Rio-Tiler-PDS: A rio-tiler plugin for Public Datasets

A rio-tiler plugin to read from publicly-available datasets.

Important This is the new module for rio-tiler missions specific (ref: https://github.com/cogeotiff/rio-tiler/issues/195)

Documentation: https://cogeotiff.github.io/rio-tiler-pds/

Source Code: https://github.com/cogeotiff/rio-tiler-pds

Installation

You can install rio-tiler-pds using pip

$ pip install -U pip
$ pip install rio-tiler-pds

or install from source:

pip install git+https://github.com/tangentlabs/django-oscar-paypal.git@issue/34/oscar-0.6

$ pip install -U pip
$ pip install git+https://github.com/cogeotiff/rio-tiler-pds.git

Datasets

Adding more dataset:

If you know of another publicly-available dataset that can easily be described with a "scene id", please feel free to open an issue.

Warnings

Requester-pays Buckets

On AWS, sentinel2, sentinel1, and cbers datasets are stored in requester pays buckets. This means that the cost of GET and LIST requests and egress fees for downloading files outside the AWS region will be charged to the accessing users, not the organization hosting the bucket. For rio-tiler and rio-tiler-pds to work with such buckets, you'll need to set AWS_REQUEST_PAYER="requester" in your shell environment.

Partial reading on Cloud hosted dataset

When reading data, rio-tiler-pds performs partial reads when possible. Hence performance will be best on data stored as Cloud Optimized GeoTIFF (COG). It's important to note that Sentinel-2 scenes hosted on AWS are not in Cloud Optimized format but in JPEG2000. Partial reads from JPEG2000 files are inefficient, and GDAL (the library underlying rio-tiler-pds and rasterio) will need to make many GET requests and transfer a lot of data. This will be both slow and expensive, since AWS's JPEG2000 collection of Sentinel 2 data is stored in a requester pays bucket.

Ref: Do you really want people using your data blog post.

Overview

Readers

Each dataset has its own submodule (e.g sentinel2: rio_tiler_pds.sentinel.aws)

from rio_tiler_pds.landsat.aws import L8Reader
from rio_tiler_pds.sentinel.aws import S1L1CReader
from rio_tiler_pds.sentinel.aws import (
    S2L1CReader,  # JPEG2000
    S2L2AReader,  # JPEG2000
    S2L2ACOGReader,   # COG
)
from rio_tiler_pds.cbers.aws import CBERSReader

All Readers are subclass of rio_tiler.io.BaseReader and inherit its properties/methods.

Properties

• bounds: Scene bounding box
• minzoom: WebMercator MinZoom (e.g 7 for Landsat8)
• maxzoom: WebMercator MaxZoom (e.g 12 for Landsat8)
• center: Scene center
• spatial_info: zooms, bounds and center

Methods

• info: Returns asset's (band) simple info (e.g nodata, band_descriptions, ....)
• stats: Returns asset's statistics (percentile, histogram, ...)
• metadata: info + stats
• tile: Read web mercator map tile from assets (bands)
• part: Extract part of assets (bands)
• preview: Returns a low resolution preview from assets (bands)
• point: Returns asset's pixel value for a given lon,lat

Other

• assets: List of available assets (bands) for each dataset

Scene ID

All readers take scene id as main input. The scene id is used internaly by the reader to derive the full path of the data.

e.g: Landsat on AWS

Because the Landsat AWS PDS follows a regular schema to store the data (s3://{bucket}/c1/L8/{path}/{row}/{scene}/{scene}_{asset}.TIF"), we can easily reconstruct the full asset's path by parsing the scene id.

from rio_tiler_pds.landsat.aws import L8Reader
from rio_tiler_pds.landsat.utils import sceneid_parser

sceneid_parser("LC08_L1TP_016037_20170813_20170814_01_RT")

> {
  'sensor': 'C',
  'satellite': '08',
  'processingCorrectionLevel': 'L1TP',
  'path': '016',
  'row': '037',
  'acquisitionYear': '2017',
  'acquisitionMonth': '08',
  'acquisitionDay': '13',
  'processingYear': '2017',
  'processingMonth': '08',
  'processingDay': '14',
  'collectionNumber': '01',
  'collectionCategory': 'RT',
  'scene': 'LC08_L1TP_016037_20170813_20170814_01_RT',
  'date': '2017-08-13'
}

with L8Reader("LC08_L1TP_016037_20170813_20170814_01_RT") as landsat:
    print(landsat._get_asset_url("B1"))

> s3://landsat-pds/c1/L8/016/037/LC08_L1TP_016037_20170813_20170814_01_RT/LC08_L1TP_016037_20170813_20170814_01_RT_B1.TIF

Each dataset has a specific scene id format:

File Per Band

rio-tiler-pds Readers assume that assets (eo:band) are stored in separate files.

$ aws s3 ls landsat-pds/c1/L8/013/031/LC08_L1TP_013031_20130930_20170308_01_T1/

LC08_L1TP_013031_20130930_20170308_01_T1_B1.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B10.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B11.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B2.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B3.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B4.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B5.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B6.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B7.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B8.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_B9.TIF
LC08_L1TP_013031_20130930_20170308_01_T1_BQA.TIF

When reading data or metadata, readers will merge them.

e.g

with S2L1CReader("S2A_L1C_20170729_19UDP_0") as sentinel:
    tile, mask = sentinel.tile(77, 89, 8, assets=("B01", "B02")
    assert tile.shape == (2, 256, 256)

    print(sentinel.stats(assets=("B8A", "B02")))
    > {
      'B8A': {
        'pc': [106, 9322],
        'min': 1,
        'max': 13659,
        'std': 2682.6511198930048,
        'histogram': [
          [261631, 52188, 137746, 98039, 41066, 30818, 21095, 8631, 1442, 105],
          [1.0, 1366.8, 2732.6, 4098.4, 5464.2, 6830.0, 8195.8, 9561.6, 10927.4, 12293.199999999999, 13659.0]
        ]
      },
      'B02': {
        ...
      }
    }

Changes

See CHANGES.md.

Contribution & Development

See CONTRIBUTING.md

License

See LICENSE.txt

Authors

The rio-tiler project was begun at Mapbox and has been transferred in January 2019.

See AUTHORS.txt for a listing of individual contributors.