Answers are grounded in this paper and cite their sources.
Ensemble prediction systems constitute a fundamental component of contemporary meteorological and hydrological forecasting; however, their raw outputs frequently exhibit systematic biases and dispersion deficiencies that necessitate statistical postprocessing. Member-by-member postprocessing (MBMP) provides a flexible framework for correcting ensemble forecasts, yet its conventional linear formulation applies adjustments of the same functional form across ensemble members irrespective of their deviation from the ensemble mean, an assumption that may limit its ability to represent nonlinear error structures. This study investigates both linear and nonlinear extensions of the MBMP framework for temperature forecasting. In addition to the benchmark linear formulation, alternative linear variants were developed through the incorporation of different ensemble spread measures. To further explore structural flexibility, the linear transformations were replaced with logarithmic, radical, and exponential mappings, and a novel angular formulation was introduced to modulate member variability in a nonlinear manner. The proposed methods were applied to ECMWF ensemble forecasts over 32 catchments in Québec (Canada) at lead times of 2, 5, and 9 days. The results show that, among the linear variants, the formulation employing the median absolute deviation as the spread measure provides the most consistent improvement relative to the benchmark method, whereas the nonlinear variants do not systematically outperform the standard MBMP configuration. The angular MBMP exhibits promising behavior at shorter lead times in terms of accuracy, with reduced advantages at longer forecast horizons.
No takes yet. Share an insight, caveat, or question.
Oghbaei et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: