Preprints
https://doi.org/10.5194/gmd-2024-98
https://doi.org/10.5194/gmd-2024-98
Submitted as: development and technical paper
 | 
19 Jun 2024
Submitted as: development and technical paper |  | 19 Jun 2024
Status: this preprint is currently under review for the journal GMD.

Robust handling of extremes in quantile mapping – "Murder your darlings"

Peter Berg, Thomas Bosshard, Denica Bozhinova, Lars Bärring, Joakim Löw, Carolina Nilsson, Gustav Strandberg, Johan Södling, Johan Thuresson, Renate Wilcke, and Wei Yang

Abstract. Quantile mapping is a method often used for bias adjustment of climate model data toward a reference, i.e. to construct a transformation of the model’s distribution to that of the reference. The main moments of the distributions are typically well transformed by quantile mapping, but statistical uncertainty increases towards the extreme tails, making robust transformations challenging. Because of the limited data at the extreme tails, also an empirical quantile mapping needs to make some estimation or fit a parameterized function for data beyond the calibration data range. The MIdAS bias adjustment platform is here employed to explore different methods for handling the extreme tail, which are evaluated using an indicator for extreme precipitation - the maximum daily precipitation amount per year. Different methodologies are evaluated for a large ensemble of regional climate model projections over Scandinavia. The sensitivity of the empirical quantile mapping for the tails of the distribution is demonstrated, and it is found that the behaviour is significantly different within and without the calibration period, causing severe issues with the temporal consistency of the timeseries. The sensitivity is identified to be due to differences in the activated features of the bias adjustment, within the calibration period where the empirical transfer function is applied, and outside that period, where the extrapolation method is likely applied. This means that the bias adjustment method is in a sense different between different time periods. Currently MIdAS uses separate calibrations for each day of the year, as opposed to e.g. for each calendar month, further aggravates this issue. Further, finding a robust parametrisation for the tail is not straightforward. We identify a two-step solution which works well for this problem: (i) "murder your darlings" by excluding data from the tail data in the calibration period, the extrapolation feature is activated for all time periods, even the calibration period, and (ii) applying an outlier insensitive method for linear regression works well for finding an extrapolation parametrisation for the tail.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Peter Berg, Thomas Bosshard, Denica Bozhinova, Lars Bärring, Joakim Löw, Carolina Nilsson, Gustav Strandberg, Johan Södling, Johan Thuresson, Renate Wilcke, and Wei Yang

Status: open (until 14 Aug 2024)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • CEC1: 'Comment on gmd-2024-98', Juan Antonio Añel, 20 Jun 2024 reply
    • AC1: 'Reply on CEC1', Peter Berg, 28 Jun 2024 reply
  • RC1: 'Comment on gmd-2024-98', Anonymous Referee #1, 24 Jun 2024 reply
    • AC2: 'Reply on RC1', Peter Berg, 28 Jun 2024 reply
Peter Berg, Thomas Bosshard, Denica Bozhinova, Lars Bärring, Joakim Löw, Carolina Nilsson, Gustav Strandberg, Johan Södling, Johan Thuresson, Renate Wilcke, and Wei Yang
Peter Berg, Thomas Bosshard, Denica Bozhinova, Lars Bärring, Joakim Löw, Carolina Nilsson, Gustav Strandberg, Johan Södling, Johan Thuresson, Renate Wilcke, and Wei Yang

Viewed

Total article views: 273 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
200 58 15 273 5 3
  • HTML: 200
  • PDF: 58
  • XML: 15
  • Total: 273
  • BibTeX: 5
  • EndNote: 3
Views and downloads (calculated since 19 Jun 2024)
Cumulative views and downloads (calculated since 19 Jun 2024)

Viewed (geographical distribution)

Total article views: 273 (including HTML, PDF, and XML) Thereof 273 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 11 Jul 2024
Download
Short summary
When bias adjusting climate model data using quantile mapping, one needs to prescribe what to do at the tails of the distribution, where a larger range of data is likely encountered outside the calibration period. The end result is highly dependent on the method used, and we show that one needs to exclude data in the calibration range to activate the extrapolation functionality also in that time period, else there will be discontinuities in the timeseries.