Strengths and weaknesses of the WHO urban air pollutant database

The 2018 World Health Organization (WHO) global ambient air quality database is an impressive piece of data compilation. It has particulate matter (PM10) monitoring data for 3,570 cities in 97 countries, and PM2.5 data for 2,628 cities in 81 countries. The database uses PM10 and PM2.5 data from established public air quality monitoring systems, which includes pollutants such as sulphates, nitrates and black carbon. These pollutants can penetrate deep into the lungs and the cardiovascular system posing the greatest risk to human health. Unsurprisingly, the WHO database reports relatively low levels of urban PM pollution in high-income countries in Western Europe, the Americas, the Western Pacific, and Oceania and high levels in low-and middle-income (LMI) countries in Africa, Latin America, and Southeast Asia, and even in high-income countries in Latin America. Lack of funding and inadequate staffing in LMI countries are key barriers to effective air pollution reduction. The WHO database has led people and the media to compare cities and draw inaccurate and misleading conclusions, which occurred with the WHO’s 2016 global urban ambient air quality database. In this paper, we investigate the strengths and weaknesses of the 2018 database with respect to several criteria (e.g. selection of pollutants, completeness, spatial and temporal representativeness, and quality assurance and quality control) and make recommendations for improvements.

Introduction objectives, assessing traffic or industrial impacts, planning, policy development or providing 104 public information -measurements will need to be accurate and reliable if they are to prove 105 useful. Without QA/QC, measured data will not provide a sound basis for the assessment of 106 population health effects of air pollution or for effective air quality management; as a result, 107 any investment of money, time and effort made in monitoring will have been wasted. Proper 108 QA/QC is essential in ensuring the comparability of measurements made at different 109 monitoring sites. QA/QC is therefore a basic tool in ensuring that data within a network of 110 sites are harmonised. 111 112 Spatial representativeness 113 Spatial representativeness relates to the question of where UAQMon is to take place. In cities, 114 monitoring is usually undertaken at selected sites, rather than at points on a grid. Sites should 115 be representative of specific location types covering, for example, characteristic central 116 urban, industrial, residential, population-exposure, commercial or kerbside areas. UAQMon 117 stations may differ from neighbouring urban sites affected by multiple sources. According to 118 European Union (EU) Directive 2008/50/EC, at least two fixed monitoring stations shall be 119 installed for a city with less than 250,000 inhabitants to measure the annual average of 120 atmospheric PM and one more for every 250,000 inhabitants up to 1.5 million inhabitants; for 121 up to 6 million inhabitants the directive recommends 13 sites and for urban areas with more 122 than 6 million inhabitants 15 sites (EU 2008). 123 124 Temporal representativeness 125 The EC Directive 2008/50/EC suggests a minimum data capture of 90 per cent (EU 2008). 126 WHO recommends that 50% of the valid data for the reported period should be available to 127 obtain annual average values, and at least 75% of valid data should be available to obtain 1-128 hour average values from data with a smaller averaging time (WHO 1999). 129 130

Meteorological conditions and topographic features 131
In addition to the four key criteria, we examine the comparability of air pollutant 139 concentration data taken in different years and at different seasons among cities. Some cities 140 generate most of their own air pollution (e.g. from road traffic) and can address the 141 sources, while others are downwind from industrial areas or other external sources they 142 cannot control. We look at the comparability of cities with different transboundary pollution 143 regimes. 144 145 Monitoring methods used for pollutants in one city may differ from those in other cities, 146 requiring adjustments to make the data comparable. Analysis of the data may also vary; some 147 cities may eliminate outliers (very high or low values), while others include all data 148 readings. Finally, we address the issue of pollutant selection in the WHO database and the 149 conversion of PM 2.5 (PM 10 ) to PM 10 (PM 2.5 ) and if monitoring of only one of these particle 150 ranges is monitored. Usually only few air pollutants are chosen for any given area 151 considering their potential for adverse effects on human health, animals, natural vegetation, 152 agricultural crops or the ecosystem. In general, it is necessary to first focus on those 153 pollutants, for which air quality standards and guidelines. 154 155

Strengths of the WHO databases 156
The main strength of the 2016 and 2018 WHO air quality databases is that they attempt to 157 provide a global overview of PM pollution. It compiles PM mass concentration data from 158 over 4,300 cities globally, with most data from developed countries. Less

Limitations noted by WHO 183
In the 2016 and 2018 databases data from sites close to sources such as industries, power plants, 184 highways, and urban kerbsides are not included. This is important in developing countries 185 since many people are living near to and exposed to emission from such sites. Cities of 186 inhabitants less than 100,000 are not included although the population may be exposed to 187 emissions from industrial facilities outside the urban area.   A final issue of concern is the use of PM 2.5 /PM 10 conversion factors if only one of the size of 515 PM distribution is monitored. If a local PM 2.5 /PM 10 factor is unknown, the usual approach of 516 the selected conversion factor is often around 0.5 (WHO, 2008). In the WHO 2018 database 518 data conversion from measured PM 2.5 (PM 10 ) to converted PM 10 (PM 2.5 ) is sometimes 519 performed when PM 10 (PM 2.5 ) measurements exist, for example, US and Indian cities. As 520 Table 2 shows, estimated conversion PM 2.5 /PM 10 ratios can and often differ from the 521 monitored ones. 522 particularly in developing-country cities where poor air quality poses a significant 528 threat to human health and wellbeing. Rankings and comparisons that single out the 529 "worst" do not advance this cause but instead confuse people and politicise a public health 530 issue. If we are to save lives now and protect future generations, we need to be more 531 thoughtful and precise when we talk about urban air quality. 532

533
The WHO air quality databases are attempts to gain an overview of the state of air quality in 534 cities around the world. This is important to raise awareness, measure progress and to inspire 535 action. However, compiling database of comparable air quality from cities is not without 536 challenges as demonstrated here. These include monitoring versus. non-monitoring, 537 representativeness, data coverage, background pollution, meteorological condition, seasonal 538 monitoring, issues of different monitoring methodologies, and QA/QC. The Notes in the 539 WHO global air quality databases should clearly advise the users, in particular the media, 540 against comparing and ranking cities as this is misleading. There are also political 541 consequences: if city officials fear being "named and shamed", they have a strong incentive 542 to hide their data or under-report pollution. The controversy over Beijing's air quality 543 data highlights these risks (Guardian, 2014)  In order to improve future WHO databases and have comprehensive and reliable data, it is 549 recommended that data collection be accompanied by a rigorous review of the existing 550 literature; that annual means for NO 2 and O 3 are also compiled as was the case in the WHO 551 Healthy Cities AMIS database; that data providers are requested to answer a certain set of 552 questions with respect to QA/QC of their data such as following rigorously a detailed QA/QC 553 plan. This would ensure the comparability of measurements, assessment of the accuracy and 554 precision of data and that monitoring results meet defined standards with a stated level of 555 confidence. Finally, a strong warning should be given to data users including the media to not 556 abuse the data to name and shame the "most polluted city".