Overview

Project Highlights:

  • You will join a large cross-university multidisciplinary research network working on air quality observations and modelling using novel instrumentations, mobile vans and low cost sensors, and air quality, health impact assessment and economic models, as part of a NERC funded major research project – West Midlands Air Quality Improvement Programme (£5m)
  • Air quality big data from the monitoring network and low cost sensors will be analysed by using machine learning techniques to model and visualize the pollution distribution, human exposure and to assess the effectiveness (and health benefit) of air quality control measures
  • Hand-on training will be provided to the candidate to develop skills in novel machine learning techniques and economic modelling, which will benefit future career of candidate whether in academia, industry or governmental departments

Air pollution is the single largest risk to human health, contributing to more deaths (7 million) than all other environmental risks combined (Landrigan et al., 2017). In the UK, air pollution causes up to 36,000 deaths a year premature deaths and costs the economy £20 b per year.

Understanding the spatial/temporal distribution of air pollutants is essential to estimate the human exposure and health effects. The number of air quality monitoring stations (AURN) is very limited, e.g., only 5 stations in Birmingham. Using the AURN data to estimate human exposure bears large uncertainty. Satellite can provide an estimate of ground level concentrations but again with large uncertainty. Recently machine learning algorithms are used to combine low cost sensor network and satellite data with ground-based monitoring data to provide a more accurate modelling of spatial and temporal distribution of air pollutants (Zhan et al., 2017, 2018).

Machine learning algorithms can also be used to decouple the effects of meteorology from observed air pollutant concentrations (Figure 1, Vu et al., 2019), which reflect the real trend in air quality. This information can then be used to evaluate the effectiveness of the air pollution interventions (such as clean air zone). We recently showed that a machine-learning based random forest algorithms has a superior performance than traditional statistical and air quality modelling (Vu et al., 2019) and offers an independent method.

The aim of this project is to evaluate the impact of clean air actions on air quality, health and economy. This will assist local authorities and the government to design future air pollution control strategies.

Observed monthly average and weather normalized (detrend) PM2.5 in Beijing from 2013 to 2017 using a random forest algorithm. The observed values are a result of emission and meteorology. Under a constant emission, meteorological difference can lead to order of magnitude differences in PM2.5 mass concentration. The detrend PM2.5 removes the meteorological effects, which enable us to see the decreasing trend in PM2.5 as a result of clean air actions (Vu et al., 2019).

Methodology

 

  1. Data source: long-term and real-time data from DEFRA; low cost sensor data from NERC funded WM-Air (£5m) meteorological data; aerosol optical depth from satellite instruments
  2. Methods for quantifying the real trend in concentrations of air pollutants in Birmingham (before and after the implementation of clean air zone) using a random forest algorithm (Vu et al., 2019). This will allow the real (weather normalized) trend to be obtained so that the real impact of emission change assessed.
  3. A spatially explicit machine learning technique will be used to provide a spatial distribution of key air pollutants such as NO2 and PM2.5 in Birmingham incorporating automatic monitoring data, low cost sensor network, and aerosol optical depth.
  4. Health burdens and economic cost due to air pollution before and after the implementation of the clean air zone in Birmingham will be used to evaluate its impact on health and economy

 

Training and Skills

Student will be trained to use R programme, the commonly used data analysis software package such as OpenAir, and economic models based on regression discontinuity design (Fu and Gu, 2017). Hand-on training by research staff and students will also be provided to the candidate to adapt or develop codes based on machine learning techniques to analyse the air quality data across networks and from low cost sensors.

Specific training on literature reading and review and scientific writing will be provided on a regular basis during weekly supervisory meetings.

Timeline

Year 1: Learn to use R and machine learning techniques (such as random forest algorithms), download air quality, meteorology, and satellite aerosol optical depth data from DEFRA, obtain low cost sensor network data from WM-Air; Apply the machine learning algorithms to visualize the spatial and temporal distribution of PM2.5 and NO2 in Birmingham; write up for publication

Year 2: Overlay the spatial and temporal distribution of PM2.5 and NO2 with population density to estimate the human exposure to PM2.5 and NO2 in Birmingham and associated health effects; write up for publication

Year 3: Apply a random forest algorithm to decouple of meteorological effect on air quality to derive the real trend in air quality; evaluate the effectiveness of clean air zone; Apply an economic model to estimate the monetary benefit and cost of the Birmingham clean air zone; write up for publication and for thesis

Year 4: write up for publication and for thesis

 

Partners and collaboration (including CASE)

This is jointly supervised by a senior air quality officer the Birmingham City Council. We are well engaged in research within the WM-Air project including deploying the low cost PM2.5 sensors and to evaluate the effectiveness of clean air zone. This project will apply a novel method that has been successfully used in Beijing (Vu et al., 2019) to Birmingham to support policy processes.

Further Details

Contact:

Dr. Zongbo SHI

School of Geography Earth and Environmental Sciences

University of Birmingham

Birmingham

B15 2TT

United Kingdom

Email: z.shi@bham.ac.uk

Telephone: 01214149128

Webpage: https://www.birmingham.ac.uk/staff/profiles/gees/shi-zongbo.aspx