Decompose
Apply a continuous wavelet transform to observed and forecast precipitation.
WQM · Wavelet-based Quantile Mapping
Correct forecast bias across scales.
WQM post-processes precipitation forecasts in the time-frequency domain, correcting amplitude distributions at individual scales before reconstructing an ensemble of rainfall forecasts.
Methodology
Decompose → Correct → Rephase → Reconstruct.
Correct
Use quantile mapping to correct the wavelet amplitude distribution at each scale.
Rephase
Preserve corrected structure while phase shuffling represents timing uncertainty.
Reconstruct
Invert the adjusted time-frequency fields into an ensemble of precipitation forecasts.
Online demonstration
Inspect a verified reference run.
Official sample
Switch between the two bundled rainfall stations and compare observations, raw forecasts, deterministic bias correction and five ensemble members over the held-out half of each record.
Fixed reproducible settings
QDM · Morlet CWT · 50/50 calibration/validation split · 0.1 precipitation threshold · M2 phase shuffling · block size 3 · seed 2021.
Software and publication
Use the maintained R implementation.
WQM for R
Install the complete package for custom data, alternative settings and reproducible analysis.
GitHub ↗ · CRAN ↗Monthly Weather Review · 2023
A New Method for Postprocessing Numerical Weather Predictions Using Quantile Mapping in the Frequency Domain.
Read the method paper ↗Explore the WQM reference run.
Compare corrected precipitation forecasts and ensemble members without uploading data or starting a server-side job.