Skip to content

Dataset · M3LEO-miniset

Satellite imagery and geodata of the United States

M3LEO-miniset/conus

A random sample of 5,000 multimodal satellite tiles of the contiguous United States for remote sensing tasks.

What's inside

The dataset includes between 10,000 and 100,000 chips generated from 5,000 randomly selected geographic tiles of the contiguous US (3,000 for training, 1,000 each for validation and test). It contains various types of data for 2020, including radar imagery (s1grd), optical RGB mosaics (s2rgbm), interferograms (gunw), and biomass data.

What you can do with it

  • Training satellite imagery analysis and Earth surface monitoring models
  • Training multimodal geo-information neural networks (combining radar and optics)
Size
10K–100K rows
Downloads this month
27.8K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("M3LEO-miniset/conus")