Input dataset for gap filling and land-cover mapping using eumap Library - 2000 to 2020

Benchmark dataset containing slope, elevation, Landsat temporal composites and night light raster layers, and the training samples (LUCAS and CORINE samples compilation) to map the land-cover in different areas of the European Union-EU.

The slope and elevation refers to Digital Terrain Model for Continental Europe, and the night light images are from VNP46A1 product (VIIRS/NPP Daily Gridded Day Night Band 500m). The temporal composites were based on GLAD Landsat ARD, considering the 4 seasons and 3 percentiles per season (25, 50 and 75), for 6 spectral (blue, green, red, NIR, SWIR1, SWIR2) and 1 thermal band, resulting at end in 88 Landsat composites per year. The images for each season were selected using the same calendar dates for all period:

• Winter: December 2 of previous year until March 20 of current year
• Spring: March 21 until June 24 of current year
• Summer: June 25 until September 12 of current year
• Fall: September 13 until December 1 of current year

The temporal composites were generated to Sentinel-2 L2A for 2018, 2019 and 2020, using the same approach (4 seasons x 3 percentiles x 6 spectral bands).

The benchmark areas were selected according to the EU tiling system, which consists of 7,042 regular tiles with 30 x 30 km. The dataset uses the ETRS89-extended / LAEA Europe as spatial reference system (EPSG:3035), and all the raster layers have 1,000 x 1,000 pixels and 30m of spatial resolution.

For all the EU the training samples will have 32 land-cover classes, varying according to the benchmark area:

• 111: Urban fabric
• 122: Road and rail networks and associated land
• 123: Port areas
• 124: Airports
• 131: Mineral extraction sites
• 132: Dump sites
• 133: Construction sites
• 141: Green urban areas
• 211: Non-irrigated arable land
• 212: Permanently irrigated arable land
• 213: Rice fields
• 221: Vineyards
• 222: Fruit trees and berry plantations
• 223: Olive groves
• 231: Pastures
• 311: Broad-leaved forest
• 312: Coniferous forest
• 321: Natural grasslands
• 322: Moors and heathland
• 323: Sclerophyllous vegetation
• 324: Transitional woodland-shrub
• 331: Beaches, dunes, sands
• 332: Bare rocks
• 333: Sparsely vegetated areas
• 334: Burnt areas
• 335: Glaciers and perpetual snow
• 411: Inland wetlands
• 421: Maritime wetlands
• 511: Water courses
• 512: Water bodies
• 521: Coastal lagoons
• 522: Estuaries
• 523: Sea and ocean

The gap filling validation data was generated by creating a mask of all nodata pixels (gaps) for each temporal composite, and then transposing that mask. All valid pixels covered by the transposed nodata mask are considered validation pixels. This method was chosen to retain the diversity of spatiotemporal nodata patterns that occur in the data. Each gap filling validation file contains 3 directory:

• raw: original temporal composite
• validation: transposed data
• filled_tmwm8: the best gap filling method that was tested

See the eumap library for more information about the gapfiling approach and land-cover mapping using this dataset.