Land-use Regression Techniques Research Articles

Implications of Nonstationary Effect on Geographically Weighted Total Least Squares Regression for PM2.5 Estimation.

Land use regression (LUR) models are used for high-resolution air pollution assessment. These models use independent parameters based on an assumption that these parameters are accurate and invariable; however, they are observational parameters derived from measurements or modeling. Therefore, the parameters are commonly inaccurate, with nonstationary effects and variable characteristics. In this study, we propose a geographically weighted total least squares regression (GWTLSR) to model air pollution under various traffic, land use, and meteorological parameters. To improve performance, the proposed model considers the dependent and independent variables as observational parameters. The GWTLSR applies weighted total least squares in order to take into account the variable characteristics and inaccuracies of observational parameters. Moreover, the proposed model considers the nonstationary effects of parameters through geographically weighted regression (GWR). We examine the proposed model’s capabilities for predicting daily PM2.5 concentration in Isfahan, Iran. Isfahan is a city with severe air pollution that suffers from insufficient data for modeling air pollution with conventional LUR techniques. The advantages of the model features, including consideration of the variable characteristics and inaccuracies of predictors, are precisely evaluated by comparing the GWTLSR model with ordinary least squares (OLS) and GWR models. The values estimated by the GWTLSR model during the spring and autumn are 0.84 and 0.91, respectively. The corresponding average values estimated by the OLS model during the spring and autumn are 0.74 and 0.69, respectively, and the values estimated by the GWR model are 0.76 and 0.70, respectively. The results demonstrate that the proposed functional model efficiently described the physical nature of the relationships among air pollutants and independent variables.

Comparative Performance of the LUR, ANN, and BME Techniques in the Multiscale Spatiotemporal Mapping of PM2.5 Concentrations in North China

Fine particle matter (PM2.5) has been receiving increasing attention by the government due to its considerable adverse health effects, especially in the north part of China. Even though a number of techniques of estimating PM2.5 exposure have been developed, what is still lacking is a systematic comparison of commonly used techniques based on classical statistics, artificial intelligence, and geostatistics. To address this need, the land use regression (LUR), the artificial neural networks (ANN), and the Bayesian maximum entropy (BME) techniques were all used to map the space–time PM2.5 concentration distribution in the highly polluted Jing-Jin-Ji region (Huabei plain of North China) during the period June 2015–May 2016. The tenfold cross-validation analysis and the entropic information theory were used to evaluate numerically the performance of the three techniques at monthly, seasonal, and annual time scales. Our results showed that the performance of each mapping technique was affected by the temporal scale and the degree of spatial heterogeneity. All three techniques were suitable for low temporal resolution (annual) datasets with low spatial variability. BME also showed a noticeable ability to analyze higher temporal resolution (monthly) datasets exhibiting high spatial heterogeneity. BME involved a single dependent variable (PM2.5) and generated complete (full-coverage) space–time PM2.5 maps, whereas LUR and ANN produced incomplete maps because of lacking independent variables (such as satellite data). Due to its self-learning feature, ANN showed better modeling performance than LUR and produced more informative maps. Overall, the ANN and BME techniques perform better than the LUR technique.

Capturing the urban canyon effect on particle number concentrations across a large road network using spatial analysis tools

Urban canyons increase overall air pollution concentrations at the street level. Yet, due to the difficulty in characterizing such geometric features over a regional area, they are rarely included in the evaluation of population exposures. This study combines measurements of ultrafine particles (UFP) with land-use regression techniques in order to capture the determinants of near-road concentrations of UFP while specifically accounting for urban canyon effects. Using the mean UFP concentration measured within a set of urban canyons (over 1000 road segments) as the dependent variable, we investigated the effects of a number of urban canyon descriptors (width, height, contiguity, vegetation, and road class) through a multivariate linear regression. Our models are adjusted for the effects of meteorology. The signs and values of the regression coefficients are qualitatively consistent with what is already known about urban canyons: increased building height (bheight = 0.004; 95% CI: 0.001, 0.007) along with decreased canyon width (bwidth = −0.007; 95% CI: −0.012, −0.002) serve to worsen the air quality therein, while increased tree cover (btrees = −1.762; 95% CI: −2.236, −1.289) improves air quality. We also observed that the difference in UFP levels on both sides of the canyon increased with increasing street width (bwidth = 0.113; 95% CI: 0.061, 0.165), and with wind direction more perpendicular to the street. We conclude that it is possible to include urban canyon effects in the development of air pollution exposure surfaces thus refining the assessment of population exposure to air pollution in urban microenvironments.

Monthly national satellite-based land-use regression of NO2 in the United States for 2000 - 2010

Background: Land use regression (LUR) is widely used for estimating within-urban variability in air pollution, typically at the urban level and for long-term averages. LUR techniques have recently been extended to national and continental scales. Aims: Our aim is a LUR for NO2 in the continental US that provides the excellent spatial resolution found in most LURs (~100-m scale), incorporates satellite estimates, covers 100% of US Census Blocks, and provides monthly average concentrations for one decade. Methods: Our work extends a previously-published national LUR for year-2006 in the US. We employ eleven years (2000-2010) of hourly NO2 measurements from US EPA monitors to calculate monthly scaling factors, which account for deviations relative to the reference year. We then create a spatially-varying (inverse-distance weighted) “scaling surface” for each month, to estimate monthly concentration deviations at non-measurement locations.Results: The resulting model captures, on average, 86% of the spatial variability and 72% of the temporal variability with low average bias (26%). Model performance is best at locations near monitors (e.g., spatial R2 = 0.90, mean bias 8% for the tertile with the lowest mean distance to the three nearest monitors). Urban and suburban locations perform moderately better than rural locations at predicting spatial (R2: 0.80 , 0.84, and 0.77, respectively) and temporal (R2: 0.78, 0.76, 0.63) variability and significantly better at predicting absolute concentrations (mean bias: 18%, 10%, 59%). Conclusions: Our approach reliably estimates monthly outdoor NO2 concentration at a spatial resolution capable of capturing within-urban and near-roadway variability in concentrations. We apply this technique to the ~8 million US Census blocks in the contiguous United States to provide a decade (2000 - 2010) of high-resolution monthly NO2 concentration estimates; these data are publicly available.

Socioeconomic status and exposure to outdoor NO2 and benzene in the Asturias INMA birth cohort, Spain

BackgroundIt is commonly assumed that low socioeconomic levels are associated with greater exposure to pollution, but this is not necessarily valid. Our goal was to examine how individual socioeconomic characteristics...

Principal component analysis optimization of a PM2.5 land use regression model with small monitoring network

The use of land-use regression (LUR) techniques for modeling small-scale variations of intraurban air pollution has been increasing in the last decade. The most appealing feature of LUR techniques is the economical monitoring requirements. In this study, principal component analysis (PCA) was employed to optimize an LUR model for PM2.5. The PM2.5 monitoring network consisted of 13 sites, which constrained the regression model to a maximum of one independent variable. An optimized surrogate of vehicle emissions was produced by PCA and employed as the predictor variable in the model. The vehicle emissions surrogate consisted of a linear combination of several traffic variables (e.g., vehicle miles traveled, speed, traffic demand, road length, and time) obtained from a road network used for traffic modeling. The vehicle-emissions surrogate produced by the PCA had a predictive capacity greater (R2=.458) than the traffic variable, Traffic Demand summarized for a 1km buffer, with best predictive capacity (R2=.341). The PCA-based method employed in this study was effective at increasing the fit of an ordinary LUR model by optimizing the utilization of a PM2.5 dataset from small-n monitoring network. In general, the method used can contribute to LUR techniques in two major ways: 1) by improving the predictive power of the input variable, by substituting a principal component for a single variable and 2) by creating an orthogonal set of predictor variables, and thus fulfilling the no colinearity assumption of the linear regression methods. The proposed PCA method, should be universally applicable to LUR methods and will expand their economical attractiveness.

Evaluation of land-use regression models used to predict air quality concentrations in an urban area

Cohort studies designed to estimate human health effects of exposures to urban pollutants require accurate determination of ambient concentrations in order to minimize exposure misclassification errors. However, it is often difficult to collect concentration information at each study subject location. In the absence of complete subject-specific measurements, land-use regression (LUR) models have frequently been used for estimating individual levels of exposures to ambient air pollution. The LUR models, however, have several limitations mainly dealing with extensive monitoring data needs and challenges involved in their broader applicability to other locations. In contrast, air quality models can provide high-resolution source–concentration linkages for multiple pollutants, but require detailed emissions and meteorological information. In this study, first we predicted air quality concentrations of PM 2.5, NO x, and benzene in New Haven, CT using hybrid modeling techniques based on CMAQ and AERMOD model results. Next, we used these values as pseudo-observations to develop and evaluate the different LUR models built using alternative numbers of (training) sites (ranging from 25 to 285 locations out of the total 318 receptors). We then evaluated the fitted LUR models using various approaches, including: 1) internal “Leave-One-Out-Cross-Validation” (LOOCV) procedure within the “training” sites selected; and 2) “Hold-Out” evaluation procedure, where we set aside 33–293 tests sites as independent datasets for external model evaluation. LUR models appeared to perform well in the training datasets. However, when these LUR models were tested against independent hold out (test) datasets, their performance diminished considerably. Our results confirm the challenges facing the LUR community in attempting to fit empirical response surfaces to spatially- and temporally-varying pollution levels using LUR techniques that are site dependent. These results also illustrate the potential benefits of enhancing basic LUR models by utilizing air quality modeling tools or concepts in order to improve their reliability or transferability.

Examining intra-urban variation in fine particle mass constituents using GIS and constrained factor analysis

Recent studies have used land use regression (LUR) techniques to explain spatial variability in exposures to PM 2.5 and traffic-related pollutants. Factor analysis has been used to determine source contributions to measured concentrations. Few studies have combined these methods, however, to construct and explain latent source effects. In this study, we derive latent source factors using confirmatory factor analysis constrained to non-negative loadings, and develop LUR models to predict the influence of outdoor sources on latent source factors using GIS-based measures of traffic and other local sources, central site monitoring data, and meteorology. We collected 3–4 day samples of nitrogen dioxide (NO 2) and PM 2.5 outside of 44 homes in summer and winter, from 2003 to 2005 in and around Boston, Massachusetts. Reflectance analysis, X-ray fluorescence spectroscopy (XRF), and high-resolution inductively-coupled plasma mass spectrometry (ICP-MS) were performed on particle filters to estimate elemental carbon (EC), trace element, and water-soluble metals concentrations. Within our constrained factor analysis, a five-factor model was optimal, balancing statistical robustness and physical interpretability. This model produced loadings indicating long-range transport, brake wear/traffic exhaust, diesel exhaust, fuel oil combustion, and resuspended road dust. LUR models largely corroborated factor interpretations through covariate significance. For example, ‘long-range transport’ was predicted by central site PM 2.5 and season; ‘brake wear/traffic exhaust’ and ‘resuspended road dust’ by traffic and residential density; ‘diesel exhaust’ by percent diesel traffic on nearest major road; and ‘fuel oil combustion’ by population density. Results suggest that outdoor residential PM 2.5 source contributions can be partially predicted using GIS-based terms, and that LUR techniques can support factor interpretation for source apportionment. Together, LUR and factor analysis facilitate source identification, assessment of spatial and temporal variability, and more refined source exposure assignment for evaluation of source contributions to health outcomes in epidemiological studies.