Top-down spatially-explicit probabilistic estimation of building energy performance at a scale

Abstract

Achieving the energy-related and environmental targets for nations and municipalities is largely dependent on the existing built stock. It plays a pivotal role in the accomplishment of these targets through the implementation of energy efficiency and flexibility programs, involving the deployment of distributed energy resource management technologies, refurbishment of building envelopes and upgrading of indoor environmental control equipment. Spatial awareness about urban energy use enables to prioritise the areas where these solutions will be most effective and balanced with the plans for new constructions. Large-scale building energy mapping, however, must cope with heterogeneity of buildings within the built stock, absence of detailed information and multiple sources of uncertainty that stem from the complex and dynamic properties of the phenomenon at a building level. One of the key challenges in the discipline is to account for these uncertainties while maintaining the rational model complexities and data needs. This study, therefore, suggests a parsimonious top-down probabilistic modelling recipe to enable geospatial energy mapping and analysis. Under such modelling principles, an inverse propagation of uncertainties is carried out from the status quo of the built stock. The proposed framework is based on probabilistic sampling with prior parametric univariate density estimation and statistical hypothesis testing. Consolidation with the exogenous influencing factors is facilitated through the measure of statistically significant difference. This approach is exemplified with the data from two sources: the cadastral system and the energy performance certificates registry. A case study developed for Trondheim (Norway) quantified the central tendency and dispersion in the distributions of the simulated bulk total annual energy use by buildings per 1×1km grid cell over the urban territory. The results suggest that best estimates of these values vary between 11MWh·y-1 and 141GWh·y-1 depending on the grid cell. A measure of dispersion in the simulated results is highly correlated with these estimates. Robust handling of uncertainties and the possibility to accommodate a variety of modelling objectives make this approach practical for energy mapping with a flexible spatial resolution that may facilitate numerous applications in energy planning. A collection of methods for univariate density estimation discussed in this study together with the empirical data are accessible through Built Stock Explorer:https://builtstockexplorer.indecol.ntnu.no. This open web application for knowledge discovery in building energy data enables to reproduce some of the results presented in the article.