Evaluating the Sensitivity of Hydrological Impacts to Different Climate Model Weighting Strategies

The use of Multi-model ensembles (MMEs) has become crucial in assessing future climate change impacts and uncertainties. These ensembles leverage simulations from various global climate models (GCMs). While the traditional "model democracy" method, where equal weights are assigned to all models, has succeeded in reproducing the mean state of historical climate, it faces challenges in hydrological impact studies. Two key criticisms prompt the investigation of model democracy: the diverse performance of GCMs across different variables and locations, and the assumption of independence among ensemble members. Shared modules and features in climate models may introduce common biases, affecting confidence in projection uncertainty and potentially increasing uncertainties in climate change predictions. To address these challenges, diverse weighting approaches are explored, assigning varying weights to GCMs based on their performance in diagnostic metrics. While equal weighting is a common approach, unequal-weighting methods aim for a more reliable ensemble mean or constrained uncertainty.

This study assesses five weighting schemes—equal weighting, random weighting, skill-based weighting, the representation of annual cycle (RAC), and Reliability Ensemble Averaging (REA)—in hydrological impact evaluations. We utilized data from A set of 22 CMIP6 GMCs, coupled with a lumped hydrological model, and one bias correction method across 3107 North American catchments during the 1971-2000 and 2071-2100 periods. To understand how weighting methods influence streamflow bias in future periods, we used a "pseudo-reality" method, which involves comparing the bias between the weighted mean of climate models and a selected model used as a reference dataset. Through multiple iterations considering climate variables and geographic regions, this research aims to uncover the complex interactions between weighting schemes and their implications for hydrological assessments.

Our findings indicate that the performance of equal weighting and other weighting methods are similar in cases where bias correction has been applied. Bias correction is commonly used in climate change impact assessments due to the inherent inaccuracies in climate models, and in such cases equal weighting approach would provide adequate results for climate change impact assessment studies. For scenarios without bias correction, applying unequal weights provides improved simulation performance and reliability. The findings of this study contribute valuable insights to the broader landscape of climate change impact studies, emphasizing the importance of tailored weighting strategies in enhancing the reliability of hydrological assessments.