Release v4.7.x

2024-09-30

Version description of the v4.7.1 release of SlideRule Earth.

Bathy Version #4.

ATL24 Changes

  • v4.7.1 - Fixed default for parameter thresh in C-Shelph.

  • v4.7.1 - Fixed calculation of min_surface_photons_per_window and min_bathy_photons_per_window in OpenOceans++

  • v4.7.1 - Updated models to latest for QTrees, CoastNet, and Ensemble

  • v4.7.1 - Implemented use of geoid_corr_h for all classifiers except for the Ensemble. The Ensemble will be updated once it is retrained on the latest round of test results that incorporate the geoid_corr_h field. Previously, all classifiers were using non-refraction corrected height, but the ensemble was being trained using the refraction corrected height. Also, any post-test-run validation of the results only had the refraction corrected heights in the output. So the geoid_corr_h field was added so that it can clearly represent non-refraction corrected heights, while ortho_h includes the refraction correction (and any future corrections we might add).

  • v4.7.1 - QTrees, CoastNet, and OpenOceans++ have been updated with the latest code and are now compiled in as header only libraries.

  • v4.7.1 - MedianFilter and C-Shelph are pulled directly from the ut-ATL24-medianfilter and ut-ATL24-C-shelph repositories

  • v4.7.1 - Version information is included in for all external classifiers (CoastNet, QTrees, OpenOceans++, MedianFilter, C-Shelph)

  • v4.7.1 - Metadata is now provided inside the Parquet and HDF5 files instead of as separate json files.

  • v4.7.1 - Uncertainty calculation fixed (migration from python to C and the update to a linear model introduced a few regressions)

Known Issues and Remaining Tasks

  • v4.7.1 - Current ensemble model is trained on refraction corrected heights though the inputs are not corrected

  • v4.7.1 - Pointnet is identifying very low rates of bathymetry with respect to the other classifiers - we are still working through possible causes

Development Updates

  • v4.7.1 - The ATL03 photons are stored in a table in memory and operated on in place instead of streamed

  • v4.7.1 - The Python classifiers are all executed from a single Python script that uses multiprocessing instead of having the data transferred to and from disk

Getting This Release

https://github.com/SlideRuleEarth/sliderule/releases/tag/v4.7.1