Air pollution has been identified as one of the leading causes of morbidity and mortality
worldwide. The current trend is predicted to continue until 2040 by the International Energy
Agency (IEA) forecasts. It is estimated that ambient fine particles (PM2.5) caused 103.1 million
disability-adjusted life-years (DALYs) in 2015. As indoor pollutant concentrations, including
PM2.5, can be even higher than those outdoors, the indoor environments of homes and workplaces
may significantly impact population exposure. This doctoral thesis presents a study of indoor air
quality in higher education institution (HEI) buildings. In the UK, most universities are located
in high-density urban built areas, and air pollution from urban traffic and other sources is the most
significant contributor to poor indoor air quality (IAQ). Since people spend long hours indoors
working in HEI buildings, there is a concern about chronic exposure to indoor air pollutants such
as PM2.5.
The main challenge addressed in this research is the high level of heterogeneous characteristics
observed in HEI buildings that require many input parameters in developing building stock IAQ
models to inform planning and design for better air quality. Robust HEI building stock IAQ
models are required for estimating university population exposure to indoor PM2.5 from outdoor
sources throughout the year. This thesis shows how such estimations can be achieved reliably by
a reduced set of input parameters at an HEI building stock level. The IAQ modelling focuses on
the annual heating season (November-April) when higher outdoor PM2.5 levels often appear
during winter in the UK. Based on the outputs of infiltrated PM2.5 concentrations, the HEI stock
IAQ model is applied to evaluate the impact of increasing the building envelope airtightness (Q50)
measure on population exposure.
Five buildings from the University of Sheffield (UoS) were selected and modelled in CONTAM
and EnergyPlus using available data sources, such as the Estates and Facilities Management
(EFM) and local building regulations and guidelines. The buildings are modelled with multiple
Q50 values ranging between 3 – 13 m3
/h/m2
to generate indoor PM2.5 concentrations due to
infiltration at a zone/room level (N =2,729 zones) during the heating season. An analytical
framework employing sensitivity analysis is used to examine correlations, regressions, and
sample comparisons to identify the input parameters influential on the concentrations of infiltrated
PM2.5 during the heating season. The advantage of utilising correlation coefficient tests lies in
their ability to assess the significance of input variables through the associated p-values. The result
of the sensitivity analysis shows the top five input parameters influencing infiltrated PM2.5
concentrations: (1) variability in building envelope airtightness Q50, (2) zone infiltration air
change rates (ACHINF), (3) indoor-outdoor temperature difference (∆T), (4) wind speed (v), and
(5) the area of exposed façade to zone volume ratio (Aef:Vz).
To allow for rapid assessments of the Q50 factor on the concentrations of infiltrated PM2.5 of
existing or proposed UoS buildings, metamodels for the heating season were further developed.
Informed by the latest literature, the five input parameters were examined systematically in three
machine learning (ML) regression algorithms: Generalised Additive Models (GAM), Random
Forest (RF), and Extreme Gradient Decision Trees (XGB). In terms of the best model
performance among the three, the XGB metamodel achieves an R2 value higher than 0.91 for the
heating season concentrations of infiltrated PM2.5 on the training (N =1,910), testing (N =819),
and evaluation (N=40) datasets with a model prediction accuracy greater than 90%.
As a test case, population exposures to indoor PM2.5 in a selected UoS building were estimated by
a microenvironment modelling approach to evaluate- the effects of changing the airtightness of
the building envelope. To directly compare the indoor concentrations with the World Health
Organisation's annual exposure limit of 10 µg/m3, the concentrations of indoor PM2.5 predicted
by the metamodel due to infiltration are combined with the simulated non-heating season
concentrations (May-October). The findings reveal that population exposure to indoor PM2.5
originating from outdoor sources experiences an 11% and 32% reduction when the Q50 values for
the buildings are set at 7 and 3 (m3/h/m2), respectively.
The thesis contributes to the existing knowledge by: (i) developing a novel modelling framework
for assessing indoor air quality (IAQ) of HEI buildings at an institutional level by combining
physics-based modelling and ML-based metamodelling; (ii) identifying the most influential input
parameters impacting the population's exposure to infiltrated PM2.5 in a given HEI context, and
(iii) demonstrating how an HEI stock IAQ model can be utilised to inform and evaluate the effects
of planning and design interventions (e.g., Q50 modifications) on IAQ.
