Evaluating climate model accuracy for building energy predictions: A performance-based multi-model ensemble approach

Climate change is projected to significantly raise global temperatures by 1.5°C to 4.5°C above pre-industrial levels by the end of the century, profoundly impacting building energy use due to increased cooling demands during hotter summers. Accurate modeling of future temperatures is essential for generating weather files used in building energy models, which are critical for designing energy-efficient buildings. To identify accurate models for future weather predictions, models must be evaluated on their ability to replicate historical weather data. It is assumed that models performing well in predicting historical weather data will also accurately predict future weather data. Previous research has either relied on single models or equally weighted multi-model ensembles. This study evaluates location-specific performance-based multi-model ensembles and compares them to averaged weighted multi-model ensembles.

This study leverages 27 climate models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) across 16 U.S. cities representing diverse climate zones. The CMIP6 models offer improved spatial resolution and better representation of physical processes compared to previous iterations, enhancing their predictive capability. Model performance is rigorously evaluated by comparing historical data predictions with actual weather station temperature data, enabling the identification of the best-performing models. This refined approach informs predictions for future building energy consumption under various climate change scenarios.

Results demonstrate that performance-based weighting significantly reduces bias and enhances prediction accuracy. Specifically, the performance-based method reduces the Root Mean Square Error (RMSE) Standard Deviation Ratio (RSR) by an average of 3.5% to 8.0% for mean, minimum, and maximum temperatures. Additionally, it achieves an average reduction of 42.0% to 68.3% in Normalized Mean Bias Error (NMBE) for mean, minimum, and maximum temperatures. Significant reductions in bias are particularly evident in warm and dry climate zones, which are particularly susceptible to the impacts of climate change.

Overall, this study underscores the critical role of performance-based weighting in enhancing the accuracy and reliability of climate predictions. By effectively reducing both error and bias, this approach offers a more nuanced and region-specific method for climate modeling. The significant bias reduction in warm and dry climates, which are more susceptible to the impacts of climate change, highlights the importance of accurate climate projections for optimizing future building energy efficiency and resilience. The substantial improvements observed across multiple locations and temperature metrics affirm the superiority of performance-based weighting, paving the way for more precise and trustworthy climate forecasts in future studies.