Combining land use regression models and fixed site monitoring to reconstruct spatiotemporal variability of NO2 concentrations over a wide geographical area

Geospatial artificial intelligence for estimating daytime and nighttime nitrogen dioxide concentration variations in Taiwan: A spatial prediction model

Nitrogen dioxide (NO2) is a major air pollutant primarily emitted from traffic and industrial activities, posing health risks. However, current air pollution models often underestimate exposure risks by neglecting the bimodal pattern of NO2 levels throughout the day. This study aimed to address this gap by developing ensemble mixed spatial models (EMSM) using geo-artificial intelligence (Geo-AI) to examine the spatial and temporal variations of NO2 concentrations at a high resolution of 50m. These EMSMs integrated spatial modelling methods, including kriging, land use regression, machine learning, and ensemble learning. The models utilized 26 years of observed NO2 measurements, meteorological parameters, geospatial layers, and social and season-dependent variables as representative of emission sources. Separate models were developed for daytime and nighttime periods, which achieved high reliability with adjusted R2 values of 0.92 and 0.93, respectively. The study revealed that mean NO2 concentrations were significantly higher at nighttime (9.60 ppb) compared to daytime (5.61 ppb). Additionally, winter exhibited the highest NO2 levels regardless of time period. The developed EMSMs were utilized to generate maps illustrating NO2 levels pre and during COVID restrictions in Taiwan. These findings could aid epidemiological research on exposure risks and support policy-making and environmental planning initiatives.

Development of Phase and Seasonally Dependent Land-Use Regression Models to Predict Atmospheric PAH Levels

Polycyclic aromatic hydrocarbons (PAHs) are an important class of pollutants in China. The land use regression (LUR) model has been used to predict the selected PAH concentrations and screen the key influencing factors. However, most previous studies have focused on particle-associated PAHs, and research on gaseous PAHs was limited. This study measured representative PAHs in both gaseous phases and particle-associated during the windy, non-heating and heating seasons from 25 sampling sites in different areas of Taiyuan City. We established separate prediction models of 15 PAHs. Acenaphthene (Ace), Fluorene (Flo), and benzo [g,h,i] perylene (BghiP) were selected to analyze the relationship between PAH concentration and influencing factors. The stability and accuracy of the LUR models were quantitatively evaluated using leave-one-out cross-validation. We found that Ace and Flo models show good performance in the gaseous phase (Ace: adj. R² = 0.14-0.82; Flo: adj. R² = 0.21-0.85), and the model performance of BghiP is better in the particle phase (adj. R² = 0.20-0.42). Additionally, better model performance was observed in the heating season (adj R² = 0.68-0.83) than in the non-heating (adj R² = 0.23-0.76) and windy seasons (adj R² = 0.37-0.59). Those gaseous PAHs were highly affected by traffic emissions, elevation, and latitude, whereas BghiP was affected by point sources. This study reveals the strong seasonal and phase dependence of PAH concentrations. Building separate LUR models in different phases and seasons improves the prediction accuracy of PAHs.

National Land Use Regression Model for NO2 Using Street View Imagery and Satellite Observations

Land use regression (LUR) models are widely applied to estimate intra-urban air pollution concentrations. National-scale LURs typically employ predictors from multiple curated geodatabases at neighborhood scales. In this study, we instead developed national NO₂ models relying on innovative street-level predictors extracted from Google Street View [GSV] imagery. Using machine learning (random forest), we developed two types of models: (1) GSV-only models, which use only GSV features, and (2) GSV + OMI models, which also include satellite observations of NO₂. Our results suggest that street view imagery alone may provide sufficient information to explain NO₂ variation. Satellite observations can improve model performance, but the contribution decreases as more images are available. Random 10-fold cross-validation R² of our best models were 0.88 (GSV-only) and 0.91 (GSV + OMI)─a performance that is comparable to traditional LUR approaches. Importantly, our models show that street-level features might have the potential to better capture intra-urban variation of NO₂ pollution than traditional LUR. Collectively, our findings indicate that street view image-based modeling has great potential for building large-scale air quality models under a unified framework. Toward that goal, we describe a cost-effective image sampling strategy for future studies based on a systematic evaluation of image availability and model performance.

Assessing the intra-urban variability of nitrogen oxides and ozone across a highly heterogeneous urban area

High-resolution air quality maps are critical towards assessing and understanding exposures to elevated air pollution in dense urban areas. However, these surfaces are rarely available in low- and middle-income countries that suffer from some of the highest air pollution levels worldwide. In this study, we make use of land use regressions (LURs) to generate annual and seasonal, high-resolution nitrogen dioxide (NO₂), nitrogen oxides (NO_x), and ozone (O₃) exposure surfaces for the Greater Beirut Area (GBA) in Lebanon. NO₂, NO_x and O₃ concentrations were monitored using passive samplers that were deployed at 55 pre-defined monitoring locations. The average annual concentrations of NO₂, NO_x, and O₃ across the GBA were 36.0, 89.7, and 26.9 ppb, respectively. Overall, the performance of the generated models was appropriate, with low biases, high model robustness, and acceptable R² values that ranged between 0.66 and 0.73 for NO₂, 0.56 and 0.60 for NO_x, and 0.54 and 0.65 for O₃. Traffic-related emissions as well as the operation of a fossil-fuel power plant were found to be the main contributors to the measured NO₂ and NO_x levels in the GBA, whereas they acted as sinks for O₃ concentrations. No seasonally significant differences were found for the NO₂ and NO_x pollution surfaces; as their seasonal and annual models were largely similar (Pearson's r > 0.85 for both pollutants). On the other hand, seasonal O₃ pollution surfaces were significantly different. The model results showed that around 99% of the population of the GBA were exposed to NO₂ levels that exceeded the World Health Organization defined annual standard.

Effects of low exposure to traffic related air pollution on childhood asthma onset by age 10 years

Although NO₂, a major traffic related air pollutant, has been associated with onset of childhood asthma, young children may be more susceptible to traffic related air pollution exposure compared to other individuals. We linked data from National Longitudinal Survey of Children and Youths Cycle 1-5 (1994-2003) and the National Air Pollution Surveillance Program to determine the association between NO₂ exposure and either early or late onset childhood asthma phenotypes. Children diagnosed with asthma from age 0-3 were defined as having early onset asthma. Children diagnosed with asthma from age 4-9 were defined as having late onset asthma. Mean NO₂ exposure for each quartile was 6.31 ppb, 9.45 ppb, 11.83 ppb, and 17.9 ppb. Higher levels of NO₂ exposure were more strongly associated with early childhood asthma (Quartile 3 OR: 2.11, 95% CI: 1.29, 3.44, Quartile 4 OR: 2.16, 95% CI: 1.27, 3.68) compared to the lowest level of NO₂ exposure (Quartile 1). No such association was observed with risk of late childhood asthma onset. Asthma susceptibility to NO₂ exposure may vary with the childhood developmental stage, and young children may be susceptible to NO₂ exposure at levels well below national and international guidelines. Our study emphasizes the importance of considering the timing of childhood asthma onset in future studies and confirms the increased risk of early onset of childhood asthma associated even with relatively low NO₂ exposure levels.

A land use regression model of nitrogen dioxide and fine particulate matter in a complex urban core in Lanzhou, China

BACKGROUND

Land use regression (LUR) models have been widely used to estimate air pollution exposures at high spatial resolution. However, few LUR models were developed for rapidly developing urban cores, which have substantially higher densities of population and built-up areas than the surrounding areas within a city's administrative boundary. Further, few studies incorporated vertical variations of air pollution in exposure assessment, which might be important to estimate exposures for people living in high-rise buildings.

OBJECTIVE

A LUR model was developed for the urban core of Lanzhou, China, along with a model of vertical concentration gradients in high-rise buildings.

METHODS

In each of four seasons in 2016-2017, NO₂ was measured using Ogawa badges for 2 weeks at 75 ground-level sites. PM_2.5 was measured using DataRAM for shorter time intervals at a subset (N = 38) of the 75 sites. Vertical profile measurements were conducted on 9 stories at 2 high-rise buildings (N = 18), with one building facing traffic and another facing away from traffic. The average seasonal concentrations of NO₂ and PM_2.5 at ground level were regressed against spatial predictors, including elevation, population, road network, land cover, and land use. The vertical variations were investigated and linked to ground-level predictions with exponential models.

RESULTS

We developed robust LUR models at the ground level for estimated annual averages of NO₂ (R²: 0.71, adjusted R²: 0.67, and Leave-One-Out Cross Validation (LOOCV) R²: 0.64) and PM_2.5 (R²: 0.77, adjusted R²: of 0.73, and LOOCV R²: 0.67) in the urban core of Lanzhou, China. The LUR models for the estimated seasonal averages of NO₂ showed similar patterns. Vertical variation of NO₂ and PM_2.5 differed by windows orientation with respect to traffic, by season or by time of a day. Vertical variation functions incorporated the ground-level LUR predictions, in a form that could allow for exposure assessment in future epidemiological investigations.

CONCLUSIONS

Ground-level NO₂ and PM_2.5 showed substantial spatial variations, explained by traffic and land use patterns. Further, vertical variation of air pollution levels is significant under certain conditions, suggesting that exposure misclassification could occur with traditional LUR that ignores vertical variation. More studies are needed to fully characterize three-dimensional concentration patterns to accurately estimate air pollution exposures for residents in high-rise buildings, but our LUR models reinforce that concentration heterogeneity is not captured by the limited government monitors in the Lanzhou urban area.

Mapping and Statistical Analysis of NO2 Concentration for Local Government Air Quality Regulation

With the growing interest in healthy living worldwide, there has been an increasing demand for more accurate measurements of the concentrations of air pollutants such as NO2. In particular, analyzing the characteristics and sources of air pollutants by region could improve the effectiveness of environmental policies applied in accordance with the environmental characteristics of individual regions. In this study, a detailed nationwide NO2 concentration map was generated using the cokriging interpolation technique, which integrates ground observations and satellite image data. The root-mean-square standardized (RMSS) error for this technique was close to 1, which indicates high accuracy. Using spatially interpolated NO2 concentration data, an administrative unit map was generated. When comparing the data for four NO2 data sources (observation data, satellite image data, detailed national data interpolated using cokriging, and NO2 concentrations averaged by an administrative unit based on the interpolated NO2 concentration data), the average concentrations were highest for remote sensing data. Land use regression (LUR) models of urban and non-urban regions were then developed to analyze the characteristics of the NO2 concentration by region using NO2 concentrations for the administrative units.

Assessing the accuracy of long-term air pollution estimates produced with temporally adjusted short-term observations from unstructured sampling

More commonly air pollution observations are obtained with unstructured monitoring, where either a research grade monitor or low-cost sensor is irregularly relocated throughout the study area. This unstructured data is commonly observed in community science programs. Often the objective is to apply these data to estimate a long-term concentration, which is achieved using a temporal adjustment to correct for the irregular sampling. Temporal adjustments leverage information from a stationary continuous reference monitor, in combination with short-term monitoring data, to estimate long-term pollutant concentrations. We assess the performance of temporal adjustment approaches to predict long-term pollutant concentrations using data representing unstructured sampling. A series of monitoring campaigns are simulated from air pollution data obtained from regulatory monitoring networks in four different cities (Paris, France; Taipei, Taiwan; Toronto, Canada; and Vancouver, Canada) for eight different pollutants (CO, NO, NO_x, NO₂, O₃, PM₁₀, PM_2.5, and SO₂). These simulated campaigns have randomized monitoring locations and sampling times to simulate the irregular nature of crowd sourced or mobile monitoring data. The number of consecutive samples reported, and selection of the reference monitor used to adjust observations, are varied in this study. The accuracy of estimates is assessed by comparing the estimated long-term concentration to the observed long-term concentration from the complete regulatory monitoring dataset. This study found that a common temporal adjustment applied in research performed significantly worse than other adjustments including a Naïve Temporal Approach where no data adjustment occurred. Increasing the sample size improved the accuracy of estimates, which showed decreasing benefit with increased sample lengths. Lastly, controlling for land use conditions of the reference monitor did not consistently improve the long-term estimates, which suggests that land use pairing of mobile and reference monitors does not significantly influence the predictive power of temporal adjustment approaches. Temporal adjustments can reduce the error in long-term concentration estimates of air pollution using incomplete data, but this benefit cannot be assumed across all approaches, pollutants or sampling programs.

GIS Multi-Criteria Analysis by Ordered Weighted Averaging (OWA): Toward an Integrated Citrus Management Strategy

This study proposes a site location assessment model for citrus cropland using multi-criteria evaluation (MCE) and the combination of a set of factors for suitability mapping and delineating the suitable areas for citrus production in Ramsar, Iran. It defines an incorporated method for the suitability mapping of the most appropriate sites for citrus cultivars with an emphasis on the multi-criteria decision analysis (MCDA) process. The combination of geographic information system (GIS) and a modified version of the analytic hierarchy process (AHP) based on the ordered weighted averaging (OWA) technique is also emphasized. The OWA is based on two principles, namely: the weights of relative criterion significance and the order weights. Therefore, the participatory technique was employed to outline the set of standards and the important criterion. The results derived from the GIS–OWA technique indicate that the cultivation of citrus is feasible only in limited areas, which make up 6.7% of the total area near the Caspian Sea. This investigation has shown that the GIS–OWA model can be integrated into MCDA to select the optimal site for citrus production. The present research highlights how multi-criteria in GIS can play a considerable role in decision making for evaluating the suitability of selected sites for citrus production.

Spatiotemporal land use random forest model for estimating metropolitan NO2 exposure in Japan

Adequate spatial and temporal estimates of NO₂ concentrations are essential for proper prenatal exposure assessment. Here, we develop a spatiotemporal land use random forest (LURF) model of the monthly mean NO₂ over four years in a metropolitan area of Japan. The overall objective is to obtain accurate NO₂ estimates for use in prenatal exposure assessments. We use random forests to convey the non-linear relationship between NO₂ concentrations and predictor variables, and compare the prediction accuracy with that of a linear regression. In addition, we include the distance decay effect of emission sources on NO₂ concentrations for more efficient model construction. The prediction accuracy of the LURF model is evaluated through a leave-one-monitor-out cross validation. We obtain a high R² value of 0.79, which is better than that of the conventional land use regression model using linear regression (R² of 0.73). We also evaluate the LURF model via a temporal and overall cross validation and obtain R² values of 0.84 and 0.92, respectively. We successfully integrate temporal and spatial components into our model, which exhibits higher accuracy than spatial models constructed individually for each month. Our findings illustrate the advantage of using a LURF to model the spatiotemporal variability of NO₂ concentrations.

A land use regression model for explaining spatial variation in air pollution levels using a wind sector based approach

Estimating pollutant concentrations at a local and regional scale is essential in environmental and health policy decision making. Here we present a novel land use regression (LUR) modelling methodology that exploits the high temporal resolution of fixed-site monitoring (FSM) to produce a national-scale air quality model for the key pollutant NO₂. The methodology partitions concentration time series from a national FSM network into wind-dependent sectors or "wedges". A LUR model is derived using predictor variables calculated within the directional wind sectors, and compared against the long-term average concentrations within each sector. Validation results, based on 15 FSM training sites, show that the model captured 78% of the spatial variability in NO₂ across the Republic of Ireland. This compares favourably to traditional LUR models based on purpose-designed monitoring campaigns despite using approximately half the number of monitoring points. Results also demonstrate the value of incorporating the relative position of emission source and receptor into the empirical LUR model structure. We applied the model at a high-resolution across the Republic of Ireland to enable applications such as the study of environmental exposure and human health, assessing representativeness of air quality monitoring networks and informing environmental management and policy makers. While the study focuses on Ireland, the methodology also has potential applicability for other criteria pollutants where appropriate FSM and meteorological networks exist.

Combining Measurements from Mobile Monitoring and a Reference Site To Develop Models of Ambient Ultrafine Particle Number Concentration at Residences

Significant spatial and temporal variation in ultrafine particle (UFP; <100 nm in diameter) concentrations creates challenges in developing predictive models for epidemiological investigations. We compared the performance of land-use regression models built by combining mobile and stationary measurements (hybrid model) with a regression model built using mobile measurements only (mobile model) in Chelsea and Boston, MA (USA). In each study area, particle number concentration (PNC; a proxy for UFP) was measured at a stationary reference site and with a mobile laboratory driven along a fixed route during an ∼1-year monitoring period. In comparing PNC measured at 20 residences and PNC estimates from hybrid and mobile models, the hybrid model showed higher Pearson correlations of natural log-transformed PNC ( r = 0.73 vs 0.51 in Chelsea; r = 0.74 vs 0.47 in Boston) and lower root-mean-square error in Chelsea (0.61 vs 0.72) but no benefit in Boston (0.72 vs 0.71). All models overpredicted log-transformed PNC by 3-6% at residences, yet the hybrid model reduced the standard deviation of the residuals by 15% in Chelsea and 31% in Boston with better tracking of overnight decreases in PNC. Overall, the hybrid model considerably outperformed the mobile model and could offer reduced exposure error for UFP epidemiology.

Simplifying field-scale assessment of spatiotemporal changes of soil salinity

Monitoring soil salinity (EC_e) is important for planning and implementing agronomic and irrigation practices. Salinity can be measured through soil sampling directed by geospatial measurements of apparent soil electrical conductivity (EC_a). Using data from a long-term (1999-2012) monitoring study at a 32.4-ha saline field located in California, USA, two established field-scale approaches to map and monitor soil salinity using EC_a are reviewed: one that relies on a single EC_a survey to identify locations that can be repeatedly sampled to infer the frequency distribution of EC_e; and another based on repeated EC_a surveys that are calibrated, each time, to EC_e estimation using ground-truth data from soil samples. The reviewed approaches are very accurate and reliable, but require extensive soil sampling. Subsequently, we propose a novel approach - temporal analysis of covariance (t-ANOCOVA) modeling - that results in accurate spatiotemporal salinity estimations using EC_a surveys with a significant reduction in the number of soil samples needed for calibration of EC_a to EC_e. In this modeling framework, the EC_e-EC_a relationship is described with a log-transformed linear function. The regression slope indicates the magnitude of the contribution of EC_e to EC_a and is assumed to remain constant over time, while the intercept represents the secondary factors influencing EC_a that are not related to EC_e (e.g., soil tillage). Once the t-ANOCOVA slope is established for a field, in subsequent surveys as few as three soil samples are used to estimate a time-specific t-ANOCOVA intercept so that EC_a measurements can be converted to EC_e estimations. Our results suggest that this approach is reliable at low salinity values (i.e., where common crops can grow). The t-ANOCOVA approach requires further validation before real-world implementations, but represents a significant step towards the use of EC_a mobile sensor technology for inexpensive soil salinity monitoring at high temporal resolution.