Comparing Bayesian Model Averaging and Reliability Ensemble Averaging in Post-Processing Runoff Projections under Climate Change

This study investigated the strength and limitations of two widely used multi-model averaging frameworks—Bayesian model (BMA) reliability ensemble (REA), in post-processing runoff projections derived from coupled hydrological models climate downscaling models. The performance weight distributions five ensembles were thoroughly compared, including simple equal-weight averaging, BMA, REAs optimizing mean (REA-mean), maximum (REA-max), minimum (REA-min) monthly runoff. results suggest that REA BMA both can synthesize individual models’ diverse skills with comparable reliability, despite their different strategies assumptions. While weighs candidate by predictive baseline period, also forces to approximate a convergent projection towards long-term future. type incorporation uncertain future weighting criteria, as well differences parameter estimation (i.e., expectation maximization (EM) algorithm Markov Chain Monte Carlo sampling method REA), tend cause larger uncertainty ranges ensembles. Moreover, our show objectives could much discrepancy than induced criteria or algorithms. Among three ensembles, REA-max most resembled because EM converges aggregated error, thus emphasize simulation high flows. REA-min achieved better terms inter-annual temporal pattern, yet at cost compromising accuracy capturing behaviors. Caution should be taken strike balance among features interest.