Statistical post-processing of hydrological forecasts using Bayesian model averaging

TL;DR

This paper introduces a doubly truncated Bayesian model averaging method for post-processing hydrological ensemble forecasts, improving predictive accuracy for water levels by accounting for variable bounds and systematic errors.

Contribution

The paper presents a novel doubly truncated BMA approach tailored for water level forecasts, enhancing probabilistic prediction accuracy over existing methods.

Findings

Doubly truncated BMA outperforms raw ensemble forecasts.

The method reduces biases and dispersion errors.

Improves water level forecast reliability.

Abstract

Accurate and reliable probabilistic forecasts of hydrological quantities like runoff or water level are beneficial to various areas of society. Probabilistic state-of-the-art hydrological ensemble prediction models are usually driven with meteorological ensemble forecasts. Hence, biases and dispersion errors of the meteorological forecasts cascade down to the hydrological predictions and add to the errors of the hydrological models. The systematic parts of these errors can be reduced by applying statistical post-processing. For a sound estimation of predictive uncertainty and an optimal correction of systematic errors, statistical post-processing methods should be tailored to the particular forecast variable at hand. Former studies have shown that it can make sense to treat hydrological quantities as bounded variables. In this paper, a doubly truncated Bayesian model averaging (BMA) method, which allows for flexible post-processing of (multi-model) ensemble forecasts of water level, is introduced. A case study based on water level for a gauge of river Rhine, reveals a good predictive skill of doubly truncated BMA compared both to the raw ensemble and the reference ensemble model output statistics approach.