pyeogpr

Python based machine learning library to use Earth Observation data to map biophysical traits using Gaussian Process Regression (GPR) models.

Features

• Access to openEO is required. Works best with the Copernicus Data Space Ecosystem. Register here or here
• Hybrid retrieval methods were used: the Gaussian Process Regression retrieval algorithms were trained on biophysical trait specific radiative transfer model (RTM) simulations
• Built-in gap-filling to avoid cloud covers
• Runs "in the cloud" with the openEO API. No local processing is needed.
• Resulting maps in .tiff or netCDF format

Get started

You can install pyeogpr using pip. Read the documentation

pip install pyeogpr

Basic example:

import pyeogpr

# Your region of interest
bounding_box = [
         -73.98605881463239,
          40.763066527718536,
          -73.94617017216025,
          40.80083669627726
        ]

# Time window for processing Satellite observations
time_window = ["2022-07-01", "2022-07-07"]

dc = pyeogpr.Datacube(
    "SENTINEL2_L2A",  # Satellite sensor
    "FVC",            # Fractional Vegetation Cover
    bounding_box,
    time_window,
    cloudmask=True
)

dc.construct_datacube("dekad")  # Initiates openEO datacube

dc.process_map()  # Starts GPR processing

To download the GPR processed map go to the openEO portal:

You can use QGIS or Panoply to visualize. IMPORTANT: The data range is off, due to few pixels being outliers. Set the data range manually for the corresponding variable e.g. FVC--> 0 to 1.

Satellites and biophysical variables

You can select from a list of trained variables developed for the following satellites:

Sentinel-2 L1C

Sentinel-2 L2A

Sentinel-3 OLCI L1B

Cite as

Dávid D.Kovács. (2024). pyeogpr (zenodo). Zenodo. https://doi.org/10.5281/zenodo.13373838

Supported by the European Union (European Research Council, FLEXINEL, 101086622) project.