-
Notifications
You must be signed in to change notification settings - Fork 170
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Main module blending #233
Main module blending #233
Conversation
* Add xarray dependency * MCH importer returns an xarray Dataset * Remove plot lines * Remove import * Adapt readers to xarray format * Rewrite as more general decorator * Add missing metadata * Adapt io tests * Mrms bounding box (pySTEPS#222) * Fix bounding box coordinates * Add missing metadata * Import xarray DataArray Ignore quality field * Black * Do not hardcode metadata * Address review comments by ruben * Add a legacy option to the io functions A "legacy" options is added to revert back the importers and readers behavior to version 1. This is a temporary solution to allow the examples, and other functions, to run as usual (v1.*). Hopefully, this is will allow a easier transition into version 2 during the development process and will allow testing functions that were not updated to v2. * Fix compatibility problems with tests Many of the tests were updated to use the legacy data structures (v1). The tests that still contains issues, were tagged with a TODO message and they are skipped. This will allow incremental changes to be tested in the new v2. IMPORTANT: once the v2 branch is stable, we may remove the legacy compatibility from the code base and the tests. * Update dependencies * Ignore plugins test Co-authored-by: Andres Perez Hortal <16256571+aperezhortal@users.noreply.github.com>
…nt (in progress..)
…on of skill values to climatological values for weights determination
Good point. We implemented this as a place to store the decomposed NWP fields and NWP velocity fields, but seeing how it used now, we can reach the same with 'just' the |
I completely forgot to have a new look at the examples forecast after the latest changes. The minor code changes will follow soon. Indeed, something seems to go wrong with either the masking or prob_matching.. I don't see the same effect (at least it is not as clear if it is there) in the test forecasts with the RMI data, so I wonder what goes wrong. |
While working on the tests for |
Very nice @RubenImhoff! The animation is very promising! I'm having a closer look at the probability matching. Do I understand it right that you basically perform a linear blending of radar extrapolation and NWP based on the weights from the second cascade level in order to build the reference field for the probability matching? This is now a bit of a detail, but the problem with this is that you would loose the peak values, especially when both weights are around 0.5. An alternative approach would be a simple resampling: for each grid point you sample either the radar or the NWP value (and this could be based on the same weights you are already using). This way you would build a new target distribution that is not smoother than the original radar and NWP fields. I have some code for this (I used it in the EnKF approach). Should I give it a try to implement it in your code? |
Great idea, @dnerini, and thanks if you want to implement that! |
Yesterday, @ladc, @cvelascof and me had a nice talk with Alan Seed about the blending code and the weights implementation. He was happy to see that we're making this open source in pysteps and thought that the results so far look promising. Based on our talk, some main suggestions from Alan (and things to work on probably in the coming days):
|
Hi guys, sorry for the super late reply. I think we made the distinction because the one path is aimed at files for internal use, including input and output (also e.g. the weights file, pre-decomposed NWP, ...) and the other is where we store only the output products for users. But I also understand if you would prefer to reduce the number of different directories or parameters in the config file, and also store these in |
The xarray overhaul (among other things) is indeed not a small task, and it takes some manpower to work on pysteps v2. If you think it's doable to integrate the blending into pysteps v1.6, then it might be better to go for that option indeed. I hope this does not create too much duplicate work?
|
It should be doable in my opinion. To simplify the task, I suggest to exclude the nwp importers from the branch and publish them as separate pysteps plugins: https://pysteps.readthedocs.io/en/latest/developer_guide/importer_plugins.html The rest of the code is already very much in line with v1 I believe? I'd also suggest a minimal refactoring to reorganize the main method in few components (I'd be happy to give it a try). If we all agree, I'd start refactoring this PR so that it can be merged into master rather than pysteps-v2. @RubenImhoff very cool that you could exchange with Alan about your work! One comment from my side on your list of improvements: at this point I'd try to limit adding complexity to this implementation. We want a robust base blending method that works, possibly as close as possible to the reference publication. Further improvements can come incrementally later on (with your publication? ;-)), and as separate modules, to avoid adding too much complexity to the base method. For example: climatological skill. This is an important component for an operational system (it has to run in any situation), but I would not include it in the base method. Instead, we can make sure that the method can accept is as an optional argument, and leave to the user to compute it in his/her implementation. Same idea with the log sine transformation: would be nice to experiment with it, but this can be included in a separate PR. |
Sounds good, Daniele, and I agree! I'll upload a couple of minor changes and implement your idea to make the decomposed NWP input optional (so, you could also start with a non-decomposed NWP forecast) and then we're ready to go. |
…th_workdir from pystepsrc
Refactoring of main steps blending code
I've just merged Daniele's great refactoring work. Should we continue the review here, or should I merge this branch in the steps_blending branch and renew that for pysteps 1.6 (and review from there)? |
…ng changes to prevent errors in blending.utils
Very good. Yes, please merge into steps_blending and we'll continue from there for the integration into master |
mmmh maybe that's not a good idea if we want to merge into master after that. Let's instead merge into a new branch from master, I'll do it now. |
* First basic functions to implement STEPS blending * Add compute of blend means,sigmas and recompose * pysteps.io with xarray (#219) * Add xarray dependency * MCH importer returns an xarray Dataset * Remove plot lines * Remove import * Adapt readers to xarray format * Rewrite as more general decorator * Add missing metadata * Adapt io tests * Mrms bounding box (#222) * Fix bounding box coordinates * Add missing metadata * Import xarray DataArray Ignore quality field * Black * Do not hardcode metadata * Address review comments by ruben * Add a legacy option to the io functions A "legacy" options is added to revert back the importers and readers behavior to version 1. This is a temporary solution to allow the examples, and other functions, to run as usual (v1.*). Hopefully, this is will allow a easier transition into version 2 during the development process and will allow testing functions that were not updated to v2. * Fix compatibility problems with tests Many of the tests were updated to use the legacy data structures (v1). The tests that still contains issues, were tagged with a TODO message and they are skipped. This will allow incremental changes to be tested in the new v2. IMPORTANT: once the v2 branch is stable, we may remove the legacy compatibility from the code base and the tests. * Update dependencies * Ignore plugins test Co-authored-by: Andres Perez Hortal <16256571+aperezhortal@users.noreply.github.com> * Add blend_optical_flow * changes to steps blending procedure - weights according to adjusted BPS2006 method * changes to blending procedures - adjust weights from original BPS2006 method * Determine spatial correlation of NWP model forecast * First attempt to make correlations and thus weights lead time dependent (in progress..) * Change back to original BPS2006 blending formulation and add regression of skill values to climatological values for weights determination * Reformat code with Black * Skill score script imports climatological correlation-values from file now * Small changes to skill score script * Add skill score tests and an interface * Add skill score tests and an interface * Small change to docstring * Bom import xarray (#228) * Add import_bom_rf3 using xarray * Add tests to xarray version * Fix mrms importer tests * Pass **kwargs to internal functions * Add nwp_importers to read bom nwp sample data * Add bom nwp data to source file * Add tests for bom_nwp reader * Fix pystepsrc Co-authored-by: Andres Perez Hortal <16256571+aperezhortal@users.noreply.github.com> * Functions to store and compute climatological weights (#231) * Implement the functions get_default_weights, save_weights, calc_clim_weights. These functions are used to evolve the weights in the scale- and skill-dependent blending with NWP in the STEPS blending algorithm. The current weights, based on the correlations per cascade level, are regressed towards these climatological weights in the course of the forecast. These functions save the current and compute the climatological weights (a running mean of the weights of the past n days, where typically n=30). First daily averages are stored and these are then averaged over the running window of n days. * Add tests for pysteps climatological weight io and calculations. * Add path_workdir to outputs section in pystepsrc file and use it as a default path to store/retrieve blending weights. * Minor changes to docstrings, changes to skill scores and testing scripts * Completed documentation for blending clim module, cleanup. Co-authored-by: RubenImhoff <r.o.imhoff@live.nl> * Main blending module, first steps * Add simple tests * Minor changes to tester: velocity now based on rainfall field of NWP * Add utilities to decompose, store and load NWP cascades for use in blending (#232) * First version of NWP decomposition * Added saving to netCDF * Completed functions for saving and loading decomposed NWP data * Added example files for the decomposed NWP functions * Added compatibility with numpy datetime * Use default output path_workdir for tmp files in blending/utils.py. * Update documentation of NWP decomposition functions in utils.py Co-authored-by: Wout Dewettinck <wout.dewettinck@ugent.be> Co-authored-by: wdewettin <87696913+wdewettin@users.noreply.github.com> * Add importer for RMI NWP data (#234) Add importer for netcdf NWP data from RMI using xarrays. * Add test for RMI NWP data importer. * Add entry for RMI NWP data in pystepsrc. * Run black on everything: fix formatting. * Add KNMI Harmonie NWP netcdf importer and tests (#235) * Changes to v_models to make use of multiple timesteps. Changes in the velocity field over time in the NWP forecast will be taken into account now. * Fixes for KNMI importer: Add forgotten @postprocess_import() Don't call dropna on NWP data. * Avoid shadowing of pysteps.blending.utils by pysteps.utils * First attempt for probability matching and masking utility; part 1 * Changes to prob matching and masking methods; part 2 * Prob matching and masking changes; part 3. Ready for testing with real data from here on * Remove unnecessary print statements * Cleanup imports * More cleanup * Update docstrings * RMI importer for gallery example (will follow) * Reprojection functionality (#236) * Added Lesley's reprojection module to this branch * Added compatibility for three-dimensional xarrays * Add commentary to reprojection util * Changes to make reprojection of KNMI data possible * Changes after Daniele's review * Add dependencies * Changes to importers, see issue #215 * Add tests * Fix some issues * documentation * Fixes for tests * Set requirements again * Some fixes * Changes to nwp_importers after Carlos' response * Remove wrong example script * Remove rasterio dependencies from lists * First try to prevent testing error * Changes Daniele and fix knmi nwp importer * Add rasterio to tox.ini * Aesthetics * rasterio import test * Add rasterio to the test dependencies * Reset try-except functionality for rasterio import * Fix for failing test on windows python 3.6 * add importerskip rasterio Co-authored-by: Wout Dewettinck <wout.dewettinck@ugent.be> * Fixes in nwp importers * Revert "Merge branch 'steps_blending' into pysteps-v2" (#239) This reverts commit 2c639f8, reversing changes made to bccb8fc. * Merge latest version pysteps-v2 into steps_blending branch (#237) * Update docstrings * More cleanup * Cleanup imports * Cleanup imports * More cleanup * Update docstrings * Update references Mention the work of Ravuri et al (2021, Nature) as an example of work using cGANs to generate ensembles * Clean up page * Reprojection functionality (#236) * Added Lesley's reprojection module to this branch * Added compatibility for three-dimensional xarrays * Add commentary to reprojection util * Changes to make reprojection of KNMI data possible * Changes after Daniele's review * Add dependencies * Changes to importers, see issue #215 * Add tests * Fix some issues * documentation * Fixes for tests * Set requirements again * Some fixes * Changes to nwp_importers after Carlos' response * Remove wrong example script * Remove rasterio dependencies from lists * First try to prevent testing error * Changes Daniele and fix knmi nwp importer * Add rasterio to tox.ini * Aesthetics * rasterio import test * Add rasterio to the test dependencies * Reset try-except functionality for rasterio import * Fix for failing test on windows python 3.6 * add importerskip rasterio Co-authored-by: Wout Dewettinck <wout.dewettinck@ugent.be> * Revert "Merge branch 'steps_blending' into pysteps-v2" (#239) This reverts commit 2c639f8, reversing changes made to bccb8fc. Co-authored-by: ned <daniele.nerini@meteoswiss.ch> Co-authored-by: dnerini <daniele.nerini@gmail.com> Co-authored-by: Wout Dewettinck <wout.dewettinck@ugent.be> * NWP skill calculation only within radar domain * Update docs * Add example for gallery examples * Fix docstrings example * Remove additional normalization step * Fixes for the tests * update docs * changes to post-processing rainfall field and docstrings * Update contributing guidelines (#241) - Improve grammar. - Make the guide more concise. Remove unused/unnecessary rules. - Indicate more clearly which parts of the guidelines are inspired by other projects (before they were only mentioned at the end). - Change "Travis-CI" references by "GitHub Actions". * Advect noise cascade * Allow for moving domain mask of extrapolation component * minor fixes * Linear blending (#229) * Implemented linear blending function * Added example file and test * Added compatibility for NWP ensembles The PR is ready to go. Making the code xarray ready will be done in a separate PR. Co-authored-by: RubenImhoff <r.o.imhoff@live.nl> * weights calculation adjustment outside radar domain if only one model present * allow for mirroring of advected noise cascade * implementation of weights following Seed et al. (2013) * Allow for decomposed NWP precip and NWP velocity fields: part 2 * Store decomposed fields with compression * changes after first review Daniele * Remove unnecessary print statement * fixes to blending utils and implementation of blending utils tests * remove unnecessary lines * Fix one time step shift of extrapolation skill prior to blending * minor changes to blending climatology, blending weights and remove path_workdir from pystepsrc * Make NWP forecast decomposition prior to blending function optional * Use pathlib * Extract methods * Minor changes to docstrings * Access climatological skill file for multiple NWP model and date string changes to prevent errors in blending.utils Co-authored-by: Carlos Velasco <carlos.velasco@bom.gov.au> Co-authored-by: ned <daniele.nerini@meteoswiss.ch> Co-authored-by: Andres Perez Hortal <16256571+aperezhortal@users.noreply.github.com> Co-authored-by: Ruben Imhoff <Ruben.Imhoff@deltares.nl> Co-authored-by: Carlos Velasco <cvelascof@gmail.com> Co-authored-by: Lesley De Cruz <lesley.decruz+git@gmail.com> Co-authored-by: Wout Dewettinck <wout.dewettinck@ugent.be> Co-authored-by: wdewettin <87696913+wdewettin@users.noreply.github.com> Co-authored-by: Lesley De Cruz <lesley.decruz@meteo.be> Co-authored-by: dnerini <daniele.nerini@gmail.com>
First 'steps' towards a main module for a blended nowcasts-NWP forecast following the original STEPS approach as described in Bowler et al. (2006) and Seed et al. (2013).
Part of issue #212