Supervised classification teaches a machine-learning model to recognise these fingerprints:
- Collect training samples — the analyst marks known examples of each class (this polygon is water, that one is forest).
- Train a classifier — an algorithm such as Random Forest or a Support Vector Machine learns the spectral pattern of each class from those samples.
- Classify every pixel — the trained model assigns each pixel in the scene to its most likely class, producing a wall-to-wall thematic map.
Why accuracy assessment is non-negotiable
A colourful map is worthless if it is wrong, so a credible LULC study never stops at the map itself. Analysts set aside independent validation samples the model never saw during training and compare its predictions against them in a confusion matrix. From that matrix come two essential figures: overall accuracy (the percentage of validation points classified correctly) and the Kappa coefficient, which discounts the agreement you would expect by pure chance. A map without these numbers should be treated with caution; a map with an overall accuracy of, say, 90% and a Kappa of 0.85 can be reported and defended.
What LULC maps make possible
Once you can classify one date, you can classify several and subtract them — revealing exactly how much forest became farmland, or farmland became city, over a decade. This change detection is how researchers quantify environmental transformation with dated, measurable evidence rather than impression. From carbon accounting to urban planning to conservation, the humble land-cover map remains one of remote sensing's most powerful outputs.