Newer
Older
- After last minute community feedback (thank you [Sergi Bech Sala](https://renkulab.io/gitlab/sbechsal7) and [Llorenç Lledó](https://renkulab.io/gitlab/llledo)), we decided to modify the way we average the RPS for RPSS. Following [Weigel et al. 2007](https://doi.org/10.1175/MWR3280.1) [#50](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/50), [!23](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/merge_requests/23) implements `RPSS=1-<RPS_ML>/<RPS_clim>`, whereas before we had `RPSS=<1-RPS_ML/RPS_clim>`, where angle brackets `<>` denote the average of the scores over a given number of forecast–observation pairs. Furthermore, we now penalize `NaNs` where numerical values where expected by `RPS=2`.
- We restart the [`s2saichallengescorer`](https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image/-/tree/master), which automatically fetches the new scores of previous submissions, so your old submissions are re-evaluated. If you want to evaluate your score yourself, please use the updated [`skill_by_year`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/scripts.py#L200).
---
**NOTE**
After community feedback, we changed the ground truth observations files. Please update these files into your training pipeline if you used them before:
- [hindcast-like-observations_2000-2019_biweekly_tercile-edges.nc](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/data/hindcast-like-observations_2000-2019_biweekly_tercile-edges.nc)
- [hindcast-like-observations_2000-2019_biweekly_deterministic.zarr](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/tree/master/data/hindcast-like-observations_2000-2019_biweekly_deterministic.zarr)
- [hindcast-like-observations_2000-2019_biweekly_terciled.zarr](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/tree/master/data/hindcast-like-observations_2000-2019_biweekly_terciled.zarr)
---
- add `eccodes` to [`environment.yml`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/environment.yml) to resolve `cfgrib` `ImportError` (https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/26, !15, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- Create categories by `groupby('week')`. (https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/29, !18 [Aaron Spring](https://renkulab.io/gitlab/aaron.spring) with help from [Sergi Bech Sala](https://renkulab.io/gitlab/sbechsal7) and [Llorenç Lledó](https://renkulab.io/gitlab/llledo))
- Regrid observations `conservative`ly with [xesmf](https://pangeo-xesmf.readthedocs.io/) (https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/32, !18, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring) with help from [Arlan Dirkson](https://renkulab.io/gitlab/arlan.dirkson))
- Observations now have data at `longitude=0` (https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/35, !18, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring) with help from [Sergi Bech Sala](https://renkulab.io/gitlab/sbechsal7))
- Missing `tp` observations after leap days added (https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/31, !18, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring) with help from [David Landry](https://renkulab.io/gitlab/david.landry))
- `conservative`ly regridded and `groupby('week')` observations are now also used by `s2saichallengescorer` (https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image/-/merge_requests/4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
### 2021-06-19: `v0.3.1` *release*
- Fix `sortby('forecast_time')` for renku datasets (as before). `renku dataset s2s-ai-challenge` tag equals repo tag (!15, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
### 2021-06-19: `v0.3.0` *release*
- Recreate biweekly renku datasets with `float32` single precision and newly uploaded data to [`climetlab_s2s_ai_challenge`](https://github.com/ecmwf-lab/climetlab-s2s-ai-challenge/releases/tag/0.7.0). Before `lead_time` in `tp` had a one day shift. These new renku datasets will be used by the [`s2saichallengescorer`](https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image/-/merge_requests/3). (#3, #5, !14, https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge/-/issues/22, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring)). Averaged recalibrated ECMWF RPSS skill value to beat at least: `RPSS = -0.043`, see [`RPSS_verification.ipynb`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/RPSS_verification.ipynb). You should also beat `climatology`, i.e. `RPSS = 0`.
- Order of processing gridded `RPSS` to final score: (#7, !9, [s2s-ai-competition-scoring-image!2](https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image/-/merge_requests/2), [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
1. `RPSS`
2. penalize where `NaN` submitted but value expected #7
3. `clip(-10,1)`: prevent too negative values
4. mean over `forecast_time`
5. spatially weighted mean [90N-60S]
6. mean over `lead_time` and `data_vars`
- Dont forget to `git add current_notebook.ipynb` also to ensure that consistent training pipeline and submission file are tagged, added to notebooks. (!9, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- Rerun [`ML_train_and_predict.ipynb`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/ML_train_and_predict.ipynb) (!9, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- Fix typo in safeguards in [ML_forecast_template.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/ML_forecast_template.ipynb): "We did NOT use `test` explicitly in training or implicitly in incrementally adjusting parameters." (!8, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- Add notebooks showcasing accessing output of different models from different sources: (!2, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- S2S-Project models:
- from from European Weather Cloud:
- [`climetlab-s2s-ai-challenge`](https://github.com/ecmwf-lab/climetlab-s2s-ai-challenge/) [recommended], see [`climetlab-s2s-ai-challenge` notebooks](https://github.com/ecmwf-lab/climetlab-s2s-ai-challenge/tree/main/notebooks)
- `curl` & `wget`, see [wget_curl.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/wget_curl.ipynb)
- `intake`, see [intake.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/intake.ipynb)
- `IRIDL` including overview, see [IRIDL.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/IRIDL.ipynb)
- `intake` catalogs for `IRIDL` [`SubX`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/SubX_catalog.yml) and [`S2S`](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/S2S_catalog.yml) see [IRIDL.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/IRIDL.ipynb) (#9 !10, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- SubX-Project models: `IRIDL` including overview, see [IRIDL.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/IRIDL.ipynb)
- How to access password-protected S2S-Project output from `IRIDL` with `xarray`? see [IRIDL.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/IRIDL.ipynb)
- Access NOAA CPC observations via opendap from `IRIDL`, see [IRIDL.ipynb](https://renkulab.io/gitlab/aaron.spring/s2s-ai-challenge-template/-/blob/master/notebooks/data_access/IRIDL.ipynb) (#9 !10, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- fix `netcdf4` version to `1.5.4` for `opendap` to work lazily with `xarray` (!2, !7, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
After this `v0.2` release, this `CHANGELOG.md` will describe all changes made in this template repository.
- update `README` how to join competition, please `git pull` if you forked before
- find status of your submission in `s2s-ai-competition-scoring-image` https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image/-/blob/master/README.md
- calculate `RPSS` with respect to climatology (not ECMWF anymore) ([Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- update `scorer`: https://renkulab.io/gitlab/tasko.olevski/s2s-ai-competition-scoring-image ([Tasko Olevski](https://renkulab.io/gitlab/tasko.olevski), [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- Averaged ECMWF RPSS skill value to beat at least: `-0.0070`
- update `README` how to join competition (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- git lfs track zarr: `git lfs track "**/*.zarr/**"` (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- add notebooks: (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- create renku datasets: `renku_datasets_biweekly.ipynb`
- RPSS verification: `RPSS_verification.ipynb`
- ML train and predict based on weatherbench: `ML_train_and_predict.ipynb`
- mean bias reduction: `mean_bias_reduction.ipynb`
- template for training and predictions: `ML_forecast_template.ipynb`
- add renku dataset `s2s-ai-challenge` with files: (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- `hindcast-like-observations_2000-2019_biweekly_deterministic.zarr`
- `forecast-like-observations_2020_biweekly_deterministic.zarr`
- `hindcast-like-observations_2000-2019_biweekly_tercile-edges.nc`
- `hindcast-like-observations_2000-2019_biweekly_terciled.zarr`
- `forecast-like-observations_2020_biweekly_terciled.nc`
- `ecmwf_forecast-input_2020_biweekly_deterministic.zarr`
- `ecmwf_hindcast-input_2000-2019_biweekly_deterministic.zarr`
- `ecmwf_recalibrated_benchmark_2020_biweekly_terciled.nc`
- add reproducibility section below in training (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- how to deal with this dry mask? provide as renku dataset? now implicitly masked in categorized observations `obs_p`
- justify if training takes more than a week (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
- show RPS for all years. ~~ToDo: take RPSS~~ (!4, [Aaron Spring](https://renkulab.io/gitlab/aaron.spring))
### 2021-05-10
- `git lfs track “submissions/*.nc”`, so submission netcdf files are not carried in `git` but `git lfs`
- add `notebooks/ML_prediction.ipynb` as a template for submission notebooks with safeguards.