Quantile regression ensemble for summer temperatures time series and its impact on built environment studies

The occurrence of heat waves, heavy rainfall, or drought periods have an increasing trend nowadays as consequence of Earth's global warming. This seriously affects natural habitats and directly impacts on the human environment. Architects and engineers use different approaches to model reference and future weather conditions to achieve building designs that resilient and energy efficient. However, modelling extreme weather events within those future conditions is a relative few explored field. This paper introduces the analysis of regression models for summer temperature time series that facilitate the representation of those extreme events; this is based on upper and lower quantiles, instead of the ordinary regression which is conditioned by the mean. The advantage of this proposal is to focus on finding patterns for both higher temperatures during the day and warmer temperatures during the night. These two temperatures are key in the study of overheating and therefore thermal related morbidity as the night time temperature will show the possibility of night purging. Advances on quantile regression models coming from both resampling and ensemble approaches are further investigated and compared with the state of the art. The results are sound in terms of statistical robustness and knowledge of inputs that specifically affect to temperature extremes. The new method has been tested with real data of temperatures in London for a period of 50 years. The quantile regression technique provides more meaningful information for use in building simulation than standard methods.