Air pollution is a growing concern for human health, biodiversity and natural environment in large urban areas. It is, therefore, vital to monitor and model air quality (AQ) in urban areas to understand its spatiotemporal variabilities and its main drivers. Traditionally it was not possible to develop high-resolution AQ maps in urban areas due to sparse reference network. However, since the emergence of low-cost sensors (LCS), it has become possible to structure a dense network of sensors and develop high-resolution AQ maps. This is what this PhD project intends to achieve by: (a) analysing the suitability of LCS for urban AQ monitoring and how their measurements can be further improvement using advance calibration techniques, (b) deploying a dense network of AQ sensors based on multiple criteria and using sensors of different grades, (c) employing various AQ modelling and mapping techniques including geostatistical interpolations, land-use regression (LUR) and dispersion modelling, and (d) using data fusion approaches to fuse measured and estimated pollutant concentrations. A multi-criteria Air Quality Monitoring Network (AQMN) was structured based on economic, social and environmental indicators. The network was made of several layers of sensors including reference sensors, LCS (e.g., AQMesh pods and Envirowatch E-MOTEs) and IoT (internet of things) sensors. The data from the designed AQMN was used in AQ mapping, models validations, analysing spatiotemporal variability of pollutants and sensor calibration. Reference sensors were used as standard to calibrate measurements of the LCS employing multiple linear regression and generalised additive model. LUR models were developed for the first time in Sheffield using several land-use and emission related variables. In contrast to previous studies that mostly used linear techniques, here nonlinear regression approaches were also used for developing LUR models, which outperformed the linear counterparts. LUR models were trained and validated using annual average NO2 concentrations from diffusion tubes as well as from LCS. The models were cross-validated by comparing estimated and measured concentrations. LUR model demonstrated that among predictor variables altitude had negative significant effect, whereas major roads, minor roads and commercial areas had positive significant effect on NO2 concentrations. Furthermore, an Airviro dispersion model was developed and several emission scenarios were tested, which showed that NOx concentrations were mainly controlled by road traffic, whereas PM10 concentrations were controlled by point sources. To further improve the AQ maps, modelled and measured concentrations were fused (integrated) to produce high-resolution maps in Sheffield using data fusion technique known as Universal Kriging, which estimated realistic (based on priori expectations) NO2 concentration maps that inherited spatial patterns of the pollutant from the model estimations and adjusted the modelled values against the measured concentrations. The methodology was successful in demonstrating the spatial variability and highlighting the hotspots of NO2 concentrations in Sheffield. The main findings of the project are: (a) The project proposed a nonlinear generalised additive model for low-cost sensors calibrations in outdoor environment. Low-cost sensors are a cheaper source of AQ data, however, they require robust outfield calibrations. (b) A formal approach was proposed for structuring an AQMN in urban areas, which was based on multi-criteria. (c) It was shown that LUR model based on nonlinear machine learning approach outperformed the dispersion modelling approach. (d) Data fusion techniques (such as Universal krigging) were employed to integrate model estimations with measured concentrations. Such data fusion approaches are useful tools for improving data quality and producing high-resolution AQ maps. (e) Time series modelling ARIMA with exogenous variables (ARIMAX) outperformed other linear and nonlinear time series models, and is proposed as an early warning tool for predicting potential pollution episodes in order to be proactive in adopting precautionary measures. Limited data was available on particulate matter, especially on fine and ultrafine particulates, therefore, further work is required on particulate matter monitoring, modelling and management in urban areas.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)