Environmental Research 214 (2022) 113932 Contents lists available at ScienceDirect Environmental Research journal homepage: www.elsevier.com/locate/envres Spatial modelling and inequalities of environmental noise in Accra, Ghana Sierra N. Clark a,b, Abosede S. Alli c, Majid Ezzati a,b,d,e, Michael Brauer f, Mireille B. Toledano a,b,g, James Nimo h, Josephine Bedford Moses h, Solomon Baah h, Allison Hughes h, Alicia Cavanaugh i, Samuel Agyei-Mensah j, George Owusu k, Brian Robinson i, Jill Baumgartner l,m, James E. Bennett a,b,1,**, Raphael E. Arku c,1,* a Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, London, UK b MRC Centre for Environment and Health, School of Public Health, Imperial College London, London, UK c Department of Environmental Health Sciences, School of Public Health and Health Sciences, University of Massachusetts, Amherst, USA d Regional Institute for Population Studies, University of Ghana, Accra, Ghana e Abdul Latif Jameel Institute for Disease and Emergency Analytics, Imperial College London, London, UK f School of Population and Public Health, The University of British Columbia, Vancouver, Canada g Mohn Centre for Children’s Health and Wellbeing, School of Public Health, Imperial College London, London, UK h Department of Physics, University of Ghana, Accra, Ghana i Department of Geography, McGill University, Montreal, Canada j Department of Geography and Resource Development, University of Ghana, Accra, Ghana k Institute of Statistical, Social & Economic Research, University of Ghana, Accra, Ghana l Institute for Health and Social Policy, McGill University, Montreal, Canada m Department of Epidemiology, Biostatistics, and Occupational Health, McGill University, Montreal, Canada A R T I C L E I N F O A B S T R A C T Keywords: Noise pollution is a growing environmental health concern in rapidly urbanizing sub-Saharan African (SSA) Environmental noise cities. However, limited city-wide data constitutes a major barrier to investigating health impacts as well as Land use regression implementing environmental policy in this growing population. As such, in this first of its kind study in West Socioeconomic status Africa, we measured, modelled and predicted environmental noise across the Greater Accra Metropolitan Area Road-traffic noise Intermittency ratio (GAMA) in Ghana, and evaluated inequalities in exposures by socioeconomic factors. Specifically, we measured Sub-Saharan Africa environmental noise at 146 locations with weekly (n = 136 locations) and yearlong monitoring (n = 10 loca- tions). We combined these data with geospatial and meteorological predictor variables to develop high- resolution land use regression (LUR) models to predict annual average noise levels (LAeq24hr, Lden, Lday, Lnight). The final LUR models were selected with a forward stepwise procedure and performance was evaluated with cross-validation. We spatially joined model predictions with national census data to estimate population levels of, and potential socioeconomic inequalities in, noise levels at the census enumeration-area level. Variables representing road-traffic and vegetation explained the most variation in noise levels at each site. Predicted day- evening-night (Lden) noise levels were highest in the city-center (Accra Metropolis) (median: 64.0 dBA) and near major roads (median: 68.5 dBA). In the Accra Metropolis, almost the entire population lived in areas where predicted Lden and night-time noise (Lnight) surpassed World Health Organization guidelines for road-traffic noise (Lden <53; and Lnight <45). The poorest areas in Accra also had significantly higher median Lden and Lnight compared with the wealthiest ones, with a difference of ~5 dBA. The models can support environmental epidemiological studies, burden of disease assessments, and policies and interventions that address underlying causes of noise exposure inequalities within Accra. * Corresponding author. School of Public Health and Health Sciences, University of Massachusetts, Amherst, MA, USA. ** Corresponding author. Department of Epidemiology and Biostatistics, Imperial College London, London, UK. E-mail addresses: umahx99@imperial.ac.uk (J.E. Bennett), rarku@umass.edu (R.E. Arku). 1 Joint senior authorship. https://doi.org/10.1016/j.envres.2022.113932 Received 4 April 2022; Received in revised form 20 June 2022; Accepted 16 July 2022 Available online 20 July 2022 0013-9351/© 2022 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 1. Introduction increasingly available globally; such as satellite derived land use mea- sures (Larkin et al., 2017) or locations of road networks and human Noise from anthropogenic activities is pervasive in urban settings activities from OpenStreetMap (Barrington-Leigh and Millard-Ball, and can have adverse effects on human health and wellbeing (European 2017). In urban SSA settings, where sources of noise are complex, LUR Environment Agency, 2020; Hammer et al., 2014; Kang, 2017a; World modelling is a cost-effective and attractive method for estimating noise, Health Organization, 2018). Epidemiolocal studies from cities in Europe given that emissions (e.g., time-resolved traffic flows) and building and North America have shown that exposure to noise from road-, rail- canyon (e.g., building footprints and heights) data needed for and aircraft traffic sources can lead to a range health effects, including propagation-based modelling are often not freely available or do not impacts on annoyance, sleep quality, cardiometabolic diseases, and exist at all (Sieber et al., 2017). impaired cognitive function (Basner and McGuire, 2018; Guski et al., To bridge the data and modelling gap of environmental noise in SSA 2017; Münzel et al., 2021; Thompson et al., 2022; van Kempen et al., cities and provide local data for policy formulation and environmental 2018; Vienneau et al., 2019). Modelling and mapping the spread of health assessments, we designed a LUR modelling study to predict and environmental noise, mostly in high-income cities, has revealed highly map spatial variations and inequalities in noise metrics within one of the unequal distributions across and within cities, sometimes patterned by largest and fastest growing metropolises in Africa. The models inte- socioeconomic gradients (Dale et al., 2015; Dreger et al., 2019; Euro- grated noise data from a 1-year large-scale measurement campaign pean Environment Agency, 2020). Within-city inequalities in noise within the Greater Accra Metropolitan Area (GAMA) (Clark et al., 2020; exposure could also create and/or exacerbate existing health Clark et al., 2021) with a suite of city-wide geospatial and meteoro- inequalities. logical data. The final models were used to predict long-term (annual) Cities in sub-Saharan Africa (SSA), home to some of the world’s averages of noise levels representing different periods of the day fastest-growing economies, are undergoing significant expansion and (LAeq24hr, Lden, Lday, Lnight) across the city. We also estimated census economic transformations. Growing SSA cities are now characterized by enumeration area population exposures and socioeconomic inequalities glaring urban transport problems, including traffic congestion, long in noise levels within the Accra Metropolis (~2 million people), the commute times, and traffic related noise pollution (Amegah and urban core of the GAMA. In a secondary analysis, we built regression Agyei-Mensah, 2017; Imoro Musah et al., 2020; Sietchiping et al., 2012). models to explore whether the intermittency of the sound environment, Traffic noise coexists with community/neighbor noise, such as loud, represented by the intermittency ratio metric, was associated with the pervasive music from religious activities and informal/small businesses, features of the environment and other noise metrics at measurement making noise pollution in SSA cities an emerging health concern (Baloye sites. and Palamuleni, 2015; Bediako-Akoto, 2018; Kazeem and Dahir, 2018; Wawa and Mulaku, 2015; Zakpala et al., 2014). Though common in 2. Material and methods European, North American, and increasingly in Asian cities (Aguilera et al., 2015; Council of the European Union, 2002; European Environ- 2.1. Study location ment Agency, 2020; Liu et al., 2020; Walker et al., 2017; Wang et al., 2016; Xie et al., 2011) modelling and mapping of environmental noise to Our study was conducted in the GAMA (~5 million people), the most reveal levels and spatial variations are severely lacking within the SSA densely populated area in Ghana. Accounting for over a fifth of the context. Thus, hindering local efforts to identify sources of noise, country’s urban population, the region includes Accra Metropolis as its investigate health impacts, quantify burdens of disease, and design core (estimated population in 2010 and 2019: ~1.66 million and ~2 policies and interventions to mitigate noise and reduce inequalities in million) (Ghana Statistical Service, 2019) and the port city of Tema. The exposures. Furthermore, a major barrier for conducting a global burden GAMA is the political, economic, and administrative capital of Ghana, of disease assessment due to environmental noise is a lack of exposure and while these sectors drive urban economic growth, vast inequalities data in low- and middle-income countries, and thus generating estimates in income, housing and environmental quality remain (Annim et al., in these regions can contribute to that global effort. 2012; Dionisio et al., 2010; Fobil et al., 2010). As the population and Propagation models, which are based on mathematical description of city-limits have expanded over the years, demand for transportation has emissions and transmissions of sound through the environment, have increased, with private vehicles or privately owned minibuses (known been widely used for modelling noise from road-, rail-, and aircraft- locally as trotro) as the main means of getting around (Imoro Musah traffic sources, particularly in European cities (Garg and Maji, 2014; et al., 2020). There is no train or tram services, and formal transit bus Khan et al., 2018). However, a challenge for implementing propagation services are limited. Ride-shares such as Uber and Bolt, and models in many low and middle-income regions of the world has been motorcycle-taxis (‘Okada’), are more recently being used to complement that national governments or even international corporations do not the need for public transport (Acheampong et al., 2020). Noise pollution routinely collect much of the data that are needed for the models, such in particular, has been highlighted recently as an environmental health as road-traffic counts, vehicle fleet compositions, or building height and concern in local and international media (Bediako-Akoto, 2018; Kaledzi, footprint information with fine enough spatial or temporal granularity 2018; Kazeem and Dahir, 2018; Knott and Gyamfi Asiedu, 2019). (Aguilera et al., 2015; Kang, 2017b; Sieber et al., 2017). Alternatively, land use regression (LUR) models (Hoek et al., 2008), which are 2.2. Data commonly used for the estimation of spatial variability in air pollution within cities (Hoek et al., 2008), have also increasingly been applied to 2.2.1. Environmental noise measurement and metrics noise in recent years in some high and middle-income country cities Between April 2019 and June 2020, we deployed sound level meters (Aguilera et al., 2015; Alam et al., 2017; Chang et al., 2019; Drudge (SLM) near the roadside at 146 locations, comprising of 136 rotating (7- et al., 2018; Fallah-Shorshani et al., 2018; Harouvi et al., 2018; Liu et al., day) and 10 fixed (~1-year) measurement sites (Fig. 1). The fixed sites 2020; Raess et al., 2021; Ragettli et al., 2016; Walker et al., 2017; Wang represent diverse land use, socioeconomic, and transport features. et al., 2016; Xie et al., 2011). Currently, only one environmental noise Rotating sites were selected through stratified random sampling based LUR model has been developed in SSA, for informal settlements in South on land use features and population data (World Bank, 2014). The Africa (Sieber et al., 2017). A noise LUR model derives statistical re- measurement campaign was briefly interrupted in April–May 2020 lationships between measured noise metrics and predictor variables that during Accra’s COVID-19 pandemic lockdown and subsequent represent a range of factors in the urban environment that are associated COVID-related stoppages. The noise measurements have been described with the emission, propagation and attenuation of noise. Geospatial and in detail in our previously published protocol paper (Clark et al., 2020). meteorological predictor data (Hoek et al., 2008; Khan et al., 2018) are We used Noise Sentry SLMs from Convergence Instruments (Québec, 2 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 Fig. 1. Locations of rotating and fixed measurement sites in the Greater Accra Metropolitan Area (GAMA). The GAMA, Accra Metropolis, and Tema boundaries are from the Ghana Statistical Service, road-network and water-body shapefiles are from OpenStreetMap (2019). Canada) to continuously record A-weighted sound levels (decibels 2011). (dBA)) which were integrated and logged every minute. The Noise Sentry is rugged in design, built to withstand high temperatures, and the 2.2.2. Predictor variables digital MEMs microphone is protected against water and dust, which is We collected and collated spatial and temporal predictor variables necessary for a setting like Accra. We deployed the SLMs in weather that reflected factors in the urban environment associated with the protective custom designed enclosures which we attached to poles or emission, propagation, and attenuation of sound. Details of each pre- trees near the roadside at ~ 4 m (±1 m) above ground, and at least 2 m dictor variable and its source are included in Table 1. To capture land away from the nearest façade. We undertook quality assurance and use/land cover, we used a raster dataset at 20 m resolution that mapped control (QA/QC) tests of SLM accuracy and precision throughout the four land cover classes across the GAMA from Spot 5 imagery attributed campaign (Clark et al., 2020; Clark et al., 2021), which showed good to the year 2014 (World Bank, 2014). To characterize vegetation, we agreement between the Noise Sentry SLMs and with a higher cost Type 1 calculated the Normalized Difference Vegetation Index (NDVI) from the SLM (Cirrus Optimus Red). Further details on the SLMs, data collection spectral signatures of green vegetation from 30 m resolution satellite protocol, and QA/QC practices undertaken throughout the measure- imagery. We obtained a Landsat 8 satellite product held on the U.S. ment campaign are described in the protocol paper (Clark et al., 2020). Geological Survey department website attributed to a cloud free day We calculated A-weighted equivalent continuous sound levels (LAeq,T) (cloud cover: 0.02%) in January 2020. January was considered as a for each site and date of measurement. Energy-based long-term average mid-point in the measurement campaign. Other days with Landsat 8 metrics, such as day-evening-night weighted (Lden) and daytime (Lday) imagery were unusable for this purpose due to cloud cover over the area. and night-time (Lnight) noise levels are the mostly commonly used met- However, there was minimal temporal variability of NDVI levels rics in epidemiological studies and are robustly associated with a throughout the year due to Accra’s location near the equator. To esti- number of adverse health outcomes (Basner and McGuire, 2018; Clark mate building density, we made use of a high-resolution spatial dataset et al., 2020; Guski et al., 2017; Thompson et al., 2022; van Kamp et al., of building footprints attributed to the year 2019/2020 from Maxar/- 2020; van Kempen et al., 2018). As well, our previous descriptive study, Ecopia (Price and Hallas, 2019), which we transformed into a dataset of which combined audio recordings with a deep learning acoustic classi- building centroids (spatial center-point). This transformation was done fier, found that road-transportation was a prominent sound source due to the computational intensity of processing building footprints. To identified across measurement sites. Road-transportation sounds were estimate human population density, we used population information particularly dominant in the city center (Accra Metropolis), and in from the most recent Ghana national census summarized within census commercial, business, and industrial areas (Clark et al., 2021). There- enumeration areas (Ghana Statisical Service, 2010). Census enumera- fore, throughout the paper and particularly with reference to Accra tion areas are small geographic units with average population of Metropolis, we refer to the measured and modelled data as environ- 750–800 people and area 0.03–0.04 km2 within the GAMA. To capture mental noise exposures, similar to previous noise LUR studies (Aguilera road-traffic sources of noise, we used a road-network shapefile from et al., 2015; Harouvi et al., 2018; Liu et al., 2020; Raess et al., 2021; OpenStreetMap (OSM) (OpenStreetMap, 2015) downloaded in 2019. Ragettli et al., 2016; Sieber et al., 2017; Wang et al., 2016; Xie et al., OpenStreetMap is an open-source editable global database of urban 3 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 Table 1 geographic information which has grown rapidly over the years Candidate predictor variables for LUR model selection. (OpenStreetMap, 2015). Barrington-Leigh et al. estimated that Open- Variable type Categories Spatial Source (Date) StreetMap had 83% global coverage of roads as of 2016, and 45% calculation coverage in Ghana (Barrington-Leigh and Millard-Ball, 2017). Though, Road-network Major roads; Total length OpenStreetMap we expect the road-network completeness in Accra in 2019 to be higher (Spatial line) secondary/ within buffer (2019) than the estimate for Ghana as Accra is a major capital city which would tertiary roads; (meters); Barrington-Leigh and likely have higher coverage than other smaller cities/rural towns across minor roads; all Euclidean Millard-Ball (2017) the country, hence the lower country-wide average. As well, roads distance and square root Barrington-Leigh et al. analysed data from 2016, and OSM is continually distance to updated and improved overtime by users. To capture aircraft noise, we nearest (meters) obtained the spatial boundaries of the Kotoka International Airport from Airport (Spatial – Euclidean Google Earth (2019) Google Earth. For locations of human activity, we identified the latitude polygon) a distance and and longitude locations of churches, mosques, hospitals, primary and square root distance to secondary schools, restaurants, shopping centers and markets, and nearest (meters) bars/nightclubs from Google Places in 2019. We also obtained locations Land cover Industrial, Area (meters2) World Bank (2014) of bus stations/terminals from Google Places as an indicator of both (raster) business, within buffer 20 m × 20 m human activity and road-transport sources. Finally, we retrieved data on commercial areas; informal elevation above sea level from a digital elevation model (DEM) for Af- high-density rica at 90 m resolution (Verdin, 2017) and data on waterways from OSM residential; (2019). formal Variations in atmospheric conditions can affect acoustic wave residential; propagation (i.e., atmospheric absorption) (Ghinet et al., 2019; Kang, ‘other’ areas (e. g., forest, water, 2017a; Truax, 1999), be sources of sound (e.g., rainfall), or influence grassland, bare human behaviors/activities that result in sound generation (Böcker soil) et al., 2013). Thus, we collected time-resolved data on temperature Locations of Schools, Presence/ Google Places (2019) (Celsius degree), wind speed (m/s), and relative humidity (%) at six of human activity hospitals, bus absence within (Spatial point) stations/ buffer; count the fixed (~yearlong) measurement sites throughout the campaign with terminals, within buffer small weather meters (Kestrel 5500, Nelsen-Kellerman, Pennsylvania, restaurants, bars USA). We also retrieved daily rainfall (mm) data from the Ghana and nightclubs, Meteorological Agency (GMA). churches, mosques, shopping centers 2.2.3. Predictor data pre-processing Normalized – Average value United States We created multiple buffers around the measurement sites which Difference within buffer Geological Survey – were based on the noise LUR literature (Aguilera et al., 2015; Ragettli Vegetation (range: 0–1; Landsat 8 imagery et al., 2016): 50 m, 100 m, 200 m, and 500 m. We then mapped the Index (raster) water (negative − 30 m × 30 m U.S values) was Geological Survey (n. spatial predictor variables to each buffer, centred by the coordinate omitted) d.) location of the measurement site, through spatial overlay. We then Human – Average value Ghana Statisical clipped the spatial predictors so that only the features of the spatial population within buffer Service (2010) predictors overlapping with each buffer remained. We calculated zonal density within (pop/km2) statistics (e.g., average, sum, area) within each buffer, depending on the enumeration areas (Spatial spatial predictor variable type (details in Table 1). Additionally, for polygon) distance variables we calculated the Euclidean distance from each Centroid of each – Count within Price and Hallas monitoring site to the nearest major and secondary road and to the building buffer (2019) airport location and applied a square root transformation to capture (Spatial point) Rivers/ Total length OpenStreetMap potential non-linear relationships. – waterways within buffer (2019) (Spatial line) (meters) 2.3. Model building and evaluation Elevation above – U.S Geological Survey sea level DEM (2017) (~90 m) We took a land use regression (LUR) approach to model and predict (raster) Verdin (2017) (meters) long-term average noise levels within the GAMA. Specifically, we con- Height of – – Measurement structed models for LAeq1hr and fit separate models for the day and night monitor off of campaign Clark et al. hours. In accordance with the Ghana Standards Authority, we defined the ground (2020) the day-time as 6:00am–9:59pm (night-time: 10:00 pm - 5:59am) (meter) (Ghana Standards Authority, 2018). We assessed the linearity of the Table includes all candidate predictor variables considered for the model se- relationships between noise levels and the (continuous) predictor vari- lection process. The final models include a subset of these predictors which ables with bivariate scatter plots. We also initially built models which survived the model selection process. assumed (i) linear and (ii) non-linear associations (e.g., splines) between a Aircraft traffic at Kotoka airport (~1.8 million passengers a year) (Ghana predictor and dependent variables and found that the predictive error Airports, 2017) is a fraction of what it is at other large airports in major cities between models was similar (Supplementary Information, Table S1). such as Heathrow in London (~80 million passengers/year) (Heathrow Airport, 2018) or Schiphol in Amsterdam (~71 million passengers/year) (Heathrow Therefore, we opted for the simpler modelling approach, which assumed Airport, 2018). linear relationships, and provided the added benefit of enhanced model interpretability. We additionally incorporated random intercepts for hour of the day to account for diurnal correlation of measured sound levels as well as random intercepts for site locations to account for any site-specific unmeasured variations. Our model selection process was aimed at identifying parsimonious 4 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 and generalizable models that also maximized predictive accuracy. We between 22:00–5:59. We restricted predictions to areas which repre- employed a two-step approach where we first chose the buffer radii for sented the measurement sites so that we did not predict out of sample. each predictor variable that had the highest correlation with the noise Thus, we excluded areas that were covered by waterbodies, and/or areas levels in each model. Consistent with other LUR modelling studies, we that were fully grassland/forest (i.e., did not contain any roads). also considered the direction of the association with our a priori as- sumptions (Aguilera et al., 2015; Fallah-Shorshani et al., 2018; Lee et al., 2.5. Population exposure to noise levels in Accra Metropolis 2017; Ragettli et al., 2016; Sieber et al., 2017). Second, we used a stepwise forward model selection process to identify models with a We estimated the percentage of the population exposed to different reduced set of spatial predictor variables (Aguilera et al., 2015; Raess levels of noise in Accra Metropolis. We first spatially overlayed the et al., 2021; Sieber et al., 2017). We began by inserting predictors which predicted noise level surfaces onto a map of enumeration areas from the had the strongest bivariate associations with the noise levels (identified 2010 census in Accra Metropolis (Fig. S1). Enumeration areas reflect the from the first step) and the process was stopped when the coefficient of location of residence at the time of the census and the smallest spatial determination (R2) was no longer improved by at least 1% (Aguilera administrative unit in Ghana. We then calculated average noise levels et al., 2015; Chang et al., 2019). We considered removing predictor within each enumeration area and estimated the number of people variables if their 95% confidence interval around the slope coefficients exposed to varying noise levels of 5 dBA increments (e.g., 50–54 dBA, crossed zero; specifically, if the magnitude of the coefficient and the 55–59 dBA, etc.) based on the population distribution in 2010. width of the confidence interval were large, we considered the estimate to be unstable and dropped the variable from the final model. The final 2.6. Socioeconomic inequalities of enumeration area noise levels in Accra models were then challenged by adding all excluded variables with their Metropolis best buffer size (i.e., one buffer size per variable type) back into the models one by one to check if an improved model could be found. We We investigated whether noise levels in Accra Metropolis were also assessed whether there was collinearity present among predictor associated with measures of enumeration area socioeconomic status variables (r > 0.8), and if found, the predictor variable that was more (SES). Measures of consumption levels ($) (Deaton, 1992; Deaton and correlated with the noise metric was retained in the model. Zaidi, 2002) and inequalities in consumption may be a cause of, and/or a We evaluated the fit and external predictions of the final models with result of, physical and social environments, opportunities, access, se- cross validation. We ran 10-fold cross-validation holding out data from curity, and empowerment within households and communities. Impor- 10% of random measurement sites (CV10%sites) and leave one site out tantly, consumption is a reflection of what people can afford and can cross validation (LOOCV). From the cross-validations, we calculated the access. Thus, our main measure of area-level SES was median log median absolute errors, mean absolute errors, the mean errors, and the equivalized household consumption (Ghanaian Cedi (GH₵)) within correlation of predicted and observed values (r and r2). Furthermore, we each enumeration area. Household consumption was first derived from evaluated whether model assumptions were upheld using diagnostic total annual real household expenditures and rent captured within the plots to see whether the residuals were normally distributed and had 2012–2013 Ghana Living Standards Survey (GLSS) (93% response rate) random and constant variance. We also checked for any temporal and (Expenditures: food, beverages and alcohol, tobacco, clothing and spatial patterns in the residuals. We checked spatial patterns by visu- footwear, housing, electricity, water, gas and other fuels, furnishings, alizing the residuals in variogram plots and calculated the Moran’s I equipment, routine maintenance, health, transport, communications, statistic of spatial autocorrelation. Finally, we evaluated potential recreation and culture, education, hotels, cafes and restaurants, and multicollinearity in the final models as a whole using variance inflation miscellaneous goods and services). To then estimate median household factors (VIF). consumption within all enumeration areas, we combined the GLSS with the 100% sample of the most recent census (2010) in small area esti- 2.3.1. Sensitivity analyses mation models (Corral et al., 2020; Elbers et al., 2003) which derived Since the primary aim of this analysis was to predict annual average relationships between consumption, area and other factors such as asset noise levels across the GAMA, we could not use the weather data for ownership, education, employment, housing quality, and spatial prediction as we only collected it at six sites in the city. However, socio-demographics. It is worth noting that the datasets used to predict we conducted sensitivity analyses to estimate the associations of noise SES (i.e., consumption) and noise levels are independent. As secondary levels with time-resolved weather conditions, specifically wind speed, SES measures, we used census data on the number of individuals within temperature, relative humidity, and rainfall, and to investigate whether each enumeration area with post-secondary education (education their inclusion in the main spatial models improved predictive accuracy. measure) and the number of unemployed individuals (unemployment We additionally modelled the final predictor variable sets with Random measure). These measures are aggregates of household level SES, indi- Forest models as a sensitivity analysis for the choice of model infra- vidual education and unemployment, summarized at the enumeration structure. Random Forest models have been shown previously to area-level, and represent proxies for area-level SES measures. improve predictive accuracy over linear regression in a noise LUR study We estimated Pearson correlations between enumeration area SES conducted in Canadian cities (Liu et al., 2020). measures with noise levels and summarized noise distributions across quintiles of the SES measures (5 groups, 20% of enumeration areas in 2.4. Predicted noise level surfaces each group). We further investigated whether differences between groups were statistically significant using difference-in-means tests with We made predictions of annual average noise levels for each hour of a p-value cut-off of <0.05. the day for an ~50 m × 50 m surface of unmeasured locations in the GAMA. Predictions of LAeq1hr were made onto 24 surfaces, each rep- 2.7. Predictors of urban sound intermittency resenting an hour of the day. From the 24, hourly surfaces, the LAeq24hr (the 24-h equivalent continuous sound level), Lden (day-evening-night There is emerging evidence that the degree of noise intermittency, sound level, a descriptor which penalizes 10 dBA for night-time and 5 characterized by the intermittency ratio metric, can be independently dBA for evening noise), Lday (day-time sound level), and Lnight (night- associated with some adverse health outcomes, or can act as an effect time sound level) metrics were logarithmically calculated. Lday was modifier of the relationship between noise levels and health outcomes calculated with respect to day-time hours between 6:00–21:59 and Lnight (Brink et al., 2019b, Brink et al., 2019a; Eze et al., 2017; Foraster et al., between 22:00–5:59. Lden was calculated with respect to day-time hours 2017; Thiesse et al., 2020; Vienneau et al., 2022). Therefore, we con- between 6:00–18:59, evening between 19:00–21:59, and night-time ducted a secondary analysis and examined intermittency ratio metrics at 5 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 each measurement site. The intermittency ratio is defined as the per- (Table 2). LAeq1hr was positively associated with road-traffic predictors centage of sound energy in the total energetic dose that is created from (i.e., length of major roads, length of secondary/tertiary roads) and the distinct sound events that exceed a threshold (Wunderli et al., 2016). presence of restaurants, and negatively associated with variables rep- Following the calculation procedure in Wunderli et al. (2016) (Equation resenting vegetation (NDVI) and formal low/medium residential land in S1), we used a threshold of +3 dBA above the LAeq,T for the time use. The variables which explained the most semi-partial variance in the period to calculate intermittency ratios for the day (IRday; fixed effect component of the LAeq1hr models were NDVI and length of 6:00am–9:59pm) and night-time (IRnight; 10:00pm–5:59am) periods for major roads. each site. We then followed the same model building process as The median absolute errors (MAEs) of the final LAeq1hr models described in Section 2.3 to identify the potential environmental factors ranged from 2.9 to 3.4 dBA with CV10%sites and the correlation of pre- associated with the degree of day and night-time intermittency ratios at dicted and observed values (r) ranged from 0.72 to 0.74 (r2: 0.51 to the measurement sites. For these models, day and night-time noise levels 0.54). The mean error (ME), a measure of bias, was close to zero, indi- (Lday, and Lnight) were additionally incorporated into the modelling cating no systematic under or over prediction (Table 3). Results from process as predictors, and we post-hoc evaluated whether land use LOOCV were very similar. We did not find evidence that model as- classifications assigned to each measurement site by the field team (see sumptions were violated, and model residuals were randomly distrib- (Clark et al., 2021) for details) modified the associations of the predictor uted. Furthermore, Moran’s I for the residuals indicated a tendency variables in the final models through interaction terms. towards spatial randomness (Range of model’s Moran’s I values: − 0.06 Analyses were conducted in R (R version 3.6.3) and some data vi- to 0.05). Variance Inflation Factors in the final models were low, be- sualizations using ArcMap ® software by Esri (Version 10.8). tween 1.0 and 2.0, indicating very low or no correlation among the variables in the final models that could inflate the coefficients. As a 3. Results sensitivity analysis, we used Random Forest models to generate pre- dictions using the final predictor variable sets and found no improve- 3.1. Noise level LUR model performance and predictor variable ment in MAE (Table S2). As an additional sensitivity analysis, we associations included time-resolved weather variables into the final spatial models and found significant associations between weather variables and The final models included between five and six spatial variables LAeq1hr, but no improvement in the overall model predictive accuracy (Table S3). Table 2 3.2. Spatial patterns of noise levels in the greater Accra Metropolitan area Mean associations of noise levels with spatial predictor variables in the final LUR modelsa. Spatial patterns of day and night-time noise levels in the GAMA were Model Predictor Predictor Buffer Coefficient [95% nearly the same, though day-time noise levels were higher by approxi- variables variable unit size (m) confidence mately 7–8 dBA (Fig. S2). Accra Metropolis, the most populated and interval] urbanized area of the GAMA, had some of the highest predicted Lden LAeq1hr for all day-time hours (dBA) (median: 64 dBA), as well as the port city of Tema in the east of GAMA Intercept – – 65.2 [64.2, 66.3] Total length of Standardizedb 100 2.5 [1.2, 3.7] (median: 62 dBA) (Fig. 2). Predicted Lden was highest near major roads major roads (meters) (median: 69 dBA), followed by secondary/tertiary roads (median: 63 Total length of Standardized 200 1.8 [0.9, 2.8] dBA), and then near minor roads (median: 60 dBA) (Table 4). The peri- secondary/ (meters) urban periphery in the north and west of the GAMA had the lowest levels tertiary roads of L (median: 58 dBA) and L (median: 50 dBA) (Fig. 2). Formal low/ Standardized 200 0.8 [-1.5, 0.1] den night − − medium (meters2) residential area 3.3. Population exposures to noise in Accra Metropolis Normalized Standardized 50 − 2.8 [-3.7, − 2.0] difference (value) vegetation index Almost the entire population in the Accra Metropolis lived in Number of Count 100 1.4 [0.5, 2.4] enumeration areas where the average Lden and Lnight exceeded the WHOs restaurants (European) guidelines for road-traffic noise (Lden: 53 dBA; Lnight: 45 dBA) Population Standardized 500 0.8 [-0.4, 2.0] (World Health Organization, 2018) (Fig. 3) and furthermore exceeded density (people/km2) 55 dBA Lden and 50 dBA Lnight. The majority of the population in the LAeq1hr for all night-time hours (dBA) Intercept – – 57.2 [55.2, 59.2] Accra Metropolis lived in enumeration areas with average Lden of 60 to Total length of Standardized 100 3.0 [1.7, 4.4] 64 dBA (31%, 515,873 people) or 65 to 69 dBA (53%, 876,098 people) major roads (meters) and average Lnight of 55 to 59 dBA (54%, 888,181 people) (Table S5, Total length of Standardized 200 2.2 [1.2, 3.3] Fig. S3). With a recent projection of around 2 million people in Accra secondary/ (meters) tertiary roads Metropolis in 2019, we expect the current numbers of people exposed to Formal low/ Standardized 200 − 1.3 [-1.9, − 0.5] be higher than our estimates which are based on the 2010 census. medium (meters2) residential area Table 3 Normalized Standardized 50 − 2.2 [-2.9, − 1.5] difference (value) Final noise level LUR model prediction accuracy from 10-fold 10% random site vegetation index hold-out cross validation (CV10%sites). Number of Count 100 1.8 [0.8, 2.9] Model r r2 Median Mean Mean restaurants absolute error absolute error a Models incorporated random effects for site and hour of the day. Mean as- error sociations of spatial predictor variables were adjusted for monitor height in the LAeq1hr for all day- 0.74 0.54 2.92 dBA 3.60 dBA − 0.34 model. The coefficients of predictor variables in the main models had the same time hours (dBA) dBA direction in bivariate models (Table S4). LAeq1hr for all 0.72 0.51 3.38 dBA 4.01 dBA − 0.41 b Continuous variables were standardized by subtracting the data mean and night-time hours dBA dividing by the data standard deviation. A 1-point change in a standardized (dBA) variable corresponds to a 1 standard deviation increase on the original scale. r2 approximates R2. 6 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 Fig. 2. Predicted noise levels in the Greater Accra Metropolitan Area. Predictions were made for a fixed height of 4 m off the ground onto an ~50 m × 50 m grid of the city and calculated from the 24 surfaces of long-term hourly averages. Grey areas on the map represent areas excluded from prediction as they are out of sample (e.g., water bodies, forest/grassland). Legend: LAeq24hr (dBA): 24-h equivalent continuous A-weighted noise level; Lden (dBA): Day-evening-night equivalent continuous A-weighted noise level. Lden was calculated with respect to day-time: 6am-6pm (13 h); evening: 7pm–9pm (3 h); night-time: 10pm-5am (8 h); Lday (dBA): Day-time equivalent continuous A-weighted noise level (6am-9:59 pm); Lnight (dBA): Night-time equivalent continuous A-weighted noise level (10:00pm-5:59 am). Table 4 Predicted noise levels in the Greater Accra Metropolitan Area (GAMA), Accra Metropolis, and stratified by road-networks. LAeq24hr Lden (dBA) Lday (dBA) Lnight (dBA) (dBA) GAMA 57.0 (54.8, 60.2 (58.2, 58.5 (56.1, 51.2 (49.6, 59.3) 62.4) 60.7) 53.3) Roadsa Major roads 65.1 (61.8, 68.5 (65.2, 66.4 (63.0, 59.9 (56.6, 68.4) 71.8) 69.8) 63.4) Secondary/ 60.3 (57.5, 63.4 (60.9, 61.7 (58.8, 54.3 (52.3, tertiary roads 63.3) 66.4) 64.7) 57.2) Minor roads 57.2 (55.1, 60.3 (58.4, 58.5 (56.3, 51.3 (49.7, 59.4) 62.4) 60.8) 53.2) Accra Metropolis 61.2 (58.0, 64.1 (61.1, 62.7 (59.4, 54.4 (51.8, 64.2) 67.0) 65.6) 57.4) Data summarized as median and interquartile ranges (IQR). a 100 m buffers were created around each road type and average noise levels were calculated amongst all the points within the 100 m buffers corresponding to each road-type. 3.4. Area-level socioeconomic inequalities of noise in Accra Metropolis We observed an inverse relationship between enumeration area Fig. 3. Cumulative densities of the proportion of the Accra Metropolis noise levels and our primary metric of SES (consumption) in Accra population living in enumeration areas (EA) with varying noise levels. The Metropolis. The poorest enumeration areas (bottom 20% of SES distri- solid grey vertical line and the dashed black vertical line shows the Lden and bution) had statistically significant (p 0.01) higher L (median: 69 Lnight limits for road-traffic noise based on WHO guidelines for the European < den region, respectively (World Health Organization, 2018). dBA) compared with the wealthiest enumeration areas in the top 20% (median: 64 dBA) with a stepwise gradient for enumeration areas in between (Fig. 4). The same trend held for night-time noise levels. Though, even within a SES quintile, there was considerable variation in 7 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 Fig. 4. Distribution of enumeration area (EA) Lden and Lnight across quintiles (20% increments) of EA socioeconomic status (SES) in Accra Metropolis. SES: EA median log equivalized household consumption. The upper and lower limits of the black box represent the interquartile range of the distribution and the hor- izontal line within the box represents the median. Each colored point represents an EA average noise level (dBA). noise levels. background. In both the day and night-time models, noise levels (Lday An inverse, but slightly weaker, relationship was found for noise and Lnight) were significantly positively associated with intermittency levels and the number of individuals with post-secondary education ratios, and the magnitudes of the associations were modified by land use within enumeration areas (Table 5). The enumeration areas in the classifications at each measurement site (Table S6). lowest quintile of this distribution had a median Lden of 69 dBA compared with the wealthiest enumeration areas at 65 dBA. The 4. Discussion weakest relationship was found for the number of unemployed in- dividuals in enumeration areas (Table 5). Environmental noise has been increasingly recognized as an envi- ronmental exposure of public health importancein growing SSA cities. However, there is scarce city-level data on environmental noise expo- 3.5. Predictors of intermittency ratios sure to aid local policy and decision making or investigate and quantify health effects. Our study is the first of its kind in SSA to model, map, and Intermittency ratios for both the day and night-time hours were investigate city-wide socioeconomic inequalities of predicted environ- negatively associated with predictor variables representing roads with mental noise exposures within a major African metropolis. We found large and constant traffic flows, such as the length of major and sec- that nearly all areas in the GAMA had Lden and Lnight levels which ondary/tertiary roads within buffers around measurement sites exceeded international guidelines. The highest levels were in the city (Table S6). However, the intermittency ratio for the day-time hours was center and near major roads. Noise levels were not equally spread across positively associated with the length of minor roads within buffers, neighborhoods as we found evidence that lower SES neighborhoods likely capturing sparse and intermittent sounds of road-traffic on these generally had higher levels compared with their wealthier counterparts. types of roads (Table S6). NDVI was positively associated with the Noise levels in Accra were positively associated with traffic-related intermittency ratio in the day-time hours, possibly due to the low variables, particularly major roads (highways, motorways), similar to background sound levels in areas with higher vegetation, and thus the previous LUR studies in North America (Fallah-Shorshani et al., 2018; ability of day-time intermittent sound events to emerge from Liu et al., 2020; Ragettli et al., 2016; Walker et al., 2017), Europe (Aguilera et al., 2015; Alam et al., 2017), Asia and the Middle East Table 5 (Chang et al., 2019; Harouvi et al., 2018; Wang et al., 2016), and South Correlation between enumeration area (EA) Lden and Lnight and EA socio- Africa (Sieber et al., 2017). Multi-lane and higher-speed roads can economic status metrics. Table shows Pearson correlation coefficients and facilitate higher traffic volumes, and attract a fleet composition with a 95% confidence intervals around correlation estimates. higher percentage of heavy vehicles that can produce higher noise levels Median log equivalized Number of individuals Number of in these areas (Curran et al., 2013). The mechanisms by which motor household with post-secondary unemployed vehicles generate noise is multi-faceted (Kang, 2017a) and include en- consumption educationa individualsa gine sounds, tire contact with the road and driver behavior such as Lden − 0.45 [-0.49, − 0.41] − 0.34 [-0.38, − 0.30] − 0.21 [-0.25, honking (Vijay et al., 2015). Thus, interventions to reduce road-traffic − 0.17] noise can take on many forms, including vehicle emissions reduction Lnight − 0.39 [-0.43, − 0.36] − 0.29 [-0.33, − 0.25] − 0.18 [-0.23, (e.g., modifications to engines and tire materials), land use planning and − 0.14] transport management (e.g., separation between roads and buildings), a The relationship was the same between the number and the percentage of the modification or creation of structures such as noise barriers or green individuals with post-secondary education or who were unemployed within vegetation (Curran et al., 2013; Kang, 2017a), and behavioral change each enumeration area. 8 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 interventions (e.g., ban on horns/honking) (Ali and Tamura, 2003; a global audience (World Health Organization, 2018). Furthermore, we Gettleman, 2020; Vijay et al., 2015). In Accra, it was estimated that 20% found that almost the entire population in Accra Metropolis (~2 million of roads are still unpaved, particularly in the poorer neighborhoods people in 2019) lived in areas where Lden and Lnight exceeded 55 and 50 (ASIRT, 2014); thus modifying pavement material (Curran et al., 2013; dBA, respectively. Based on evidence from WHO commissioned sys- Donavan, 2005) could potentially reduce some road-transport noise, tematic reviews (2018) (Basner and McGuire, 2018; Guski et al., 2017), particularly on higher-speed roads (Eurocities, 2015). As well, given the at and above these noise levels over 11% (11–49%) and over 4% dominance of transport by private vehicles, the local government in (4–12%) of the population are likely to be highly annoyed (Lden) and Accra could consider changes to urban design, placement of key ser- sleep disturbed (Lnight) from road-traffic noise exposure, respectively vices, safety measures and public messaging that inspire modal shifts (World Health Organization, 2018). As noise epidemiological research is towards cycling and walking and mass transit, as a mechanism to reduce currently lacking in Africa, future research in Accra can utilize our noise road-traffic noise in the city. These interventions also have added ben- exposure surfaces to generate local evidence of the health effects of efits related to reductions in vehicular air pollution and greenhouse gas noise; this will have the effect of strengthening and diversifying the emissions and an increase in physical activity through active transport global literature base on noise health effects around the world. (Giles-Corti et al., 2010; Nieuwenhuijsen, 2021). Recent measures to We found that in Accra, the poorest enumeration areas had higher curb vehicular air pollution emissions in Ghana, such as the regulation median Lden and Lnight levels compared with the wealthier ones. Previous and taxes imposed on the import of old (and often noisy) vehicles into noise studies conducted in Europe, North America, and China using SES the country, may have an indirect impact on road-transport related measures derived from material deprivation indicators, such as income, noise. deprived living area, or mean dwelling value found similar trends (Casey Noise levels were generally higher in the city-core (Accra Metropolis) et al., 2017; Dale et al., 2015; European Environment Agency, 2020; and in industrial (Tema Metropolis) areas (Drudge et al., 2018; Harouvi Lam and Chan, 2008). For our education measure, an indicator which et al., 2018; Ragettli et al., 2016; Sieber et al., 2017) compared with may also reflect behavioral aspects, median noise levels were similarly outlying peri-urban and formal residential areas, where vegetation, lower in enumeration areas with a higher number of individuals with which is a natural noise attenuator (Halim et al., 2015), was more post-secondary education. Our analysis of area-level education and abundant. Previous research on SSA cities suggests that outdoor noise noise reflects results from a similar study in Montreal Canada (Dale sources in these settings extend beyond road, rail, and aircraft trans- et al., 2015). Though this relationship has had mixed results among portation and can include outdoor religious activities, social even- studies that looked at individual-level associations in Europe (Dreger ts/gatherings, formal and informal commercial activities, and large and et al., 2019). In Accra, poorer communities are likely burdened by small-scale industrial activities (Olayinka, 2012; Samagwa et al., 2010; multiple environmental pollutants in addition to noise, as previous Zakpala et al., 2014). Additionally, the number of restaurants in an area, studies have found higher levels of PM2.5 air pollution concentrations in which may serve as a proxy for general ‘neighborhood’ sources of noise lower SES neighborhoods (Dionisio et al., 2010; Zhou et al., 2011). and commercial activities in our models, was positively associated with In a secondary analysis, we calculated intermittency ratios for each noise levels. Previous research from South Africa found that Lden levels measurement site and explored the potential associations of environ- were significantly positively associated with high neighborhood noise mental features and noise levels within LUR models. The challenge of annoyance (Sieber et al., 2018). It is also common for restaurants in modelling a metric, such as the intermittency ratio, with spatial LUR residential and commercial areas of Accra to play music from loud- models is that the predictor variables are often temporally static, thus speakers, and many restaurants in the city are ‘open-air’ concept, making it difficult to capture intermittency which is inherently time providing nearby residents with little protection from exposure to sound dependent. Predictors which vary in both time and space may have generated by the restaurants. Previous research from Tanzania studied allowed us to capture noise intermittency better. Furthermore, the noise at restaurants and found elevated levels both indoors and outdoors medium-low predictive accuracy of the intermittency ratio models could (Samagwa et al., 2010), in part due to music being played. The be due to the measurement of the metric itself. We calculated inter- perception of different types of city sounds can vary widely between mittency ratios with sound level data integrated every minute; thus, we countries and are intricately linked to social, cultural, and contextual would have missed some infrequent sound events/peaks that would factors related to the time and place in which the sound is perceived, as have been lost in the integration. We can also only interpret our results well as personal preferences and demographics (Deng et al., 2020; Kang, within the context of the fixed event cutoff that we used (+3 dBA). If we 2010, 2017c). Beyond a few small studies in SSA related to religious had modelled intermittency ratios with stricter cutoffs, the estimated noise making (often accompanied by loud music) (Armah et al., 2010; intermittency ratios in the GAMA would be lower (Clark et al., 2021). Zakpala et al., 2014), and music from commercial shops (Ebare et al., 2011), there is scarce research exploring whether elevated human 4.1. Strengths and limitations speech and outdoor music sounds, generated within a restaurant envi- ronment, would be considered unwanted noise or just ‘sounds of city Our research is one of the first to develop a LUR model of noise in SSA life’ by the local population in Accra. Therefore, future soundscape (Sieber et al., 2017) and the first to do so in West Africa. The models research studies conducted within this understudied environment would incorporated a suite of geospatial predictor variables and leveraged shed light on local perceptions of different types of sounds, and situa- noise measurements from a large-scale and long-term data collection tions and circumstances which impact perception. campaign. Finally, the comparison of predicted noise levels with small Almost all areas within 100 m of major and secondary/tertiary roads area SES measures is the first study to our knowledge to characterize and within the Accra Metropolis (main city center), where road-traffic inequalities of noise in a SSA setting. These models and the predicted noise sources are highly prevalent (Clark et al., 2021), had predicted noise surfaces provide opportunities for major environmental epidemi- noise levels which exceeded the World Health Organization (WHO) ologic studies that would provide locally sound and globally relevant guidelines for road traffic noise (Lden (53 dB), Lnight (45 dB)) (World data on noise health effects within this understudied region. The noise Health Organization, 2018). Chronic exposure to road-traffic noise exposure surfaces can also be used to conduct environmental burden of beyond these guideline thresholds is associated with adverse health ef- disease assessments, which can feed into local noise policy and decision fects including sleep disturbance, annoyance, and cardiovascular dis- making. eases. While the guidelines were developed for the European region, Our research has several limitations. While we did include a wide with the majority of the evidence underpinning them from European variety of spatial predictor variables in the study, we were not able to and North American countries, the WHO report does state that the obtain spatially and/or temporally resolved information on traffic vol- guidelines can be considered applicable in other regions and suitable for ume and fleet composition. Inclusion of this information could have 9 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 improved model predictive accuracy, particularly from traffic-related editing. Jill Baumgartner: Methodology, Writing – review & editing. sources. Further, we were not able to capture potential small-scale James Bennett: Conceptualization, Formal analysis, Methodology, Su- variations in noise propagation due to sound reflection or absorption pervision, Software, Writing – review & editing. Raphael Arku: in the built environment as we did not have data for building height and Conceptualization, Project administration, Supervision, Writing – re- material and ground material (Kang, 2017a). We also made the view & editing. assumption that the spatial predictors were stationary in time and representative of the period when the noise measurements were taken. Funding sources This assumption may not be true for all spatial predictors as the census data which was used to estimate population counts was generated in This work was supported by the Pathways to Equitable Healthy Cities 2010 and the dataset used to estimate land cover dates to 2014. The grant from the Wellcome Trust [209376/Z/17/Z]. For the purpose of temporal misalignment of some of the predictor variables may be open access, the authors have applied a CC BY public copyright license especially relevant for a rapidly urbanizing context such as the GAMA, to any Author Accepted Manuscript version arising from this submis- and particularly its peripheries outside of the city center. With respect to sion. SC was supported by a Canadian Institutes for Health Research SES inequalities of noise, our analysis was at the enumeration area level, (CIHR) PhD scholarship as well as an Imperial College Presidents PhD and we recognize that associations at the individual level may be scholarship. Infrastructure support for the Department of Epidemiology different. There is also a temporal misalignment between the noise and and Biostatistics at Imperial College was provided by the NIHR Imperial SES data. We used SES metrics estimated from the 2012 GLSS and the Biomedical Research Centre (BRC). 2010 census as they were the most recent data of its kind, though the noise data were collected in 2019/2020. It is possible that the spatial distribution of SES in some parts of the GAMA in 2010 differ to present Declaration of competing interest day realities (2019/2020). Though we expect this to be minimal in the city center (Accra Metropolis) where we conducted the SES analysis, as The authors declare that they have no known competing financial the major changes to within city migration, land use, and urban plan- interests or personal relationships that could have appeared to influence ning, are taking place at the peripheries of the GAMA (Addae and the work reported in this paper. Oppelt, 2019). Future work incorporating the 2020 census is warranted to verify if trends have remained the same or changed. The 2020 census Acknowledgements was delayed due to the COVID-19 Pandemic but may be completed and data released in a few of years. We thank the many Accra residents who generously allowed us to install monitors on their property and the Physics Department staff at 5. Conclusion the University of Ghana for their assistance in setting up the laboratory. This work is supported by the Pathways to Equitable Healthy Cities The measured and predicted noise levels exceeded international grant from the Wellcome Trust [209376/Z/17/Z]. For the purpose of health-based guidelines almost everywhere in the Greater Accra open access, the authors have applied a CC BY public copyright license Metropolitan Area. At these levels, it is likely that common adverse to any Author Accepted Manuscript version arising from this submis- health impacts attributable to environmental noise exposures, such as sion. SC was supported by a Canadian Institutes for Health Research annoyance, sleep disturbance, and cardiovascular diseases, are experi- (CIHR) PhD scholarship as well as an Imperial College Presidents PhD enced within the city. Furthermore, noise levels varied unequally across scholarship. Infrastructure support for the Department of Epidemiology the city and poorer neighborhoods were generally worse off in terms of and Biostatistics at Imperial College was provided by the NIHR Imperial noise levels than the wealthiest neighborhoods. The spatial and social Biomedical Research Centre (BRC). inequalities in environmental noise in Accra further highlight the need for local government to consider the equity impacts of urban planning Appendix A. Supplementary data and policy decision making. This is particularly the case as inequalities in noise exposure, compounded with socioeconomic inequalities and Supplementary data to this article can be found online at https://doi. other environments exposures (e.g., air pollution), could further org/10.1016/j.envres.2022.113932. entrench health inequalities in Accra. City-level actions are needed to tackle this environmental exposure in Accra though changes in infra- References structure, services and regulations that could also have broader and equitable benefits for health and wellbeing. Acheampong, R.A., Siiba, A., Okyere, D.K., Tuffour, J.P., 2020. Mobility-on-demand: an empirical study of internet-based ride-hailing adoption factors, travel characteristics and mode substitution effects. Transport. Res. C Emerg. Technol. 115, 102638 Credit Author Statement https://doi.org/10.1016/j.trc.2020.102638. Addae, B., Oppelt, N., 2019. Land-use/land-cover change analysis and urban growth Sierra N Clark: Conceptualization, Data curation, Formal analysis, modelling in the greater Accra metropolitan area (GAMA), Ghana. Urban Sci. 3, 26. https://doi.org/10.3390/urbansci3010026. Funding acquisition, Methodology, Project administration, Visualiza- Aguilera, I., Foraster, M., Basagaña, X., Corradi, E., Deltell, A., Morelli, X., Phuleria, H.C., tion, Writing – original draft, Writing – review & editing. Abosede S Alli: Ragettli, M.S., Rivera, M., Thomasson, A., Slama, R., Künzli, N., 2015. Application of Data curation, Project administration, Writing – review & editing. Majid land use regression modelling to assess the spatial distribution of road traffic noise in three European cities. J. Expo. Sci. Environ. Epidemiol. 25, 97–105. https://doi.org/ Ezzati: Conceptualization, Methodology, Supervision, Funding acquisi- 10.1038/jes.2014.61. tion, Resources, Writing – review & editing. Michael Brauer: Method- Alam, M.S., Corcoran, L., King, E.A., McNabola, A., Pilla, F., 2017. Modelling of intra- ology, Supervision, Writing – review & editing. Mireille B Toledano: urban variability of prevailing ambient noise at different temporal resolution. Noise Mapp. 4, 20–44. https://doi.org/10.1515/noise-2017-0002. Supervision, Writing – review & editing. James Nimo: Data curation, Ali, S.A., Tamura, A., 2003. Road traffic noise levels, restrictions and annoyance in Writing – review & editing. Josephine Bedford Moses: Data curation, Greater Cairo, Egypt. Appl. Acoust. 64, 815–823. https://doi.org/10.1016/S0003- Writing – review & editing. Solomon Baah: Data curation, Writing – 682X(03)00031-8. Amegah, A.K., Agyei-Mensah, S., 2017. Urban air pollution in sub-saharan Africa: time review & editing. Allison Hughes: Data curation, Resources, Writing – for action. Environ. Pollut. 220, 738–743. https://doi.org/10.1016/j. review & editing. Alicia Cavanaugh: Data curation, Formal analysis, envpol.2016.09.042. Writing – review & editing. Samuel Agyei-Mensah: Data curation, Re- Annim, S.K., Mariwah, S., Sebu, J., 2012. Spatial inequality and household poverty in Ghana. Econ. Syst. 36, 487–505. https://doi.org/10.1016/j.ecosys.2012.05.002. sources, Writing – review & editing. George Owusu: Resources, Writing Armah, F.A., Odoi, J.O., Yawson, D.O., Yengoh, G.T., Afrifa, E.K.A., Pappoe, A.N.M., – review & editing. Brian Robinson: Data curation, Writing – review & 2010. Mapping of noise risk zones derived from religious activities and perceptions 10 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 in residential neighbourhoods in the Cape Coast metropolis, Ghana. Environ. Ebare, M.N., Omuemu, V.O., Isah, E.C., 2011. Assessment of noise levels generated by Hazards 9, 358–368. https://doi.org/10.3763/ehaz.2010.0003. music shops in an urban city in Nigeria. Publ. Health 125, 660–664. https://doi.org/ ASIRT, 2014. Road Travel Report: Republic of Ghana. Rockville. 10.1016/j.puhe.2011.06.009. Baloye, D.O., Palamuleni, L.G., 2015. A comparative land use-based analysis of noise Elbers, C., Lanjouw, J.O., Lanjouw, P., 2003. Micro-level estimation of poverty and pollution levels in selected urban centers of Nigeria. Int. J. Environ. Res. Publ. Health inequality. Econometrica 71, 355–364. https://doi.org/10.1111/1468-0262.00399. 12, 12225–12246. https://doi.org/10.3390/ijerph121012225. European Environment Agency, 2020. Environmental Noise in Europe - 2020, European Barrington-Leigh, C., Millard-Ball, A., 2017. The world’s user-generated road map is Environment Agency. more than 80% complete. PLoS One 12, e0180698. https://doi.org/10.1371/ Eurocities, 2015. Low-noise Road Surfaces. Brussels. journal.pone.0180698. Eze, I.C., Foraster, M., Schaffner, E., Vienneau, D., Héritier, H., Rudzik, F., Thiesse, L., Basner, M., McGuire, S., 2018. WHO environmental noise guidelines for the European Pieren, R., Imboden, M., Eckardstein, A. von, Schindler, C., Brink, M., Cajochen, C., region: a systematic review on environmental noise and effects on sleep. Int. J. Wunderli, J.M., Röösli, M., Probst-Hensch, N., 2017. Long-term exposure to Environ. Res. Publ. Health 15, 519. https://doi.org/10.3390/ijerph15030519. transportation noise and air pollution in relation to incident diabetes in the Bediako-Akoto, R.D.O., 2018. Noise pollution: a country at risk. Daily Graphic SAPALDIA study. Int. J. Epidemiol. 46, 1115–1125. https://doi.org/10.1093/ije/ Newspaper (Ghana). dyx020. Böcker, L., Dijst, M., Prillwitz, J., 2013. Impact of everyday weather on individual daily Fallah-Shorshani, M., Minet, L., Liu, R., Plante, C., Goudreau, S., Oiamo, T., travel behaviours in perspective: a literature review. Transp. Rev. 33, 71–91. Smargiassi, A., Weichenthal, S., Hatzopoulou, M., 2018. Capturing the spatial https://doi.org/10.1080/01441647.2012.747114. variability of noise levels based on a short-term monitoring campaign and comparing Brink, M., Schäffer, B., Vienneau, D., Foraster, M., Pieren, R., Eze, I.C., Cajochen, C., noise surfaces against personal exposures collected through a panel study. Environ. Probst-Hensch, N., Röösli, M., Wunderli, J.M., 2019a. A survey on exposure-response Res. 167, 662–672. https://doi.org/10.1016/j.envres.2018.08.021. relationships for road, rail, and aircraft noise annoyance: differences between Fobil, J., May, J., Kraemer, A., 2010. Assessing the relationship between socioeconomic continuous and intermittent noise. Environ. Int. 125, 277–290. https://doi.org/ conditions and urban environmental quality in Accra, Ghana. Int. J. Environ. Res. 10.1016/j.envint.2019.01.043. Publ. Health 7, 125–145. https://doi.org/10.3390/ijerph7010125. Brink, M., Schäffer, B., Vienneau, D., Pieren, R., Foraster, M., Eze, I.C., Rudzik, F., Foraster, M., Eze, I.C., Schaffner, E., Vienneau, D., Héritier, H., Endes, S., Rudzik, F., Thiesse, L., Cajochen, C., Probst-Hensch, N., Röösli, M., Wunderli, J.M., 2019b. Self- Thiesse, L., Pieren, R., Schindler, C., Schmidt-Trucksäss, A., Brink, M., Cajochen, C., reported sleep disturbance from road, rail and aircraft noise: exposure-response Wunderli, J.M., Röösli, M., Probst-Hensch, N., 2017. Exposure to road, railway, and relationships and effect modifiers in the SiRENE study. Int. J. Environ. Res. Publ. aircraft noise and arterial stiffness in the SAPALDIA study: annual average noise Health 16. https://doi.org/10.3390/ijerph16214186. levels and temporal noise characteristics. Environ. Health Perspect. 125, 1–8. Casey, J., Morello-Frosch, R., Mennitt, J.D., Fristrup, K., Ogburn, L.E., James, P., 2017. https://doi.org/10.1289/EHP1136. Race/ethnicity, socioeconomic status, residential segregation, and spatial variation Garg, N., Maji, S., 2014. A critical review of principal traffic noise models: strategies and in noise exposure in the contiguous United States. Environ. Health Perspect. 125, 10. implications. Environ. Impact Assess. Rev. 46, 68–81. https://doi.org/10.1016/j. https://doi.org/10.1289/EHP898. eiar.2014.02.001. Chang, T.Y., Liang, C.H., Wu, C.F., Chang, L. Te, 2019. Application of land-use regression Gettleman, J., 2020. Mumbai police play trick on honking drivers [WWW Document]. models to estimate sound pressure levels and frequency components of road traffic New York Times. https://www.nytimes.com/2020/02/04/world/asia/mumbai-hor noise in Taichung, Taiwan. Environ. Int. 131, 104959 https://doi.org/10.1016/j. n-honking.html?auth=login-google accessed 12.19.2020. envint.2019.104959. Ghana Airports, 2017. Flight statistics for 2017 [WWW Document]. URL. https://www. Clark, C., Crumpler, C., Notley, H., 2020. Evidence for environmental noise effects on gacl.com.gh/traffic-stats/#1523274746929-18a28425-0abb. accessed 12.4.2021. health for the United Kingdom policy context: a systematic review of the effects of Ghana Statisical Service, 2010. Ghana Population and Housing Census. environmental noise on mental health, wellbeing, quality of life, cancer, dementia, Ghana Statistical Service, 2019. Greater Accra Population: Population by Sex and District birth, reproductive outcomes, and cognition. Int. J. Environ. Res. Publ. Health 17. 2010 and 2019 [WWW Document]. URL. https://statsghana.gov.gh/nationalaccount https://doi.org/10.3390/ijerph17020393. _macros.php?Stats=MTA1NTY1NjgxLjUwNg==/webstats/s679n2sn87. accessed Clark, S.N., Alli, A., Brauer, M., Ezzati, M., Baumgartner, J.C., Toledano, M.B., 3.15.2021. Hughes, A., Nimo, J., Moses, J., Terkpertey, S., Vallarino, J., Agyei-Mensah, S., Ghinet, S., Price, A., Daigle, G.A., Stinson, M.R., Grewal, A., Wickramasinghe, V., 2019. Agyemang, E., Nathvani, R., Muller, E., Bennett, J., Wang, J., Beddows, A., Kelly, F., Atmospheric propagation of aircraft acoustic signature from high altitude. INTER- Barratt, B., Beevers, S., Arku, R.E., 2020. High-resolution spatiotemporal NOISE 2019 MADRID - 48th Int. Congr. Exhib. Noise Control Eng. measurement of air and environmental noise pollution in sub-saharan african cities: Giles-Corti, B., Foster, S., Shilton, T., Falconer, R., 2010. The co-benefits for health of Pathways to Equitable Healthy Cities Study protocol for Accra, Ghana. BMJ Open investing in active transportation. NSW Public Health Bull. 21, 122–127. https://doi. 1–10. https://doi.org/10.1136/bmjopen-2019-035798. org/10.1071/NB10027. Clark, S.N., Alli, A.S., Nathvani, R., Huges, A., Ezzati, M., Brauer, M., Toledano, M.B., Guski, R., Schreckenberg, D., Schuemer, R., 2017. WHO environmental noise guidelines Baumgartner, J., Bennett, J.E., Nimo, J., Moses, J.B., Baah, S., Agyei-Mensah, S., for the European region: a systematic review on environmental noise and annoyance. Owusu, G., Croft, B., Arku, R.E., 2021. Space-time characterization of community Int. J. Environ. Res. Publ. Health 14, 1539. https://doi.org/10.3390/ noise and sound sources in Accra, Ghana. Sci. Rep. 1–12. https://doi.org/10.1038/ ijerph14121539. s41598-021-90454-6. Halim, H., Abdullah, R., Ali, A.A.A., Nor, M.J.M., 2015. Effectiveness of existing noise Corral, P., Molina, I., Nguyen, M., 2020. Pull Your Small Area Estimates Up by the barriers: comparison between vegetation, concrete hollow block, and panel concrete. Bootstraps. Policy Research. World Bank, Washington, DC. Working Paper; No. Procedia Environ. Sci. 30, 217–221. https://doi.org/10.1016/j.proenv.2015.10.039. 9256. Hammer, M.S., Swinburn, T.K., Neitzel, R.L., 2014. Environmental noise pollution in the Council of the European Union, 2002. Directive of the European Parliament and of the United States: developing an effective public health response. Environ. Health Council of 25 June 2002 Relating to the Assessment and Management of Perspect. 122, 115. https://doi.org/10.1289/ehp.1307272. Environmental Noise. Harouvi, O., Ben-Elia, E., Factor, R., de Hoogh, K., Kloog, I., 2018. Noise estimation Curran, J.H., Ward, H.D., Shum, M., Davies, H.W., 2013. Reducing cardiovascular health model development using high-resolution transportation and land use regression. impacts from traffic-related noise and air pollution: intervention strategies. Environ. J. Expo. Sci. Environ. Epidemiol. 28, 559–567. https://doi.org/10.1038/s41370- Health Rev. 56, 31–38. https://doi.org/10.5864/d2013-011. 018-0035-z. Dale, L.M., Goudreau, S., Perron, S., Ragettli, M.S., Hatzopoulou, M., Smargiassi, A., Heathrow Airport, 2018. Facts and Figures. LHR Airports Limited [WWW Document]. htt 2015. Socioeconomic status and environmental noise exposure in Montreal, Canada. ps://www.heathrow.com/company/about-heathrow/company-information/facts- BMC Publ. Health. https://doi.org/10.1186/s12889-015-1571-2. and-figures. Deaton, A., 1992. Consumption and permanent income. In: Understanding Consumption. Hoek, G., Beelen, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P., Briggs, D., Oxford University Press. https://doi.org/10.1093/0198288247.001.0001. 2008. A review of land-use regression models to assess spatial variation of outdoor Deaton, A., Zaidi, S., 2002. Guidelines for Constructing Consumption Aggregates for air pollution. Atmos. Environ. 42, 7561–7578. https://doi.org/10.1016/j. Welfare Analysis. World Bank, Washington, DC. Working Paper No. 135. atmosenv.2008.05.057. Deng, L., Kang, J., Zhao, W., Jambrošić, K., 2020. Cross-national comparison of Imoro Musah, B., Peng, L., Xu, Y., 2020. Urban congestion and pollution: a quest for soundscape in urban public open spaces between China and Croatia. Appl. Sci. 10, cogent solutions for Accra city. IOP Conf. Ser. Earth Environ. Sci. 435 https://doi. 960. https://doi.org/10.3390/app10030960. org/10.1088/1755-1315/435/1/012026. Dionisio, K., Arku, R.E., Hughes, A.F., Jose Vallarin, O., Carmichael, H., Spengler, J.D., Kaledzi, I., 2018. Ghana Asks Mosques to Turn Down the Noise and Use WhatsApp for Agyei-Mensah, S., Ezzati, M., 2010. Air pollution in Accra neighborhoods: spatial, Call to Prayer. DW [WWW Document]. URL. https://www.dw.com/en/ghana-asks- socioeconomic, and temporal patterns. Environ. Sci. Technol. 44, 2270–2276. mosques-to-turn-down-the-noise-and-use-whatsapp-for-call-to-prayer/a-43373007. https://doi.org/10.1021/es903276s. (Accessed 2 April 2021). Donavan, P.R., 2005. Comparative measurements of tire/pavement noise in Europe and Kang, J., 2010. Sound environment: high-versus low-density cities. In: Designing High- the United States. Noise News Int. 13, 46–53. https://doi.org/10.3397/1.3703067. Density Cities for Social, Environmental Sustainability, pp. 163–180. Dreger, S., Schüle, S.A., Hilz, L.K., Bolte, G., 2019. Social inequalities in environmental Kang, J., 2017a. Urban Sound Environment, first ed. Routledge. noise exposure: a review of evidence in the WHO european region. Int. J. Environ. Kang, J., 2017b. Macroscale accoustic modelling. In: Urban Sound Environment, p. 154. Res. Publ. Health 16. https://doi.org/10.3390/ijerph16061011. Kang, J., 2017c. Urban soundscape. In: Urban Sound Environment. Routledge, Drudge, C., Johnson, J., MacIntyre, E., Li, Y., Copes, R., Ing, S., Johnson, S., pp. 43–107. Varughese, S., Chen, H., 2018. Exploring nighttime road traffic noise: a Kazeem, Y., Dahir, A.L., 2018. African Cities Are Battling Esclating Noise Pollution - but comprehensive predictive surface for Toronto, Canada. J. Occup. Environ. Hyg. 15, Religion Stands in the Way. Quartz Africa [WWW Document]. https://qz.com/africa 389–398. https://doi.org/10.1080/15459624.2018.1442006. /1255785/ghana-mosques-to-use-whatsapp-for-call-to-prayers-to-stop-noise-pollut ion-of-african-cities/. 11 S.N. Clark et al. E n v i r o n m e n t a l R e s e a r c h 214 (2022) 113932 Khan, J., Ketzel, M., Kakosimos, K., Sørensen, M., Jensen, S.S., 2018. Road traffic air and Wunderli, J.M., Probst-Hensch, N., Röösli, M., Cajochen, C., 2020. Transportation noise pollution exposure assessment – a review of tools and techniques. Sci. Total noise impairs cardiovascular function without altering sleep: the importance of Environ. 634, 661–676. https://doi.org/10.1016/j.scitotenv.2018.03.374. autonomic arousals. Environ. Res. 182, 109086 https://doi.org/10.1016/j. Knott, S., Gyamfi Asiedu, K., 2019. ‘If You Complain They See You as Evi’: Accra’s envres.2019.109086. Religious Noise Problem. Guardian [WWW Document]. https://www.theguardian. Thompson, R., Smith, R.B., Bou Karim, Y., Shen, C., Drummond, K., Teng, C., com/cities/2019/mar/27/if-you-complain-they-see-you-as-evil-accras-religious Toledano, M.B., 2022. Noise pollution and human cognition: an updated systematic -noise-problem. review and meta-analysis of recent evidence. Environ. Int. 158, 106905 https://doi. Lam, K., Chan, P.-K., 2008. Socio-economic status and inequalities in exposure to org/10.1016/j.envint.2021.106905. transportation noise in Hong Kong. Open Environ. Sci. 2, 107–113. https://doi.org/ Sound propagation. In: Truax, B. (Ed.), 1999. Handbook for Acoustic Ecology. 10.2174/1876325100802010107. U.S Geological Survey, n.d. Landsat products [WWW Document]. URL https://www.usgs Larkin, A., Geddes, J.A., Martin, R.V., Xiao, Q., Liu, Y., Marshall, J.D., Brauer, M., .gov/core-science-systems/nli/landsat. Hystad, P., 2017. Global land use regression model for nitrogen dioxide air pollution. van Kamp, I., Simon, S., Notley, H., Baliatsas, C., van Kempen, E., 2020. Evidence Environ. Sci. Technol. 51, 6957–6964. https://doi.org/10.1021/acs.est.7b01148. relating to environmental noise exposure and annoyance, sleep disturbance, cardio- Lee, M., Brauer, M., Wong, P., Tang, R., Tsui, T.H., Choi, C., Cheng, W., Lai, P.-C., vascular and metabolic health outcomes in the context of IGCB (N): a scoping review Tian, L., Thach, T.-Q., Allen, R., Barratt, B., 2017. Land use regression modelling of of evidence regarding sources other than transport noise. Proc. 2020 Int. Congr. air pollution in high density high rise cities: a case study in Hong Kong. Sci. Total Noise Control Eng. INTER-NOISE 1–21. Environ. 592 https://doi.org/10.1016/j.scitotenv.2017.03.094. van Kempen, E., Casas, M., Pershagen, G., Foraster, M., 2018. WHO environmental noise Liu, Y., Goudreau, S., Oiamo, T., Rainham, D., Hatzopoulou, M., Chen, H., Davies, H., guidelines for the European region: a systematic review on environmental noise and Tremblay, M., Johnson, J., Bockstael, A., Leroux, T., Smargiassi, A., 2020. cardiovascular and metabolic effects: a summary. Int. J. Environ. Res. Publ. Health Comparison of land use regression and random forests models on estimating noise 15, 1–59. https://doi.org/10.3390/ijerph15020379. levels in five Canadian cities. Environ. Pollut. 256, 113367 https://doi.org/10.1016/ Verdin, K.L., 2017. Digital elevation model (DEM) from the hydrologic derivatives for j.envpol.2019.113367. modeling and analysis (HDMA) database – Africa. URL. https://www.sciencebase.go Münzel, T., Sørensen, M., Daiber, A., 2021. Transportation noise pollution and v/catalog/item/591f6d02e4b0ac16dbdde1c7. cardiovascular disease. Nat. Rev. Cardiol., 0123456789 https://doi.org/10.1038/ Vienneau, D., Eze, I.C., Probst-Hensch, N., Rsli, M., 2019. Association between s41569-021-00532-5. transportation noise and cardio-metabolic diseases: an update of the WHO meta- Nieuwenhuijsen, M.J., 2021. New urban models for more sustainable, liveable and analysis. Proc. Int. Congr. Acoust. 1543–1550. https://doi.org/10.18154/RWTH- healthier cities post covid19; reducing air pollution, noise and heat island effects and CONV-239440. increasing green space and physical activity. Environ. Int. 157, 106850 https://doi. Vienneau, D., Saucy, A., Schäffer, B., Flückiger, B., Tangermann, L., Stafoggia, M., org/10.1016/j.envint.2021.106850. Wunderli, J.M., Röösli, M., 2022. Transportation noise exposure and cardiovascular Olayinka, O.S., 2012. Noise pollution in urban areas: the neglected dimensions. Environ. mortality: 15-years of follow-up in a nationwide prospective cohort in Switzerland. Res. J. 6, 259–271. https://doi.org/10.3923/erj.2012.259.271. Environ. Int. 158 https://doi.org/10.1016/j.envint.2021.106974. OpenStreetMap, 2015. Planet dump [WWW Document]. URL. https://planet.openstreet Vijay, R., Sharma, A., Chakrabarti, T., Gupta, R., 2015. Assessment of honking impact on map.org. traffic noise in urban traffic environment of Nagpur, India. J. Environ. Heal. Sci. Eng. Price, R., Hallas, M., 2019. IN41A-02 Mapping Every Building and Road in Sub-saharan 13 https://doi.org/10.1186/s40201-015-0164-4. Africa. AGU Fall Meeting. Walker, E.D., Hart, J.E., Koutrakis, P., Cavallari, J.M., VoPham, T., Luna, M., Laden, F., Raess, M., Brentani, A., Ledebur de Antas de Campos, B., Flückiger, B., de Hoogh, K., 2017. Spatial and temporal determinants of A-weighted and frequency specific Fink, G., Röösli, M., 2021. Land use regression modelling of community noise in São sound levels—an elastic net approach. Environ. Res. 159, 491–499. https://doi.org/ Paulo, Brazil. Environ. Res. 199 https://doi.org/10.1016/j.envres.2021.111231. 10.1016/j.envres.2017.08.034. Ragettli, M.S., Goudreau, S., Plante, C., Fournier, M., Hatzopoulou, M., Perron, S., Wang, V.S., Lo, E.W., Liang, C.H., Chao, K.P., Bao, B.Y., Chang, T.Y., 2016. Temporal and Smargiassi, A., 2016. Statistical modeling of the spatial variability of environmental spatial variations in road traffic noise for different frequency components in noise levels in Montreal, Canada, using noise measurements and land use metropolitan Taichung, Taiwan. Environ. Pollut. 219, 174–181. https://doi.org/ characteristics. J. Expo. Sci. Environ. Epidemiol. 26, 597–605. https://doi.org/ 10.1016/j.envpol.2016.10.055. 10.1038/jes.2015.82. Wawa, E.A., Mulaku, G.C., 2015. Noise pollution mapping using GIS in Nairobi, Kenya. Samagwa, D., Mkoma, S., Tungaraza, C., 2010. Investigation of noise pollution in J. Geogr. Inf. Syst. 7, 486–493. https://doi.org/10.4236/jgis.2015.75039. restaurants in Morogoro municipality, Tanzania, East Africa. J. Appl. Sci. Environ. World Bank, 2014. 2014 Land Cover Classification of Accra. Ghana. Manag. 13 https://doi.org/10.4314/jasem.v13i4.55395. World Health Organization, 2018. WHO Environmental Noise Guidelines for the Sieber, C., Ragettli, M.S., Brink, M., Toyib, O., Baatjies, R., Saucy, A., Probst-Hensch, N., European Region. Geneva. Dalvie, M.A., Röösli, M., 2017. Land use regression modeling of outdoor noise Wunderli, J.M., Pieren, R., Habermacher, M., Vienneau, D., Cajochen, C., Probst- exposure in informal settlements in Western Cape, South Africa. Int. J. Environ. Res. Hensch, N., Röösli, M., Brink, M., 2016. Intermittency ratio: a metric reflecting short- Publ. Health 14, 1262. https://doi.org/10.3390/ijerph14101262. term temporal variations of transportation noise exposure. J. Expo. Sci. Environ. Sieber, C., Ragettli, M.S., Brink, M., Olaniyan, T., Baatjies, R., Saucy, A., Vienneau, D., Epidemiol. 26, 575–585. https://doi.org/10.1038/jes.2015.56. Probst-Hensch, N., Dalvie, M.A., Röösli, M., 2018. Comparison of sensitivity and Xie, D., Liu, Y., Chen, J., 2011. Mapping Urban environmental noise: a land use annoyance to road traffic and community noise between a South African and a Swiss regression method. Environ. Sci. Technol. 45, 7358–7364. https://doi.org/10.1021/ population sample. Environ. Pollut. 241, 1056–1062. https://doi.org/10.1016/j. es200785x. envpol.2018.06.007. Zakpala, R.N., Armah, F.A., Sackey, B.M., Pabi, O., 2014. Night-time decibel hell: Sietchiping, R., Permezel, M.J., Ngomsi, C., 2012. Transport and mobility in sub-Saharan mapping noise exposure zones and individual annoyance ratings in an urban African cities: an overview of practices, lessons and options for improvements. Cities environment in Ghana. Scientifica (Cairo) 1–11. https://doi.org/10.1155/2014/ 29, 183–189. https://doi.org/10.1016/j.cities.2011.11.005. 892105. Ghana Standards Authority, 2018. Health Protection - Requirements for Ambient Noise Zhou, Z., Dionisio, K.L., Arku, R.E., Quaye, A., Hughes, A.F., Vallarino, J., Spengler, J.D., Control. Accra https://doi.org/GS 1222:2018. Hill, A., Agyei-Mensah, S., Ezzati, M., 2011. Household and community poverty, Thiesse, L., Rudzik, F., Kraemer, J.F., Spiegel, K., Leproult, R., Wessel, N., Pieren, R., biomass use, and air pollution in Accra, Ghana. Proc. Natl. Acad. Sci. USA 108, Héritier, H., Eze, I.C., Foraster, M., Garbazza, C., Vienneau, D., Brink, M., 11028–11033. https://doi.org/10.1073/pnas.1019183108. 12