EnviroAtlas land cover dataset

This dataset contains imagery, land cover, and object layers for four of the cities covered by the EPA’s EnviroAtlas dataset: Pittsburgh, PA, Phoenix, AZ, Durham, NC, and Austin, TX. The dataset consists of 10 test tiles and 5 validation tiles in each city, with an additional 10 training tiles and 8 (total) validation tiles in Pittsburgh, subsetting the total area that the EnviroAtlas dataset covers in each city. The EnviroAtlas label definitions are explained in Pilant et al. 2020. In the processed prior files included in this dataset, channels correspond to:

• 0: water
• 1: impervious surface
• 2: soil and barren
• 3: trees and forest (and shrub in AZ)
• 4: grass and herbaceous

Specifically, this dataset is in the same format as the "Chesapeake Land Cover" dataset and contains the following layers:

• NAIP 1m aerial imagery (year = 2010 for Pittsburgh, P and Phoenix, AZ, and year = 2012 for Durham NC, and Austin TX
  • https://www.fsa.usda.gov/programs-and-services/aerial-photography/imagery-programs/naip-imagery/index
  • https://planetarycomputer.microsoft.com/dataset/naip
• NLCD 2016
  • https://www.mrlc.gov/data
• OpenStreetMap roads
  • https://www.openstreetmap.org/
• OpenStreetMap waterways, waterbodies, water= max(waterways, waterbodies)
  • https://www.openstreetmap.org/
• Microsoft building footprints
  • https://github.com/Microsoft/USBuildingFootprints
• EnviroAtlas 1m land cover labels:
  • https://www.epa.gov/enviroatlas/about-data
• Prior probabilities of the 1m land cover labels (with OSM and building data fused into prior)
• Prior probabilities of the 1m land cover labels (without OSM or building data)

For more information on how the OpenStreetMap layers are downloaded and processed, as well as how the prior probabilities are calculated, see the accompanying GitHub repository: 

Version 1.1 fixes an issue with the spatial index file.