Specifications

The EarthNet2021 samples, so-called data multicubes, are saved as compressed numpyarrays, so analysis-ready for usage in python. In the following we describe what exactly is contained in each multicube.

Accessing multicube

The next example shows how to access the data in one multicube example.npz:

import numpy as np

# Loading File
sample = np.load("example.npz")

# Accessing high-resolution dynamic variables (the Sentinel 2 bands)
hrd = sample["highresdynamic"]  
imgs = hrd[:,:,:4,:] # B, G, R, NIR channels
msks = hrd[:,:,-1,:] # EarthNet2021 binary quality mask

# Accessing mesoscale dynamic variables (the E-OBS weather data)
md = sample["mesodynamic"]

# Accessing high-resolution and mesoscale static variables (the EUDEM digital elevation model)
hrs = sample["highresstatic"]
ms = sample["mesostatic"]

Highresdynamic

The variables "highresdynamic" have:

  • axes (height, width, channels, time)
  • dimension (128, 128, c, t)
    • t is the 5-daily time
    • Train c = 7, t = 30
    • Test c = 5
      • context t = 10 / 20 / 70
      • target t = 20 / 40 / 140
  • Channels:
    • Train (Blue, Green, Red, Near-Infrared, Sen2Cor Cloud Mask, ESA Scene Classification, EarthNet2021 Data Quality Mask)
    • Test (Blue, Green, Red, Near-Infrared, EarthNet2021 Data Quality Mask)
  • Units:
    • B, G, R, NIR (B02, B03, B04, B8A): 0 - 2 reflectance, NaN if not available.
    • Sen2Cor Cloud Mask (CLD): 0-100 cloud probability
    • ESA Scene Classification (SCL): 0-11 categories (see here)
    • EarthNet2021 data quality mask: {0,1} binary mask, 0 if good quality, 1 if bad quality.

ProTip: Preprocess images by imgs[imgs < 0] = 0, imgs[imgs > 1] = 1, and imgs[np.isnan(imgs)] = 0.

Mesodynamic

The variables "mesodynamic" have:

  • axes (height, width, channels, time)
  • dimension (80, 80, 5, t)
    • t is the daily time, md[:,:,:,4] fits the sentinel 2 date hrd[:,:,:,0].
    • Train t = 30
    • Test context t = 150 / 300 / 1050, target t = 0
  • Channels:
    • Precipitation (RR), Sea pressure (PP), Mean temperature (TG), Minimum temperature (TN), Maximum temperature (TX)
    • for more see here
  • Units:
    • All data has been rescaled to lay between 0 and 1, transformation rules:
      • Temperatur (°C) = 5000(2temp - 1)
      • Rain (mm) = 50 * rain
      • Pressure (hPa) = 200 * pressure + 900

ProTip: Missing data in the E-OBS variables is visible by those pixels where PP = 0. Note this information does not go into the high-resolution data quality mask.

Static

The static variable has:

  • axes (height, width, channels)
  • dimension (h, w, 1)
    • "highresstatic"has h = w = 128
    • "mesostatic"has h = w = 80
  • Channels:
  • Units:
    • Data has been rescaled to lay between 0 and 1, transformation rule:
      • DEM (m) = 2000 * (2*dem - 1)