Developed Land Use Regression Models Research Articles

Winter and Summer PM2.5 Land Use Regression Models for the City of Novi Sad, Serbia

In this study, we describe the development of seasonal winter and summer (heating and non-heating season) land use regression (LUR) models for PM2.5 mass concentration for the city of Novi Sad, Serbia. The PM2.5 data were obtained through an extensive seasonal measurement campaign conducted at 21 locations in urban, urban/industrial, industrial and background areas in the period from February 2020–July 2021. At each location, PM2.5 samples were collected on quartz fibre filters for 10 days per season using a reference gravimetric pump. The developed heating season model had two predictors, the first can be associated with domestic heating over a larger area and the second with local traffic. These predictors contributed to the adjusted R2 of 0.33 and 0.55, respectively. The developed non-heating season model had one predictor which can be associated with local traffic, which contributed to the adjusted R2 of 0.40. Leave-one-out cross-validation determined RMSE/mean absolute error for the heating and non-heating season model were 4.04/4.80 μg/m3 and 2.80/3.17 μg/m3, respectively. For purposes of completeness, developed LUR models were also compared to a simple linear model which utilizes satellite aerosol optical depth data for PM2.5 estimation, and showed superior performance. The developed LUR models can help with quantification of differences between seasonal levels of air pollution, and, consequently, air pollution exposure and association between seasonal long-term exposure and possible health risk implications.

Land Use Regression Models for Particle Number Concentration and Black Carbon in Lanzhou, Northwest of China

It is necessary to predict the spatial variation in particle number concentration (PNC) and black carbon (BC) because they are considered air pollutants associated with traffic and many diseases. In this study, land use regression (LUR) models for PNC and BC were developed based on a mobile monitoring campaign in January 2020 in Lanzhou, and the performance of models was evaluated with hold-out validation (HV) and leave-one-out cross-validation (LOOCV). The results show that the adjusted R2 of the LUR models for PNC and BC are 0.51 and 0.53, respectively. The R2 of HV and LOOCV are 0.43 and 0.44, respectively, for the PNC model and 0.42 and 0.50, respectively, for the BC model. The performances of the LUR models are of a moderate level. The spatial distribution of the predicted PNC is related to the distance from water bodies. The high PNC is related to industrial pollution. The BC concentration decreases from south to north. High BC concentrations are associated with freight distribution centres and coal-fired power plants. The range of PNC particle sizes in this study is larger than in most studies. As one of few studies in Lanzhou to develop LUR models of air pollutants, it is important to accurately estimate pollutant concentrations to improve air quality and provide health benefits for residents.

Development of historic monthly land use regression models of SO2, NOx and suspended particulate matter for birth cohort ELSPAC

Vulnerable windows in child development in utero and after birth are critical time points for uncovering the links between environment and health. Particular attention is paid to the first 1000 days of life from conception to the second year of life.The ELSPAC (European Longitudinal Study of Pregnancy and Childhood) birth cohort, launched in the early 1990s, is a rich source of longitudinal data about health and life events, based mainly in Brno, Czechia. There are currently no air quality concentration maps that can be used to assess exposure to air pollutants for this period of the 1990s in Central Europe. Simply transferring current models to the 1990's is burdened with the error introduced by the temporal change in emission sources and land use of the area. Therefore, Czech air quality monitoring data were used to develop monthly land use regression (LUR) models, which combine collected spatial variables with monitoring data to predict the variation in exposures to pollutants. Monthly pollutant concentrations were regressed against the GIS-based potential predictor variables to develop LUR models, following a supervised forward linear regression, with several predefined constraints.We constructed 180 LUR monthly models for sulphur dioxide (SO2), nitrogen oxides (NOx) and suspended particulate matter (SPM) for 1990–1994, that completely cover the first 1000 days for all ELSPAC study participants. The final models showed, on average reasonably good performance (adjusted R2 = 0.59 with hold-out validation (HOV) R2 = 0.40 for SO2; adjusted R2 = 0.75 with HOV R2 = 0.35 for NOx; and adjusted R2 = 0.61 with HOV R2 = 0.31 for SPM; with a mean number of stations of 74, 38 and 41, respectively). For these models, roads and greenness were predominantly selected as the best predictors.The modelled exposures will serve in many subsequent ELSPAC epidemiological studies, but our models may be also used in other Czech and possibly other Central European cities in that period.

Land-Use Regression Modeling to Estimate NO2 and VOC Concentrations in Pohang City, South Korea

Land-use regression (LUR) has emerged as a promising technique for air pollution modeling to obtain the spatial distribution of air pollutants for epidemiological studies. LUR uses traffic, geographic, and monitoring data to develop regression models and then predict the concentration of air pollutants in the same area. To identify the spatial distribution of nitrogen dioxide (NO2), benzene, toluene, and m-p-xylene, we developed LUR models in Pohang City, one of the largest industrialized areas in Korea. Passive samplings were conducted during two 2-week integrated sampling periods in September 2010 and March 2011, at 50 sampling locations. For LUR model development, predictor variables were calculated based on land use, road lengths, point sources, satellite remote sensing, and population density. The averaged mean concentrations of NO2, benzene, toluene, and m-p-xylene were 28.4 µg/m3, 2.40 µg/m3, 15.36 µg/m3, and 0.21 µg/m3, respectively. In terms of model-based R2 values, the model for NO2 included four independent variables, showing R2 = 0.65. While the benzene and m-p-xylene models showed the same R2 values (0.43), toluene showed a lower R2 value (0.35). We estimated long-term concentrations of NO2 and VOCs at 167,057 addresses in Pohang. Our study could hold particular promise in an epidemiological setting having significant health effects associated with small area variations and encourage the extended study using LUR modeling in Asia.

Application of land use regression modelling to describe atmospheric levels of semivolatile organic compounds on a national scale

Despite the success of passive sampler-based monitoring networks in capturing global atmospheric distributions of semivolatile organic compounds (SVOCs), their limited spatial resolution remains a challenge. Adequate spatial coverage is necessary to better characterize concentration gradients, identify point sources, estimate human exposure, and evaluate the effectiveness of chemical regulations such as the Stockholm Convention on Persistent Organic Pollutants. Land use regression (LUR) modelling can be used to integrate land use characteristics and other predictor variables (industrial emissions, traffic intensity, demographics, etc.) to describe or predict the distribution of air concentrations at unmeasured locations across a region or country. While LUR models are frequently applied to data-rich conventional air pollutants such as particulate matter, ozone, and nitrogen oxides, they are rarely applied to SVOCs.The MONET passive air sampling network (RECETOX, Masaryk University) continuously measures atmospheric SVOC levels across Czechia in monthly intervals. Using monitoring data from 29 MONET sites over a two-year period (2015–2017) and a variety of predictor variables, we developed LUR models to describe atmospheric levels and identify sources of polycyclic aromatic hydrocarbons (PAHs), polychlorinated biphenyls (PCBs) and DDT across the country. Strong and statistically significant (R2 > 0.6; p < 0.05) models were derived for PAH and PCB levels on a national scale. The PAH model retained three predictor variables – heating emissions represented by domestic fuel consumption, industrial PAH point sources, and the hill:valley index, a measure of site topography. The PCB model retained two predictor variables – site elevation, and secondary sources of PCBs represented by soil concentrations. These models were then applied to Czechia as a whole, highlighting the spatial variability of atmospheric SVOC levels, and providing a tool that can be used for further optimization of sampling network design, as well as evaluating potential human and environmental chemical exposures.

Land use regression modelling of community noise in São Paulo, Brazil

Noise pollution has negative health consequences, which becomes increasingly relevant with rapid urbanization. In low- and middle-income countries research on health effects of noise is hampered by scarce exposure data and noise maps. In this study, we developed land use regression (LUR) models to assess spatial variability of community noise in the Western Region of São Paulo, Brazil.We measured outdoor noise levels continuously at 42 homes once or twice for one week in the summer and the winter season. These measurements were integrated with various geographic information system variables to develop LUR models for predicting average A-weighted (dB(A)) day-evening-night equivalent sound levels (Lden) and night sound levels (Lnight). A supervised mixed linear regression analysis was conducted to test potential noise predictors for various buffer sizes and distances between home and noise source. Noise exposure levels in the study area were high with a site average Lden of 69.3 dB(A) ranging from 60.3 to 82.3 dB(A), and a site average Lnight of 59.9 dB(A) ranging from 50.7 to 76.6 dB(A). LUR models had a good fit with a R2 of 0.56 for Lden and 0.63 for Lnight in a leave-one-site-out cross validation. Main predictors of noise were the inverse distance to medium roads, count of educational facilities within a 400 m buffer, mean Normalized Difference Vegetation Index (NDVI) within a 100 m buffer, residential areas within a 50 m (Lden) or 25 m (Lnight) buffer and slum areas within a 400 m buffer. Our study suggests that LUR modelling with geographic predictor data is a promising and efficient approach for noise exposure assessment in low- and middle-income countries, where noise maps are not available.

Measurements of NOx and Development of Land Use Regression Models in an East-African City

Air pollution causes premature mortality and morbidity globally, but these adverse health effects occur over proportionately in low- and middle-income countries. Lack of both air pollution data and knowledge of its spatial distribution in African countries have been suggested to lead to an underestimation of health effects from air pollution. This study aims to measure nitrogen oxides (NOx), as well as nitrogen dioxide (NO2), to develop Land Use Regression (LUR) models in the city of Adama, Ethiopia. NOx and NO2 was measured at over 40 sites during six days in both the wet and dry seasons. Throughout the city, measured mean levels of NOx and NO2 were 29.0 µg/m3 and 13.1 µg/m3, respectively. The developed LUR models explained 68% of the NOx variances and 75% of the NO2. Both models included similar geographical predictor variables (related to roads, industries, and transportation administration areas) as those included in prior LUR models. The models were validated by using leave-one-out cross-validation and tested for spatial autocorrelation and multicollinearity. The performance of the models was good, and they are feasible to use to predict variance in annual average NOx and NO2 concentrations. The models developed will be used in future epidemiological and health impact assessment studies. Such studies may potentially support mitigation action and improve public health.

Development of land-use regression models to estimate particle mass and number concentrations in Taichung, Taiwan

Land-use regression (LUR) models have been used to estimate particle mass concentration (PMC), but few studies apply it to predict particle number concentration (PNC) at different sizes. This study aimed to determine both PMC and PNC throughout one year to establish predictive models in Taichung, Taiwan. The annual averages of PM10, PM2.5, and PM1 were 71 ± 46 μg/m3, 44 ± 35 μg/m3, and 32 ± 28 μg/m3, respectively. The PNC at size ranges of <0.5 μm, 0.5–1 μm, 1–2.5 μm, 2.5–10 μm, and ≥10 μm were 715098 ± 664879 counts/L, 29053 ± 30615 counts/L, 1009 ± 659 counts/L, 647 ± 347 counts/L, and 3 ± 3 counts/L, respectively. The model-explained variance (R2) values of PM10, PM2.5, and PM1 were 0.42, 0.53, and 0.51, respectively. The magnitude of the R2 values ranged from 0.31 to 0.50 for the PNC with the highest R2 between 0.5 and 1 μm. The differences between the model R2 and the leave-one-out cross-validation R2 ranged from 4% to 8% for PMC and from 3% to 10% for PNC. This study developed LUR models with moderate performance to estimate PMC and PNC at different sizes in an Asian metropolis. The built LUR models may be improved by combining with other open data to increase the predictive capacity.

Using Street View Imagery to Predict Street-Level Particulate Air Pollution.

Land-use regression (LUR) models are frequently applied to estimate spatial patterns of air pollution. Traditional LUR often relies on fixed-site measurements and GIS-derived variables with limited spatial resolution. We present an approach that leverages Google Street View (GSV) imagery to predict street-level particulate air pollution (i.e., black carbon [BC] and particle number [PN] concentrations). We developed empirical models based on mobile monitoring data and features extracted from ∼52 500 GSV images using a deep learning model. We tested theory- and data-driven feature selection methods as well as models using images within varying buffer sizes (50-2000 m). Compared to LUR models with traditional variables, our models achieved similar model performance using the street-level predictors while also identifying additional potential hotspots. Adjusted R2 (10-fold CV R2) with integrated feature selection was 0.57-0.64 (0.50-0.57) and 0.65-0.73 (0.61-0.66) for BC and PN models, respectively. Models using only features near the measurement locations (i.e., GSV images within 250 m) explained ∼50% of air pollution variability, indicating PN and BC are strongly affected by the street-level built environment. Our results suggest that GSV imagery, processed with computer vision techniques, is a promising data source to develop LUR models with high spatial resolution and consistent predictor variables across administrative boundaries.

Air Pollution Measurements and Land-Use Regression in Urban Sub-Saharan Africa Using Low-Cost Sensors—Possibilities and Pitfalls

Air pollution is recognized as the most important environmental factor that adversely affects human and societal wellbeing. Due to rapid urbanization, air pollution levels are increasing in the Sub-Saharan region, but there is a shortage of air pollution monitoring. Hence, exposure data to use as a base for exposure modelling and health effect assessments is also lacking. In this study, low-cost sensors were used to assess PM2.5 (particulate matter) levels in the city of Adama, Ethiopia. The measurements were conducted during two separate 1-week periods. The measurements were used to develop a land-use regression (LUR) model. The developed LUR model explained 33.4% of the variance in the concentrations of PM2.5. Two predictor variables were included in the final model, of which both were related to emissions from traffic sources. Some concern regarding influential observations remained in the final model. Long-term PM2.5 and wind direction data were obtained from the city’s meteorological station, which should be used to validate the representativeness of our sensor measurements. The PM2.5 long-term data were however not reliable. Means of obtaining good reference data combined with longer sensor measurements would be a good way forward to develop a stronger LUR model which, together with improved knowledge, can be applied towards improving the quality of health. A health impact assessment, based on the mean level of PM2.5 (23 µg/m3), presented the attributable burden of disease and showed the importance of addressing causes of these high ambient levels in the area.

Associations between modeled residential outdoor and measured personal exposure to ultrafine particles in four European study areas

Land use regression (LUR) models for Ultrafine Particles (UFP) have been developed to assess health effects of long-term average UFP exposure in epidemiological studies. Associations between LUR modeled residential outdoor and measured long-term personal exposure to UFP have never been evaluated, adding uncertainty in interpretation of epidemiological studies of UFP. Our aim was to assess how predictions of recently developed LUR models for UFP compared to measured average personal UFP exposure in four European areas.Personal UFP exposure was measured in 154 adults from Basel (Switzerland), Amsterdam and Utrecht (the Netherlands), Norwich (United Kingdom), and Turin (Italy). Subjects performed three 24-h exposure measurements by carrying a real-time monitor measuring particles between 10 and 300 nm (MiniDisc). Subjects reported whereabouts and indoor sources of UFP in questionnaires. In Basel and the Netherlands contemporaneously residential outdoor UFP concentrations were monitored. Area-specific LUR models were applied to model residential outdoor UFP concentrations. Associations between modeled and measured UFP concentrations were assessed with linear regression.LUR model predictions were significantly associated with median but not mean personal UFP exposures, likely because of the high impact of indoor peaks on mean personal exposures. Regression slopes (±se) combined for the four areas were 0.12 ± 0.04 for median and −0.06 ± 0.17 for mean personal exposure. The LUR model explained variance of the median personal exposure less than variance of residential outdoor measurements. Associations did not change when personal exposure was calculated for the time spent at home or when presence of indoor sources was incorporated in the regression models. Regression slopes for measured residential outdoor versus personal exposure were smaller for UFP (0.16 ± 0.04) than for simultaneously measured PM2.5 and soot (0.32 ± 0.10 and 0.43 ± 0.06).Our findings provide some support for the use of LUR models to estimate long-term exposure to ambient generated UFP in epidemiological studies.

Application of land use regression to assess exposure and identify potential sources in PM2.5, BC, NO2 concentrations

BackgroundUnderstanding spatial variation of air pollution is critical for public health assessments. Land Use Regression (LUR) models have been used increasingly for modeling small-scale spatial variation in air pollution concentrations. However, they have limited application in China due to the lack of spatially resolved data. ObjectiveBased on purpose-designed monitoring networks, this study developed LUR models to predict fine particulate matter (PM2.5), black carbon (BC) and nitrogen dioxide (NO2) exposure and to identify their potential outdoor-origin sources within an urban/rural region, using Taizhou, China as a case study. MethodTwo one-week integrated samples were collected at 30 PM2.5 (BC) sites and 45 NO2 sites in each two distinct seasons. Samples of 1/3 of the sites were collected simultaneously. Annual adjusted average was calculated and regressed against pre-selected GIS-derived predictor variables in a multivariate regression model. ResultsLUR explained 65% of the spatial variability in PM2.5, 78% in BC and 73% in NO2. Mean (±Standard Deviation) of predicted PM2.5, BC and NO2 exposure levels were 48.3 (±6.3) μg/m3, 7.5 (±1.4) μg/m3 and 27.3 (±8.2) μg/m3, respectively. Weak spatial corrections (Pearson r = 0.05–0.25) among three pollutants were observed, indicating the presence of different sources. Regression results showed that PM2.5, BC and NO2 levels were positively associated with traffic variables. The former two also increased with farm land use; and higher NO2 levels were associated with larger industrial land use. The three pollutants were correlated with sources at a scale of ≤5 km and even smaller scales (100–700m) were found for BC and NO2. ConclusionWe concluded that based on a purpose-designed monitoring network, LUR model can be applied to predict PM2.5, NO2 and BC concentrations in urban/rural settings of China. Our findings highlighted important contributors to within-city heterogeneity in outdoor-generated exposure, and indicated traffic, industry and agriculture may significantly contribute to PM2.5, NO2 and BC concentrations.

A Nationwide Land Use Regression Model for Ultrafine Particles

TPS 651: Air pollution exposure modeling 1, Exhibition Hall, Ground floor, August 27, 2019, 3:00 PM - 4:30 PM Background: Ultrafine Particles (UFP) are highly variable in space and time and are therefore difficult to predict on a large scale. Previous modelling studies have been published for single metropolitan areas. National models have not been developed which limits exposure for nationwide cohorts. Objectives: The aim of the paper is to create nationwide Land Use Regression (LUR) Model for UFP to be used in epidemiological studies. Method: We used a mobile platform to sample UFP concentrations across the Netherlands for a period of 16 months (2016-2017), with a total of 14.392 road segments sampled 3 times on average. Average concentrations per road segment (~43 seconds of data) and extracted GIS variables were used to develop LUR models using a supervised forward regression approach, as well as a deconvolution, LASSO and random forest algorithm. The deconvolution approach develops separate models for regional, urban and local scales. All models were tested on longer-term UFP measurements in 2014 at 42 sites with an average of 3 times 24 hours of measurements. Results: Out of the four prediction models, the LASSO model predicted the longer-term UFP measurements best (R2 =0.68). The supervised LUR model, deconvoluted approach and random forest algorithm predicted 51%, 62% and 56% of the variation in the longer-term UFP measurements respectively. Conclusions: We were able to develop nationwide models for UFP with good performance. The models will be applied in nationwide epidemiological studies. More modelling sophisticated procedures improved predictive power. The deconvoluted approach improved the standard linear model by distinguishing separate spatial scales.

Is a Land Use Regression Model Capable of Predicting the Cleanest Route to School?

Land Use Regression (LUR) modeling is a widely used technique to model the spatial variability of air pollutants in epidemiology. In this study, we explore whether a LUR model can predict home-to-school commuting exposure to black carbon (BC). During January and February 2019, 43 children walking to school were involved in a personal monitoring campaign measuring exposure to BC and tracking their home-to-school routes. At the same time, a previously developed LUR model for the study area was applied to estimate BC exposure on points along the route. Personal BC exposure varied widely with mean ± SD of 9003 ± 4864 ng/m3. The comparison between the two methods showed good agreement (Pearson’s r = 0.74, Lin’s Concordance Correlation Coefficient = 0.6), suggesting that LUR estimates are capable of catching differences among routes and predicting the cleanest route. However, the model tends to underestimate absolute concentrations by 29% on average. A LUR model can be useful in predicting personal exposure and can help urban planners in Milan to build a healthier city for schoolchildren by promoting less polluted home-to-school routes.

Mapping urban air quality using mobile sampling with low-cost sensors and machine learning in Seoul, South Korea

Recent studies have demonstrated that mobile sampling can improve the spatial granularity of land use regression (LUR) models. Mobile sampling campaigns deploying low-cost (<$300) air quality sensors could potentially offer an inexpensive and practical approach to measure and model air pollution concentration levels. In this study, we developed LUR models for street-level fine particulate matter (PM2.5) concentration levels in Seoul, South Korea. 169 h of data were collected from an approximately three week long campaign across five routes by ten volunteers sharing seven AirBeams, a low-cost ($250 per unit), smartphone-based particle counter, while geospatial data were extracted from OpenStreetMap, an open-source and crowd-generated geographical dataset. We applied and compared three statistical approaches in constructing the LUR models – linear regression (LR), random forest (RF), and stacked ensemble (SE) combining multiple machine learning algorithms – which resulted in cross-validation R2 values of 0.63, 0.73, and 0.80, respectively, and identification of several pollution ‘hotspots.’ The high R2 values suggest that study designs employing mobile sampling in conjunction with multiple low-cost air quality monitors could be applied to characterize urban street-level air quality with high spatial resolution, and that machine learning models could further improve model performance. Given this study design's cost-effectiveness and ease of implementation, similar approaches may be especially suitable for citizen science and community-based endeavors, or in regions bereft of air quality data and preexisting air monitoring networks, such as developing countries.

Annual, seasonal, and morning rush hour Land Use Regression models for black carbon in a school catchment area of Milan, Italy

IntroductionThe European Environment Agency has identified Northern Italy as one of the most polluted areas in Europe. Among air contaminants, black carbon (BC) has been identified as a sensitive marker of traffic related air pollution. This study aims to investigate the spatial distribution of BC in the catchment area of an elementary school of Milan, the biggest city in Northern Italy, using Land Use Regression (LUR) models and focusing especially on Morning Rush Hour (MRH). MethodsTwo recruitment campaigns were performed asking schoolchildren's parents and residents of the study area to host a monitoring site in their own dwellings. Finally, 34 monitoring sites and 1 reference site were sampled. BC was measured in two seasonal campaigns using eight micro-aethalometers. Six seasonal and annual LUR models were developed, 3 focused on MRH. ResultsOverall, median BC was 3247 and 1309 ng/m3 in the cold and warm season, respectively. In both seasons, there was a significant spatial variation between the monitoring sites. MRH values were higher than the daily values with median concentrations of 4227 and 2331 ng/m3, respectively. Developed LUR models showed that BC variability is well explained only by traffic variables; R2 ranged from 0.52 to 0.79 and from 0.65 to 0.81, for seasonal/annual and MRH LUR models respectively. Discussion.LUR models based on traffic variables explain most of the measured BC distribution variability for both warm and cold season. MRH represents a critical moment for BC during all the year, with an increase of 1000 ng/m3 respective to the daily median value and differences in magnitude according to location. Our results highlight that the mobility issue is one of the most important challenges to reduce air pollution in the city of Milan and this is of particular concern for elementary schoolchildren that commute to school during MRH.

Development of spatiotemporal models to predict ambient ozone and NOx concentrations in Tianjin, China

Nitrogen oxides (NOx) and ozone (O3) are important air pollutants that are associated with adverse health effects. Land-use regression (LUR) models have been widely developed to estimate air pollution concentrations. Due to data availability, however, such models are usually not applied in developing countries. We aimed to characterize NOx and O3 concentrations and develop LUR models to predict their spatial and temporal distributions using publicly-available data in Tianjin, a heavily polluted city in China. Seasonal samples were collected across Tianjin at 29 locations for O3 and 49 locations for NOx. Heavy-duty vehicle counts estimated from 0.5 m × 0.5 m satellite images correlated well with field-measured counts, thus supporting the use of high-resolution satellite images to assess vehicle traffic. Concentrations of NOx were highest in winter, while the opposite pattern was observed for O3. The majority of the variance in NOx was explained by season (36.2%) and heavy vehicle traffic (19.8%). For O3, the variance was explained by season (80.7%) in a pooled model, and by distance to roads (43.4%) and distance to coal plants (26.2%) in a summer model. Cross-validation showed reasonable practicability for NOx (R2 = 0.53 with field-measured heavy-duty vehicle count; R2 = 0.46 with satellite-based heavy-duty vehicle count) and O3 (R2 = 0.90 for pooled model; R2 = 0.70 for summer model) models. This study provides utility for researchers investigating air pollution in regions where field-measured vehicle traffic data are not available, as well as for policy makers and public health officials seeking to understand the sources and spatial distribution of air pollution in Tianjin.

Land Use Regression models for 60 volatile organic compounds: Comparing Google Point of Interest (POI) and city permit data

Land Use Regression (LUR) models of Volatile Organic Compounds (VOC) normally focus on land use (e.g., industrial area) or transportation facilities (e.g., roadway); here, we incorporate area sources (e.g., gas stations) from city permitting data and Google Point of Interest (POI) data to compare model performance. We used measurements from 50 community-based sampling locations (2013–2015) in Minneapolis, MN, USA to develop LUR models for 60 VOCs. We used three sets of independent variables: (1) base-case models with land use and transportation variables, (2) models that add area source variables from local business permit data, and (3) models that use Google POI data for area sources. The models with Google POI data performed best; for example, the total VOC (TVOC) model has better goodness-of-fit (adj-R2: 0.56; Root Mean Square Error [RMSE]: 0.32 μg/m3) as compared to the permit data model (0.42; 0.37) and the base-case model (0.26; 0.41). Area source variables were selected in over two thirds of models among the 60 VOCs at small-scale buffer sizes (e.g., 25 m–500 m). Our work suggests that VOC LUR models can be developed using community-based sampling and that models improve by including area sources as measured by business permit and Google POI data.

Comparison of model estimates from an intra-city land use regression model with a national satellite-LUR and a regional Bayesian Maximum Entropy model, in estimating NO2 for a birth cohort in Sydney, Australia

BackgroundMethods for estimating air pollutant exposures for epidemiological studies are becoming more complex in an effort to minimise exposure error and its associated bias. While land use regression (LUR) modelling is now an established method, there has been little comparison between LUR and other recent, more complex estimation methods. Our aim was to develop a LUR model to estimate intra-city exposures to nitrogen dioxide (NO2) for a Sydney cohort, and to compare those with estimates from a national satellite-based LUR model (Sat-LUR) and a regional Bayesian Maximum Entropy (BME) model. MethodsSatellite-based LUR and BME estimates were obtained using existing models. We used methods consistent with the European Study of Cohorts for Air Pollution Effects (ESCAPE) methodology to develop LUR models for NO2 and NOx. We deployed 46 Ogawa passive samplers across western Sydney during 2013/2014 and acquired data on land use, population density, and traffic volumes for the study area. Annual average NO2 concentrations for 2013 were estimated for 947 addresses in the study area using the three models: standard LUR, Sat-LUR and a BME model. Agreement between the estimates from the three models was assessed using interclass correlation coefficient (ICC), Bland-Altman methods and correlation analysis (CC). ResultsThe NO2 LUR model predicted 84% of spatial variability in annual mean NO2 (RMSE: 1.2 ppb; cross-validated R2: 0.82) with predictors of major roads, population and dwelling density, heavy traffic and commercial land use. A separate model was developed that captured 92% of variability in NOx (RMSE 2.3 ppb; cross-validated R2: 0.90). The annual average NO2 concentrations were 7.31 ppb (SD: 1.91), 7.01 ppb (SD: 1.92) and 7.90 ppb (SD: 1.85), for the LUR, Sat-LUR and BME models respectively. Comparing the standard LUR with Sat-LUR NO2 cohort estimates, the mean estimates from the LUR were 4% higher than the Sat-LUR estimates, and the ICC was 0.73. The Pearson's correlation coefficients (CC) for the LUR vs Sat-LUR values were r = 0.73 (log-transformed data) and r = 0.69 (untransformed data). Comparison of the NO2 cohort estimates from the LUR model with the BME blended model indicated that the LUR mean estimates were 8% lower than the BME estimates. The ICC for the LUR vs BME estimates was 0.73. The CC for the logged LUR vs BME estimates was r = 0.73 and for the unlogged estimates was r = 0.69. ConclusionsOur LUR models explained a high degree of spatial variability in annual mean NO2 and NOx in western Sydney. The results indicate very good agreement between the intra-city LUR, national-scale sat-LUR, and regional BME models for estimating NO2 for a cohort of children residing in Sydney, despite the different data inputs and differences in spatial scales of the models, providing confidence in their use in epidemiological studies.

Leveraging Google Place of Interest (POI) Data, Crowdsourcing, and Machine Learning to Predict Urban NO2 Concentrations for the Contiguous U.S.

Background: Land Use Regression (LUR) has been extensively applied to model intra-urban variations of air quality across the US. Predictor variables are typically derived from government sources (e.g., US Census) and do not include detailed information on land use or urban form. We developed LUR models that combine enhanced data on urban form and land use with machine learning to compare to traditional LUR models.Methods: Our LUR models are based on data from three sources: (1) a crowdsourced measure of urban form, (2) land use information derived from Google Place of Interest (POI) data, and (3) satellite observations of air quality. We developed models using US EPA NO2 monitoring data (n&amp;#61;426 locations) during 2010-2015. The measure of urban form (Local Climate Zones [LCZs]) was obtained using a crowdsourcing platform (Amazon Mechanical Turk) to allow multiple users (n&amp;#61;10) to classify satellite imagery into 17 categories of urban development (e.g., compact; sprawl). We web-scraped Google POI data (e.g., gas stations; restaurants) at various buffers (100-1,000 meters) to add detailed information on land use. We also included satellite observations of NO2 abundance. Using only these variables, we developed LUR models using both conventional (stepwise regression) and machine learning (e.g., bagging; gradient boosting).Results: Model fit was better for the machine learning models (gradient boosting; R2: 0.59; mean absolute error [MAE]: 2.46) as compared to the conventional models (i.e., stepwise regression; R2: 0.39 and MAE: 3.35) when using only the LCZ urban form measure and satellite NO2 abundance. Adding Google POI data improved model fit. For gradient boosting model-R2 increased to 0.65 (MAE: 2.28); model-R2 increased to 0.61 for stepwise regression (MAE: 2.34).Conclusions: Our work suggests that using generalizable data on urban form (e.g., Google POI; LCZs) to develop LUR models may produce similar results to models developed using conventional data sources.