ESA-PhiLab-Edge/PhilEOBench-land_cover_classification

Simulated PhiSat Bench Dataset - Landcover

This repository contains a simulated PhiSat2 dataset derived from Sentinel-2 data for landcover classification, with a spatial resolution of 10 meters per pixel.

Landcover Labels

The landcover labels in this dataset follow a classification scheme defined by the WorldCover PUM V1.1 guidelines. Each pixel is labeled with the dominant landcover class. The key classes include:

Code	Class Name	Description
10	Tree Cover	Areas with ≥10% tree cover (includes plantations, flooded forests, excluding mangroves).
20	Shrubland	Dominated by natural shrubs, typically <5 m tall.
30	Grassland	Herbaceous vegetation like grasslands, steppes, prairies, and pastures.
40	Cropland	Annual crops harvested within 12 months; greenhouses are classified as built-up.
50	Built-up	Built environments, including buildings, roads, and other artificial infrastructure.
60	Bare/Sparse Vegetation	Areas with exposed soil, sand, or rocks; usually with <10% vegetative cover year-round.
70	Snow and Ice	Regions that are persistently covered by snow or ice.
80	Permanent Water Bodies	Water bodies present for at least 9 months per year (can be fresh or salt water).
90	Herbaceous Wetland	Areas dominated by herbaceous plants that are permanently or seasonally flooded.
95	Mangroves	Intertidal zones with salt-tolerant tree vegetation.
100	Moss and Lichen	Regions covered by lichens or mosses (often found in tundra or alpine zones).

Use Case

This dataset is intended for landcover classification tasks, including applications in urban planning, environmental monitoring, and resource management.