Annual average land-use regression (LUR) models have been widely used to assess spatial patterns of air pollution exposures. However, they fail to capture diurnal variability in air pollution and consequently might result in biased dynamic exposure assessments. In this study we aimed to model average hourly concentrations for two major pollutants, NO2 and PM2.5, for the Netherlands using the LUR algorithm. We modelled the spatial variation of average hourly concentrations for the years 2016-2019 combined, for two seasons, and for two weekday types. Two modelling approaches were used, supervised linear regression (SLR) and random forest (RF). The potential predictors included population, road, land use, satellite retrievals, and chemical transport model pollution estimates variables with different buffer sizes. We also temporally adjusted hourly concentrations from a 2019 annual model using the hourly monitoring data, to compare its performance with the hourly modelling approach. The results showed that hourly NO2 models performed overall well (5-fold cross validation R2 = 0.50-0.78), while the PM2.5 performed moderately (5-fold cross validation R2 = 0.24-0.62). Both for NO2 and PM2.5 the warm season models performed worse than the cold season ones, and the weekends' worse than weekdays'. The performance of the RF and SLR models was similar for both pollutants. For both SLR and RF, variables with larger buffer sizes representing variation in background concentrations, were selected more often in the weekend models compared to the weekdays, and in the warm season compared to the cold one. Temporal adjustment of annual average models performed overall worse than both modelling approaches (NO2 hourly R2 = 0.35-0.70; PM2.5 hourly R2 = 0.01-0.15). The difference in model performance and selection of variables across hours, seasons, and weekday types documents the benefit to develop independent hourly models when matching it to hourly time activity data.
Mobile monitoring campaigns have effectively captured spatial hyperlocal variations in long-term average concentrations of regulated and unregulated air pollutants. However, their application in estimating spatiotemporally varying maps has rarely been investigated. Tackling this gap, we investigated whether mobile measurements can assess long-term average nitrogen dioxide (NO2) concentrations for each hour of the day. Using mobile NO2 data monitored for 10 months in Amsterdam, we examined the performance of two spatiotemporal land use regression (LUR) methods, Spatiotemporal-Kriging and GTWR (Geographical and Temporal Weighted Regression), alongside two classical spatial LUR models developed separately for each hour. We found that mobile measurements follow the general pattern of fixed-site measurements, but with considerable deviations (indicating collection uncertainty). Leveraging heterogeneous spatiotemporal autocorrelations, GTWR smoothed these deviations and achieved an overall performance of an R2 of 0.49 and a Mean Absolute Error of 6.33 µg/m3, validated by long-term fixed-site measurements (out-of-sample). The other models tested were more affected by the collection uncertainty. We highlighted that the spatiotemporal variations captured in mobile measurements can be used to reconstruct long-term average hourly air pollution maps. These maps facilitate dynamic exposure assessments considering spatiotemporal human activity patterns.
Detailed spatial models of monthly air pollution levels at a very fine spatial resolution (25 m) can help facilitate studies to explore critical time-windows of exposure at intermediate term. Seasonal changes in air pollution may affect both levels and spatial patterns of air pollution across Europe. We built Europe-wide land-use regression (LUR) models to estimate monthly concentrations of regulated air pollutants (NO2, O3, PM10 and PM2.5) between 2000 and 2019. Monthly average concentrations were collected from routine monitoring stations. Including both monthly-fixed and -varying spatial variables, we used supervised linear regression (SLR) to select predictors and geographically weighted regression (GWR) to estimate spatially-varying regression coefficients for each month. Model performance was assessed with 5-fold cross-validation (CV). We also compared the performance of the monthly LUR models with monthly adjusted concentrations. Results revealed significant monthly variations in both estimates and model structure, particularly for O3, PM10, and PM2.5. The 5-fold CV showed generally good performance of the monthly GWR models across months and years (5-fold CV R2: 0.31-0.66 for NO2, 0.4-0.79 for O3, 0.4-0.78 for PM10, 0.46-0.87 for PM2.5). Monthly GWR models slightly outperformed monthly-adjusted models. Correlations between monthly GWR model were generally moderate to high (Pearson correlation >0.6). In conclusion, we are the first to develop robust monthly LUR models for air pollution in Europe. These monthly LUR models, at a 25 m spatial resolution, enhance epidemiologists to better characterize Europe-wide intermediate-term health effects related to air pollution, facilitating investigations into critical exposure time windows in birth cohort studies.
Mobile air quality measurements are collected typically for several seconds per road segment and in specific timeslots (e.g., working hours). These short-term and on-road characteristics of mobile measurements become the ubiquitous shortcomings of applying land use regression (LUR) models to estimate long-term concentrations at residential addresses. This issue was previously found to be mitigated by transferring LUR models to the long-term residential domain using routine long-term measurements in the studied region as the transfer target (local scale). However, long-term measurements are generally sparse in individual cities. For this scenario, we propose an alternative by taking long-term measurements collected over a larger geographical area (global scale) as the transfer target and local mobile measurements as the source (Global2Local model). We empirically tested national, airshed countries (i.e., national plus neighboring countries) and Europe as the global scale in developing Global2Local models to map nitrogen dioxide (NO2) concentrations in Amsterdam. The airshed countries scale provided the lowest absolute errors, and the Europe-wide scale had the highest R2. Compared to a "global" LUR model (trained exclusively with European-wide long-term measurements), and a local mobile LUR model (using mobile data from Amsterdam only), the Global2Local model significantly reduced the absolute error of the local mobile LUR model (root-mean-square error, 6.9 vs 12.6 µg/m3) and improved the percentage explained variances compared to the global model (R2, 0.43 vs 0.28, assessed by independent long-term NO2 measurements in Amsterdam, n = 90). The Global2Local method improves the generalizability of mobile measurements in mapping long-term residential concentrations with a fine spatial resolution, which is preferred in environmental epidemiological studies.
Air Pollutants , Air Pollution , Air Pollutants/analysis , Particulate Matter/analysis , Nitrogen Dioxide/analysis , Environmental Monitoring/methods , Air Pollution/analysis , Machine Learning
Previous European land-use regression (LUR) models assumed fixed linear relationships between air pollution concentrations and predictors such as traffic and land use. We evaluated whether including spatially-varying relationships could improve European LUR models by using geographically weighted regression (GWR) and random forest (RF). We built separate LUR models for each year from 2000 to 2019 for NO2, O3, PM2.5 and PM10 using annual average monitoring observations across Europe. Potential predictors included satellite retrievals, chemical transport model estimates and land-use variables. Supervised linear regression (SLR) was used to select predictors, and then GWR estimated the potentially spatially-varying coefficients. We developed multi-year models using geographically and temporally weighted regression (GTWR). Five-fold cross-validation per year showed that GWR and GTWR explained similar spatial variations in annual average concentrations (average R2 = NO2: 0.66; O3: 0.58; PM10: 0.62; PM2.5: 0.77), which are better than SLR (average R2 = NO2: 0.61; O3: 0.46; PM10: 0.51; PM2.5: 0.75) and RF (average R2 = NO2: 0.64; O3: 0.53; PM10: 0.56; PM2.5: 0.67). The GTWR predictions and a previously-used method of back-extrapolating 2010 model predictions using CTM were overall highly correlated (R2 > 0.8) for all pollutants. Including spatially-varying relationships using GWR modestly improved European air pollution annual LUR models, allowing time-varying exposure-health risk models.
