Adequate spatial and temporal estimates of NO concentrations are essential for proper prenatal exposure assessment. Here, we develop a spatiotemporal land use random forest (LURF) model of the monthly mean NO over four years in a metropolitan area of Japan. The overall objective is to obtain accurate NO estimates for use in prenatal exposure assessments. We use random forests to convey the non-linear relationship between NO 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 NO 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 R value of 0.79, which is better than that of the conventional land use regression model using linear regression (R of 0.73). We also evaluate the LURF model via a temporal and overall cross validation and obtain R 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 NO concentrations.

During the transboundary transport of anthropogenic heavy metals by mineral particles providing reaction sites, the divalent metal salt PbSO4 can be converted to PbCO3 in the presence of water. We carried out laboratory experiments to study the transformation process under various conditions by incorporating test particles comprising CaCO3 of a particulate mineral component, PbSO4, and NaCl. After the immersion of PbSO4 particles in contact with CaCO3 particles in a water droplet, the conversion of PbSO4 into PbCO3 was confirmed by the change in morphology of the original particles to stick or needle form; the percentages of the chemical forms relative to the total Pb were determined by X-ray absorption near edge structure (XANES) analysis. Approximately 60-80% of PbSO4 was converted to PbCO3 after 24 h. A small amount of stick particles was detected when NaCl particles attached to PbSO4/CaCO3 particles were exposed to air with a relative humidity (RH) of 80-90% for 24 h. XANES measurements of the samples revealed that the molar percentage of PbCO3 relative to the total Pb content was 4%. Field experiments were also conducted to determine the chemical forms of the Pb particles during the Kosa (Asian dust storm) event. Samples were collected from two remote sites in Japan and Korea. The mass size distribution of Pb aerosols collected in Japan was bimodal with two peaks in the coarse mode; the enrichment factor of Pb suggested that its source was anthropogenic. Pb L3 edge XANES measurements of both samples indicated that they had similar shapes. These measurements also indicated that the major Pb components for the samples collected in Japan were PbO, PbSO4 PbCl2, and PbCO3, with molar percentages of 44%, 30%, 21%, and 5%, respectively. No significant differences were found between the component ratios of the samples collected in Japan and Korea, suggesting that definite transformation did not occur during the transport of the Kosa particles from Korea to Japan. On the basis of these observations, we postulate that the transformation process either occurred mainly before the particles arrived at Korea or did not take place after the particles left continental Asia.

Air monitoring network design is a critical issue because monitoring stations should be allocated properly so that they adequately represent the concentrations in the domain of interest. Although the optimization methods using observations from existing monitoring networks are often applied to a network with a considerable number of stations, they are difficult to be applied to a sparse network or a network under development: there are too few observations to define an optimization criterion and the high number of potential monitor location combinations cannot be tested exhaustively. This paper develops a hybrid of genetic algorithm and simulated annealing to combine their power to search a big space and to find local optima. The hybrid algorithm as well as the two single algorithms are applied to optimize an air monitoring network of PM2.5, NO2 and O3 respectively, by minimization of the mean kriging variance derived from simulated values of a chemical transport model instead of observations. The hybrid algorithm performs best among the algorithms: kriging variance is on average about 4% better than for GA and variability between trials is less than 30% compared to SA. The optimized networks for the three pollutants are similar and maps interpolated from the simulated values at these locations are close to the original simulations (RMSE below 9% relative to the range of the field). This also holds for hourly and daily values although the networks are optimized for annual values. It is demonstrated 3 that the method using the hybrid algorithm and the model simulated values for the calculation of the mean kriging variance is of benefit to the optimization of air monitoring networks.

The application of regression kriging to air pollutants in Japan was examined for the purpose of providing a practical method to obtain a spatial distribution with sufficient accuracy and a high spatial resolution of 1 × 1 km. We used regulatory air monitoring data from the years 2009 and 2010. Predictor variables at 1 × 1 km resolution were prepared from various datasets to perform regression kriging. The prediction performance was assessed by indicators, including root mean squared error (RMSE) and R 2 , calculated from the leave-one-out cross validation results, and was compared to the results obtained from a linear regression method, often referred to as land use regression (LUR). Regression kriging wellexplained the spatial variability of NO 2 , with R 2 values of 0.77 and 0.78. Ozone (O 3 ) was moderately explained, with R 2 values of 0.52 and 0.66. The reason for this difference in performance between NO 2 and O 3 might be the characteristics of these pollutants -primary or secondary. Regression kriging outperformed the linear regression method with regard to RMSE and R 2 . The performance of regression kriging in this work was comparable to that found in previous studies. The results indicate that regression kriging is a practical procedure that can be applied for the prediction of the spatial distribution of air pollutants in Japan, with sufficient accuracy and a high spatial resolution.

No abstract