Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to PEP 440 and uses Semantic Versioning.

[2.1.2]

Fixed

• The search functions in autoriftcoremodule.cpp, arPixDisp_ and arSubPixDisp_, now use OpenMP's dynamic thread scheduling for improved performance.

[2.1.1]

Fixed

• GeogridRadar.cpp no longer overwrites the provided lookSide variable. This means that GeogridRadar should now work with left-looking geometries.
• Updated numpy pad calls to work with newer numpy versions. Contributed by @AgentOxygen

[2.1.0]

Added

• Support for the use of pre-generated radar-geometry topographic corrections for ISCE3 when processing Sentinel-1 scenes.
• A citation.cff file to provide citation metadata for both humans and machines (e.g., zenodo integration).

[2.0.0]

Added

• An environment.yml file for creating a developmental conda environment

• ruff.toml configuration files for formatting python code with ruff

• .clang-format configuration files for formatting C/C++ code with clang-format

• Updating the changelog will now be required for all PRs into autoRIFT's main branch.

• geogrid C++ extension for the radar workflow which calls core ISCE3 C++ code directly.

Changed

• geogrid and autoRIFT now require Python 3.10 or greater
• All project documentation has been consolidated into the README.md and docs/demo.md, which have been updated to reflect new usage.
• Python's sysconfig is now used to get the Python include directory and purelib library paths instead of predicting it from the OpenCV build info.
• Geogrid and autoRIFT now report their version numbers from the geo_autoRIFT namespace package metadata
• The radar module in the Geogrid package has been changed from Geogrid.py to GeogridRadar.py.
• The radar version of Geogrid now uses ISCE3 rather than ISCE2.
• The test*.py workflows are no longer specific to optical or radar processing:
  • testGeogridOptical.py and testGeogrid_ISCE.py have been combined into a new testGeogrid.py script
  • testautoRIFT.py and testautoRIFT_ISCE.py workflow scripts have been merged into one testautoRIFT.py script
• Significant performance improvements to functions in autoRIFT.py:
  • arPixDisp_s and arPixDisp_u (float32 and uint8 pixel-wise displacement functions):
    • Moved the looping code into autoriftmodule.cpp
    • Multi-threading is now done via OpenMP in C++, rather than by using Python's multiprocessing library
    • Removed the loop_unpacking_ functions that were used for multiprocessing
  • colfilt (column-wise filter function):
    • All the filter methods have been manually implemented with Numba's @jit
    • The filters are now applied using SciPy's LowLevelCallable and generic_filter

Fixed

• The arPixDisp_s and arSubPixDisp_s C++ functions now use normalized cross-covariance (CV_TM_CCOEFF_NORMED) in the cv::matchTemplate calls, rather than normalized cross-correlation (CV_TM_CCORR_NORMED). This matches what the arPixDisp_u and arSubPixDisp_u workflows have been using. See here for the differences between the methods.

Removed

• The geo_autoRIFT/__init__.py file and all the sconscript files have been removed with the migration to ISCE3 -- these only served to facilitate building autoRIFT as a contributed package inside ISCE2.
• topsinsar_filename.py has been removed with the migration to ISCE3. This functionality has been moved to the get_topsinsar_config function in testautoRIFT.py.
• testautoRIFT.py no longer outputs a Matlab-style offest.mat file
• The geocode-only workflow added to testGeoGrid_ISCE.py in v1.6.0 with the switch to ISCE3. A similar workflow is currently being developed for a forthcoming release.

[1.6.0]

Added

• autoRIFT can now successfully process Sentinel-1 burst mosaics generated with burst2safe. We recommend processing at least 2 bursts along track, with 3-5 bursts seeing improvements in data quality.
• A geocode-only workflow was added to testGeogrid_ISCE.py that uses only the Sentinel-1 SAFE and orbit metadata without needing to run ISCE2 through the topo step. Thanks @leiyangleon!

Changed

• variety of updates in netcdf_output.py:
  • Now following CF Conventions 1.8 and (most of) NSIDC's recommendations; variable metadata may have changed accordingly. See the netcdf_output.py diff for the full list of changes: https://github.com/nasa-jpl/autoRIFT/compare/v1.5.0..v1.6.0
  • The value of the stable_shift netCDF attribute is now 0 instead np.nan if the stable shift has not been applied
  • 'satellite' global netCDF attribute will be spelled out and follow mission styling (e.g., Landsat 9)
  • M11/M12 now have the appropriate scale factor applied to ease use (e.g., when correcting ionosphere)
  • M11/M12 will now be stored as float32 instead int16 in the output netCDF file
  • Provider's image IDs will be saved as attributes in the img_pair_info variable
  • Ensure Landsat satellite_img1 and satellite_img2 attributes in the img_pair_info variable are strings to match the convention of other missions

Deprecated

In the upcoming v2.0.0 autoRIFT release:

• Support for using ISCE2 for the radar workflows will be removed in favor of an ISCE3-based workflow
• The test*.py workflows will no longer be specific to optical or radar processing:
  • testGeogridOptical.py and testGeogrid_ISCE.py will be combined into a new testGeogrid.py script
  • testautoRIFT_ISCE.py workflow scripts will be merged into the `testautoRIFT.py script

Fixed

• autoRIFT/Geogrid now support NumPy v2.0+, see the migration guide. Thanks @benhills, @gbrencher, and @sssangha!
• Fixed autoRIFT's Wallis filter to ensure non-negative variance and image masks are correctly combined

[1.5.0]

Added

• autoRIFT/Geogrid now support processing Landsat 4, 5, 7, and 9 scene

Changed

• autoRIFT/Geogrid now explicitly requires scenes to be in the same projection
• autoRIFT will now use a default filter width of 5 pixels, except for Sentinel-1 scenes where it'll use the previous default of 21

[1.4.0]

Added

• refined the workflow and ready for scaling the production of both optical and radar data results
• netCDF packaging for production, especially improved product quality for radar images

[1.3.1]

Added

• add condition check with error message in Geogrid when the two images have different projections

Fixed

• fix bug in calculating Geogrid pixel indices
• fix bug in autoRIFT column filter indices

[1.3.0]

Changed

• Merged PR #29 for improving the memory use and runtime for both autoRIFT and Geogrid

[1.2.0]

Changed

• Merged the PRs (#18 and #19) for using urls instead of local files & returning function outputs instead of saving as txt file
• Merged the PR (#26 and #27 ) for adding dt-varying search range routine

Fixed

• Merged the PR (#20) for fixing the indexing problem in optical coregistration,
• Merged the PR (#23) for dealing with dependent/overlapping chips (grid spacing < chip size)
• Merged the PR (#24) for MISC formatting/checking issues
• Merged the PR (#25) for fixing bugs related to nodata values

[1.0.0]

Added

• parallel computing for NCC
• support of urls for input files (both images and auxiliary input files can be urls for optical data processing; only auxiliary input files can be urls for radar data processing)

Changed

• updated netcdf packaging for production

[1.0.8]

Changed

• Vsicurl has been integrated to the optical data access without the need of downloading optical images (Landsat-8 and/or Sentinel-2) as well as initialization files (DEM, slope, reference velocity, etc).

[1.0.7]

Added

• autoRIFT was successfully compiled with OpenCV v3 and OpenCV v4

Changed

• eased Geogrid output handling (the program will automatically check the optional inputs provided by the user and only generate the meaningful output based on the input combinations)
• netcdf output packaging (not publicly available at this moment) for autoRIFT final products is significantly improved, where error metrics are calculated at the stable surfaces wherever available, and radar range-projected velocities onto a priori flow (e.g. surface slope parallel) are also provided when ionosphere-induced azimuth error are much larger.

Fixed

• bug of grid offset in Geogrid outputs has been fixed
• bug of erroneous large values (due to radar geometry's loss of sensitivity to geocode pixel displacement to velocity at large surface slopes) in Geogrid outputs has been fixed

[1.0.6]

[1.0.5]

[1.0.4]

[1.0.3]

Added

• compatible to both GDAL 2 and GDAL 3

[1.0.0]

Initial release of autoRIFT (autonomous Repeat Image Feature Tracking)