Xsdba for CHIRPS-GEFS bias correction product

[nishadhka]

The CHIRPS-GEFS bias correction currently uses the routines provided here:

• Dataset: https://chc.ucsb.edu/data/chirps-gefs
• Source code (IDL-based): https://data.chc.ucsb.edu/people/marty/CHIRPS-GEFS_code/

The xsdba.EmpiricalQuantileMapping method appears to implement a very similar approach to the CHIRPS-GEFS quantile-mapping bias correction. It would be valuable to initiate an intercomparison study to evaluate how well xsdba reproduces or improves upon the existing CHIRPS-GEFS methodology.

With the advent of Zarr and analysis-ready, cloud-optimized datasets, this presents a good opportunity.

The three main datasets required for such an implementation are:

1. GEFS reanalysis dataset: https://dynamical.org/catalog/noaa-gefs-analysis/
2. GEFS real-time forecast data: GEFS GRIBs in S3: Scan GRIB and Fast Referencing Using Index Files with Zarr v2 — 24-Hour Accumulation Plot and Comparison with Dynamical.org GEFS fsspec/kerchunk#572
3. Virtual Zarr creation of CHIRPS daily data for operational use.

I am looking forward to work on this aspect to learn more about xsdba. Thank you for the library.

[coxipi]

Hi!

Thanks for your interest in xsdba and for sharing these resources. It will be interesting to see how EmpiricalQuantileMapping compares with another method if you want to share your explorations. You could also look at xsdba.DetrendedQuantileMapping and xsdba.QuantileDeltaMapping. These are methods similar to the basic quantile mapping, but are designed to take into account how statistical distributions are changing with climate change. See e.g. https://journals.ametsoc.org/view/journals/clim/28/17/jcli-d-14-00754.1.xml

[nishadhka]

Thank you for the note, started with llm comparison notes on the IDL scripts method https://data.chc.ucsb.edu/people/marty/CHIRPS-GEFS_code/ with xsdb https://github.com/icpac-igad/DevOps-hazard-modeling/blob/main/chirps-gefs/README.md, albeit on old version with xclim. The Cannon et al 2015 paper shared gives the need to address Non-Stationarity Issues with the quantile mapping methods, excited to know xsdba could address that.