Abstract. For the purpose of providing reliable and robust air quality predictions, an air quality prediction system was developed for the main air quality criteria species in South Korea (PM10, PM2.5, CO, O3 and SO2). The main caveat of the system is to prepare the initial conditions (ICs) of the Community Multiscale Air Quality (CMAQ) model simulations using observations from the Geostationary Ocean Color Imager (GOCI) and ground-based monitoring networks in northeast Asia. The performance of the air quality prediction system was evaluated during the Korea-United States Air Quality Study (KORUS-AQ) campaign period (1 May–12 June 2016). Data assimilation (DA) of optimal interpolation (OI) with Kalman filter was used in this study. One major advantage of the system is that it can predict not only particulate matter (PM) concentrations but also PM chemical composition including five main constituents: sulfate (SO42-), nitrate (NO3-), ammonium (NH4+), organic aerosols (OAs) and elemental carbon (EC). In addition, it is also capable of predicting the concentrations of gaseous pollutants (CO, O3 and SO2). In this sense, this new air quality prediction system is comprehensive. The results with the ICs (DA RUN) were compared with those of the CMAQ simulations without ICs (BASE RUN). For almost all of the species, the application of ICs led to improved performance in terms of correlation, errors and biases over the entire campaign period. The DA RUN agreed reasonably well with the observations for PM10 (index of agreement IOA =0.60; mean bias MB =-13.54) and PM2.5 (IOA =0.71; MB =-2.43) as compared to the BASE RUN for PM10 (IOA =0.51; MB =-27.18) and PM2.5 (IOA =0.67; MB =-9.9). A significant improvement was also found with the DA RUN in terms of bias. For example, for CO, the MB of −0.27 (BASE RUN) was greatly enhanced to −0.036 (DA RUN). In the cases of O3 and SO2, the DA RUN also showed better performance than the BASE RUN. Further, several more practical issues frequently encountered in the air quality prediction system were also discussed. In order to attain more accurate ozone predictions, the DA of NO2 mixing ratios should be implemented with careful consideration of the measurement artifacts (i.e., inclusion of alkyl nitrates, HNO3 and peroxyacetyl nitrates – PANs – in the ground-observed NO2 mixing ratios). It was also discussed that, in order to ensure accurate nocturnal predictions of the concentrations of the ambient species, accurate predictions of the mixing layer heights (MLHs) should be achieved from the meteorological modeling. Several advantages of the current air quality prediction system, such as its non-static free-parameter scheme, dust episode prediction and possible multiple implementations of DA prior to actual predictions, were also discussed. These configurations are all possible because the current DA system is not computationally expensive. In the ongoing and future works, more advanced DA techniques such as the 3D variational (3DVAR) method and ensemble Kalman filter (EnK) are being tested and will be introduced to the Korean air quality prediction system (KAQPS).

The role of anthropogenic NOx emissions in secondary organic aerosol (SOA) production is not fully understood but is important for understanding the contribution of emissions to air quality. Here, we examine the role of organic nitrates (RONO2) in SOA formation over the Korean Peninsula during the Korea–United States Air Quality field study in Spring 2016 as a model for RONO2 aerosol in cities worldwide. We use aircraft-based measurements of the particle phase and total (gas + particle) RONO2 to explore RONO2 phase partitioning. These measurements show that, on average, one-fourth of RONO2 are in the condensed phase, and we estimate that ≈15% of the organic aerosol (OA) mass can be attributed to RONO2. Furthermore, we observe that the fraction of RONO2 in the condensed phase increases with OA concentration, evidencing that equilibrium absorptive partitioning controls the RONO2 phase distribution. Lastly, we model RONO2 chemistry and phase partitioning in the Community Multiscale Air Quality modeling system. We find that known chemistry can account for one-third of the observed RONO2, but there is a large missing source of semivolatile, anthropogenically derived RONO2. We propose that this missing source may result from the oxidation of semi- and intermediate-volatility organic compounds and/or from anthropogenic molecules that undergo autoxidation or multiple generations of OH-initiated oxidation.

A CNN+LSTM (Convolutional Neural Network + Long Short-Term Memory) based deep hybrid neural network was established for the citywide daily PM2.5 prediction in South Korea. The structural hyperparameters of the CNN+LSTM model were determined through comprehensive sensitivity tests. The input features were obtained from the ground observations and GFS forecast. The performance of CNN+LSTM was evaluated by comparison with PM2.5 observations and with the 3-D CTM (three-dimensional chemistry transport model)-predicted PM2.5. The newly developed hybrid model estimated more accurate ambient levels of PM2.5 compared to the 3-D CTM. For example, the error and bias of the CNN+LSTM prediction were 1.51 and 6.46 times smaller than those by 3D-CTM simulation. In addition, based on IOA (Index of Agreement), the accuracy of CNN+LSTM prediction was 1.10–1.18 times higher than the 3-D CTM-based prediction. The importance of input features was indirectly investigated by sequential perturbing input variables. The most important meteorological and atmospheric environmental features were geopotential height and previous day PM2.5. The obstacles of the current CNN+LSTM-based PM2.5 prediction were also discussed. The promising result of this study indicates that DNN-based models can be utilized as an effective tool for air quality prediction.

No abstract

This study proposes an improved approach for monitoring the spatial concentrations of hourly particulate matter less than 2.5 μm in diameter (PM2.5) via a deep neural network (DNN) using geostationary ocean color imager (GOCI) images and unified model (UM) reanalysis data over the Korean Peninsula. The DNN performance was optimized to determine the appropriate training model structures, incorporating hyperparameter tuning, regularization, early stopping, and input and output variable normalization to prevent training dataset overfitting. Near-surface atmospheric information from the UM was also used as an input variable to spatially generalize the DNN model. The retrieved PM2.5 from the DNN was compared with estimates from random forest, multiple linear regression, and the Community Multiscale Air Quality model. The DNN demonstrated the highest accuracy compared to that of the conventional methods for the hold-out validation (root mean square error (RMSE) = 7.042 μg/m3, mean bias error (MBE) = −0.340 μg/m3, and coefficient of determination (R2) = 0.698) and the cross-validation (RMSE = 9.166 μg/m3, MBE = 0.293 μg/m3, and R2 = 0.49). Although the R2 was low due to underestimated high PM2.5 concentration patterns, the RMSE and MBE demonstrated reliable accuracy values (<10 μg/m3 and 1 μg/m3, respectively) for the hold-out validation and cross-validation.