Low-cost NO₂ monitoring and predictions of urban exposure using universal kriging and land-use regression modelling in Mysore, India

Highlights

• Air pollution exposure estimation is difficult in LMICs such as India.

• Utilized available predictor data and randomized pollution samples to interpolate NO₂ in Mysore.

• Demonstrated utility of LUR and kriging, feasibility of dense monitoring.

• Showed elevated ambient NO₂ levels in the highly populated city center.

• Discussed limitations of spatial methods in this resource-constrained environment.

Abstract

In Low- and Middle-Income Countries, rapid urbanization has led to poorer air quality, yet pollution monitoring networks are often sparse or non-existent. Few previous studies have sought to understand the unique predictors of air pollution exposure in Indian urban environments. Our study monitored and modeled nitrogen dioxide (NO₂) in Mysore, a rapidly urbanizing city in India. NO₂ sampling was conducted in four seasonal campaigns (each lasting 2 weeks) in 2016–2017, at 150 sites throughout Mysore. Seasonal spatial interpolation of NO₂ levels was conducted using 2 distinct models, the first utilizing a land use regression (LUR) approach and the second using universal kriging methods. Model performance was determined using adjusted R², and validated using leave-one-out cross validation. Measured NO₂ concentrations ranged from 0.3 to 51.9 ppb across the four seasons of the study period, with higher concentrations in the center of the city. In the LUR model (R² = 0.535), proximity to major roads, point sources of pollution such as industrial sites and religious points of interest (PoI), land uses with high human activity, and high population density were associated with higher levels of NO₂. Proximity to minor roads and coverage of land uses characterized by low human activity were inversely associated with air pollution. Cross-validation of results confirmed the reliability of each model. Few studies have applied spatially heterogeneous sampling to assess ambient air pollution levels in India. The combination of passive NO₂ sampling and LUR/kriging modeling methods allowed for characterization of NO₂ patterns in Mysore. While previous work indicates traffic pollution as a major contributor to ambient air pollution levels in urbanizing centers in Asia, our results indicate the influence of other pollution factors (e.g., point sources), as well as highly localized characteristics of the urban environment (e.g., proximity to religious points of interest) in urban India. Areas of Mysore consistently experienced pollution in excess of World Health Organization (WHO) health-protective guidelines for NO₂.

Introduction

Air pollution exposure studies and assessments of the corresponding health outcomes are needed in Low- and Middle-Income Countries (LMICs) worldwide, but especially in heavily populated nations that are undergoing unprecedented urbanization (Allender et al., 2010). According to the World Health Organization (WHO), 98% of cities in low- and middle-income countries with more than 100,000 inhabitants do not meet WHO health-based air quality guidelines, as compared with 56% of similar cities in high-income countries (World Health Organization, 2019a). Rapid growth of population in cities, development of heavy industries, and intensification of road traffic present significant challenges to ambient air quality in urban areas of LMICs. In addition, many basic needs such as cooking, household heating, and transit take place in these urban areas without availability of clean fuels or alternative technologies, resulting in poor air quality (Krzyzanowsk et al., 1999).

India represents a LMIC setting that experiences several of these challenges to urban air quality. Air pollution likely differs in Indian cities compared with Western cities due to variations in traffic pollution, which can be caused by fuel differences, technology differences, the types of vehicles on the roads, and the presence of heavily trafficked roadways (O'Neill et al., 2003). The results of several studies further agree that in future, the demand for transportation will exceed improvements to emissions reduction technologies (Vasconcellos and Urban Transport, 2013; Delucchi, 2000). Despite regulatory interventions, higher exposure to traffic pollution with distinct within-city gradients around major roads and highways has been shown in studies of US and European urban areas, suggesting higher exposure differentials within cities than between cities (Zhu et al., 2002; Gilbert et al., 2005). Further, the presence of heavy industry within urban areas of India is likely to impact air quality and the heterogeneity of urban air pollution (Khillare et al., 2004; Reddy et al., 1993). Additionally, the spatial gradient of air pollution may vary due to multiple industrial point sources of pollution. Compared to Western metropolitan areas, cities in India likely have diverse built structures, topography, weather and land use patterns that may influence the spatial distribution of air pollution. Actual exposure may also vary through distinctions in lifestyle and culture that affect indoor/outdoor activity patterns, as well as housing characteristics.

Understanding the variation of pollution that occurs within urban environments is important to improve the estimation of air pollution exposure, and thereby improve our knowledge of the health effects of air pollution. Many methods of estimating air pollution concentrations (including atmospheric dispersion models and satellite data-based models) are very data intensive, and the input data that are required to successfully build and implement those models may not be readily available in low-resource settings (Jerrett et al., 2005). Additionally, current satellite-based data on air pollution at the surface level for India are not highly spatially resolved, and may lead to exposure misclassification by missing the fine-scale changes or seasonal patterns in the distribution of air pollution exposure in Indian cities (Somvanshi et al., 2019). Studies using simpler methods of exposure, such as distance to the nearest major roadway or the data collected by a local regulatory monitor as a proxy of exposure at the individual's location, have potential for exposure misclassification and do not incorporate other variables that may influence exposure, such as local point sources, small industry, and highly trafficked markets (Jerrett et al., 2005).

The development of models to assess air pollution exposures within cities for health studies has been identified as a priority area for future research (Brunekreef and Holgate, 2002; Brauer et al., 2003). Locally tailored approaches for estimating heterogeneous air pollution exposures within LMICs are needed, as methods used for studies in Western countries with denser monitoring networks may not be appropriate or feasible. Multiple methods exist to address intraurban variation of air pollutants. Traditionally, urban studies estimate air pollution exposure by assigning values from the nearest available monitor or averaging values from multiple nearby monitors, an approach that uses readily available data from regulatory monitors and provides large spatial and temporal coverage (Ritz et al., 2000; Marshall et al., 2008; Bell, 2006). This approach either requires a large network of monitors, and/or relies on the assumption that the pollutant of interest is homogeneous within urban areas. However, such extensive monitoring networks may not exist in all areas of interest in India, and previous studies (Gurung et al., 2017; Weissert et al., 2018) suggest that there may be spatial heterogeneity within small areas in urban cities of developed countries as well as LMICs such as India, leading to high potential for exposure misclassification using traditional methods.

Alternative approaches have made use of low-cost, spatially dense monitoring networks for collection of field samples of health relevant gaseous pollutants (such as nitrogen dioxide, NO₂), as well as spatial analysis tools such as Geographical Information Systems (GIS), to estimate exposures to pollutants at high spatial resolution using multiple regression models (Wheeler et al., 2008), interpolation (e.g. kriging (Mulholland et al., 1998; Jerrett et al., 2001; Pikhart et al., 2001; Finkelstein et al., 2003; Mercer et al., 2011; Young et al., 2014; Adam-Poupart et al., 2014), inverse distance weighting (Marshall et al., 2008; Abbey et al., 1999)), dispersion modeling (Panasevich et al., 2009), and LUR (Gurung et al., 2017; Jerrett et al., 2005; Finkelstein et al., 2003; Mercer et al., 2011; Young et al., 2014) models.

The objectives of this study were to characterize seasonal intra-urban variation of NO₂ in Mysore, Karnataka State, India, in order to facilitate our study of the long-term respiratory health effects of NO₂ exposure in the city. We collected seasonal field samples of NO₂ utilizing multiple sampling technologies. We applied LUR modeling and universal kriging to the collected field data to build seasonal and annual average models of NO₂ air pollution levels throughout the city, with a focus on explaining the influence of important covariates on the spatial relationships in the levels of NO₂ in the city. The methods were designed to address issues central to India and similar LMICs, including limited data availability and high intra-urban spatial variability of air pollution. NO₂ was selected as the pollutant of interest given the potential for adverse health outcomes associated with short-term exposure to NO₂, including airway inflammation in healthy people (Blomberg et al., 1997) and increased respiratory symptoms in people with underlying health conditions (Weinmayr et al., 2010). Exposure to NO₂ at high concentrations is of particular concern among sensitive populations, including children, the elderly, and people with chronic respiratory ailments (Weinmayr et al., 2010; Annesi-Maesano et al., 2003). Additionally, NO₂ has high feasibility and reliability of field measurements, as well as relatively low cost of equipment and sampling.