On the accuracy of Urban Building Energy Modelling

The growing demand for energy in urban areas has led to the development of a variety of methodologies for modelling energy in buildings at large scale. However, their accuracy has yet to be thoroughly reviewed. This paper presents a systematic analysis of urban building energy models, that have been validated against measured data, using a singular taxonomy based on key attributes that could influence a model’s accuracy: application, scale, input data, computational method, calibration and validation methods. The analysis showed that the accuracy of urban building energy models is multi-dimensional, considered at a variety of temporal resolutions, spatial resolutions and measures of error, with the results demonstrating that there is no single key attribute that governs it. At the aggregate spatial and annual temporal resolutions, the accuracy, often reported in a single percent error value, can be as low as 1%, while for individual buildings at the annual resolution, the tails of the distribution of errors can reach 1000%. Models using non-calibrated physics-based computational methods were more likely to report overly large errors, while those employing Bayesian calibration consistently reported lower errors at the hourly temporal resolution, demonstrating the positive impact of calibration and in particular the Bayesian approach, on the models’ accuracy. Overall, the review has highlighted that more transparent and consistent reporting of accuracy is necessary and further research is essential for improving the evaluation of accuracy in modelling methodologies, if modern challenges are to be met through emerging applications such as energy systems integration and climate resilience.