The European Space Agency

Features and Limitations

The FORCE integration into to CDSE is in a proof-of-concept stage. It is possible to use the FORCE level 2 and Time Series Analysis (TSA) modules on a small scale. Most parameters supported by FORCE are available through the openEO interface.

Unsupported parameters are usually either managed by the implementation (parallelization options, input/output directores) or reuqire supplementary files which are not avaiable on the backend. The aoi (openEO interface) / FILE_AOI (FORCE parameter file) parameter is good example of a parameter that is exposed in a different way in openEO to avoid the file interface. Instead of a shape file, a GEOJSON string is passed.

If you encounter limitations not mentioned here, please raise an issue in the Github repository

Features

  • FORCE level 2 processing through openEO on CDSE
  • FORCE Time Series Analysis (TSA) through openEO on CDSE
  • Automatic STAC generation for FORCE datacubes
  • Discover parameters of cloudified FORCE modules using the openEO client
  • Run level 2 processing once and re-use results without downloading multiple times
  • Area of interest selection (level 2) with GEOJSON: parameter aoi of force_level2.
  • Automatic DEM download based on selected area

Limitations

  • Each FORCE Job runs on a single virtual machine -> Not suited for large scale processing in a single job
  • The executors have limited memory available. Large jobs will not run successfully. You can parallelize time series analysis computations to avoid running into memory issues (see Tips & Tricks below).
  • Many parameters relying on auxiliary files are not supported. See the guide on parametrization for details.

TSA

  • Python and R user defined functions (UDFs) are not supported
  • Spectral Mixture Analyis is not supported

Tips & Tricks

Parallelizing large Time Series Analysis jobs

Server-side parallelization of the TSA module is not currently enabled. It is possible to parallelize TSA by creating one openEO job per tile by setting the x_tile_range and y_tile_range arguments to a single tile per job. Furthermore, consider computing spatio-temporal metrics (STM) in separate jobs to reduce memory usage.

Finally, you may reduce the chunk_size which will reduce memory usage at the cost of a longer processing time.