• Journal of Korea Multimedia Society
• /
• v.24 no.10
• /
• pp.1414-1424
• /
• 2021

In this paper, the importance of input factors of a DNN (Deep Neural Network) PM_2.5 forecasting model using LRP(Layer-wise Relevance Propagation) is analyzed, and forecasting performance is improved. Input factor importance analysis is performed by dividing the learning data into time and PM_2.5 concentration. As a result, in the low concentration patterns, the importance of weather factors such as temperature, atmospheric pressure, and solar radiation is high, and in the high concentration patterns, the importance of air quality factors such as PM_2.5, CO, and NO₂ is high. As a result of analysis by time, the importance of the measurement factors is high in the case of the forecast for the day, and the importance of the forecast factors increases in the forecast for tomorrow and the day after tomorrow. In addition, date, temperature, humidity, and atmospheric pressure all show high importance regardless of time and concentration. Based on the importance of these factors, the LRP_DNN prediction model is developed. As a result, the ACC(accuracy) and POD(probability of detection) are improved by up to 5%, and the FAR(false alarm rate) is improved by up to 9% compared to the previous DNN model.

• Journal of Korean Society for Atmospheric Environment
• /
• v.34 no.2
• /
• pp.306-320
• /
• 2018

Emission inventory is the essential component for improving the performance of air quality forecasting system. This study evaluated the simulated daily mean $PM_{2.5}$ concentrations in South Korea and China for 1-year period (Sept. 2016~Aug. 2017) using air quality forecasting system which was applied by the emission inventory of E2015 (predicted CAPSS 2015 for South Korea and KORUS 2015 v1 for the other regions). To identify the impacts of emissions on the simulated $PM_{2.5}$, the emission inventory replaced by E2010 (CAPSS 2010 and MIX 2010) were also applied under the same forecasting conditions. These results showed that simulated daily mean $PM_{2.5}$ concentrations had generally suitable performance with both emission data-sets for China (IOA>0.87, R>0.87) and South Korea (IOA>0.84, R>0.76). The impacts of the changes in emission inventories on simulated daily mean $PM_{2.5}$ concentrations were quantitatively estimated. In China, normalized mean bias (NMB) showed 5.5% and 26.8% under E2010 and E2015, respectively. The tendency of overestimated concentrations was larger in North Central and Southeast China than other regions under both E2010 and E2015. Seasonal differences of NMB were higher in non-winter season (28.3% (E2010)~39.3% (E2015)) than winter season (-0.5% (E2010)~8.0% (E2015)). In South Korea, NMB showed -5.4% and 2.8% for all days, but -15.2% and -11.2% for days below $40{\mu}g/m^3$ to minimize the impacts of long-range transport under E2010 and E2015, respectively. For all days, simulated $PM_{2.5}$ concentrations were overestimated in Seoul, Incheon, Southern part of Gyeonggi and Daejeon, and underestimated in other regions such as Jeonbuk, Ulsan, Busan and Gyeongnam, regardless of what emission inventories were applied. Our results suggest that the updated emission inventory, which reflects current status of emission amounts and spatio-temporal allocations, is needed for improving the performance of air quality forecasting.

• Journal of Environmental Impact Assessment
• /
• v.27 no.2
• /
• pp.215-231
• /
• 2018

It was analyzed to identify the cause of $PM_{2.5}$ concentration changes for the fall in 2016 and 2017 in South Korea using ground measurement data such as meterological variables and $PM_{2.5}$, AOD from GOCI satellite, and WRF-CMAQ modeling system. The result of ground measurement data showed that the $PM_{2.5}$ concentrations for the fall in 2017 decreased by 12.3% ($3.0{\mu}g/m^3$) compared to that of 2016. The difference of $PM_{2.5}$ concentrations between 2016 and 2017 mainly occurred for 11 Oct. - 20 Oct. (CASE1) and 15 Nov. - 19 Nov. (CASE2) when weather conditions were difficult to long-range transport from foreign regions and favored atmospheric ventilation in 2017 compared to 2016. Simulated $PM_{2.5}$ concentrations in 2017 decreased by 64.0% ($23.1{\mu}g/m^3$) and 35.7% ($12.2{\mu}g/m^3$) during CASE1 and CASE2, respectively. These results corresponded to the changes in observed $PM_{2.5}$ concentrations such as 53.6% for CASE1 and 47.8% for CASE2. It is implied that the changes in weather conditions affected significantly the $PM_{2.5}$ concentrations for the fall between 2016 and 2017. The contributions to decreases in $PM_{2.5}$ concentrations was assessed as 52.8% by long-range transport from foreign regions and 47.2% by atmospheric ventilation effects in domestic regions during CASE1, whereas their decreases during CASE2 were affected by 66.4% from foreign regions and 33.6% in domestic regions.

• Journal of Korean Society for Atmospheric Environment
• /
• v.34 no.5
• /
• pp.687-696
• /
• 2018

In order to develop effective emission abatement strategies for coal-fired power plants, we analyzed the shutdown effects of coal-fired power plants on $PM_{2.5}$ concentration in June by employing air quality model for the period from 2013 to 2016. WRF (Weather Research and Forecast) and CMAQ(Community Multiscale Air Quality) models were used to quantify the impact of emission reductions on the averaged $PM_{2.5}$ concentrations in June over Chungcheongnam-do area in Korea. The resultant shutdown effects showed that the averaged $PM_{2.5}$ concentration in June decreased by 1.2% in Chungcheongnam-do area and decreased by 2.3% in the area where the surface air pollution measuring stations were located. As a result of this study, it was confirmed that it is possible to analyze policy effects considering the change of meteorology and emission and it is possible to quantitatively estimate the influence at the maximum impact region by utilizing the air quality model. The results of this study are expected to be useful as a basic data for analyzing the effect of $PM_{2.5}$ concentration change according to future emission changes.

• Communications for Statistical Applications and Methods
• /
• v.25 no.2
• /
• pp.199-215
• /
• 2018

Interest in $PM_{10}$ concentrations have increased greatly in Korea due to recent increases in air pollution levels. Therefore, we consider a forecasting model for next day $PM_{10}$ concentration based on the principal elements of air pollution, weather information and Beijing $PM_{2.5}$. If we can forecast the next day $PM_{10}$ concentration level accurately, we believe that this forecasting can be useful for policy makers and public. This paper is intended to help forecast a daily mean $PM_{10}$, a daily max $PM_{10}$ and four stages of $PM_{10}$ provided by the Ministry of Environment using various data mining techniques. We use seven models to forecast the daily $PM_{10}$, which include five regression models (linear regression, Randomforest, gradient boosting, support vector machine, neural network), and two time series models (ARIMA, ARFIMA). As a result, the linear regression model performs the best in the $PM_{10}$ concentration forecast and the linear regression and Randomforest model performs the best in the $PM_{10}$ class forecast. The results also indicate that the $PM_{10}$ in Seoul is influenced by Beijing $PM_{2.5}$ and air pollution from power stations in the west coast.

• Journal of Korea Multimedia Society
• /
• v.20 no.2
• /
• pp.363-370
• /
• 2017

Data assimilation is an initializing method for air quality forecasting such as PM10. It is very important to enhance the forecasting accuracy. Optimal interpolation is one of the data assimilation techniques. It is very effective and widely used in air quality forecasting fields. The technique, however, requires too much memory space and long execution time. It makes the PM10 air quality forecasting difficult in real time. We propose a fast optimal interpolation data assimilation method for PM10 air quality forecasting using a new kernel tridiagonal sparse matrix and CUDA massively parallel processing architecture. Experimental results show the proposed method is 5~56 times faster than conventional ones.

• Journal of Korean Society for Atmospheric Environment
• /
• v.31 no.5
• /
• pp.430-436
• /
• 2015

Numerical air quality forecasting suffers from the large uncertainties of input data including emissions, boundary conditions, earth surface properties. Data assimilation has been widely used in the field of weather forecasting as a way to reduce the forecasting errors stemming from the uncertainties of input data. The present study aims at evaluating the effect of input data on the air quality forecasting results in Korea when data assimilation was invoked to generate the initial concentrations. The forecasting time was set to 36 hour and the emissions and initial conditions were chosen as tested input parameters. The air quality forecast model for Korea consisting of WRF and CMAQ was implemented for the test and the chosen test period ranged from November $2^{nd}$ to December $1^{st}$ of 2014. Halving the emission in China reduces the forecasted peak value of $PM_{10}$ and $SO_2$ in Seoul as much as 30% and 35% respectively due to the transport from China for the no-data assimilation case. As data assimilation was applied, halving the emissions in China has a negligible effect on air pollutant concentrations including $PM_{10}$ and $SO_2$ in Seoul. The emissions in Korea still maintain an effect on the forecasted air pollutant concentrations even after the data assimilation is applied. These emission sensitivity tests along with the initial condition sensitivity tests demonstrated that initial concentrations generated by data assimilation using field observation may minimize propagation of errors due to emission uncertainties in China. And the initial concentrations in China is more important than those in Korea for long-range transported air pollutants such as $PM_{10}$ and $SO_2$. And accurate estimation of the emissions in Korea are still necessary for further improvement of air quality forecasting in Korea even after the data assimilation is applied.

• Journal of Korea Multimedia Society
• /
• v.23 no.6
• /
• pp.747-755
• /
• 2020

In this paper, we deal with outlier data problems that occur when constructing a PM_2.5 fine dust forecasting system using a neural network. In general, when learning a neural network, some of the data are not helpful for learning, but rather disturbing. Those are called outlier data. When they are included in the training data, various problems such as overfitting occur. In building a PM_2.5 fine dust concentration forecasting system using neural network, we have found several outlier data in the training data. We, therefore, remove them, and then make learning 3 ways. Over_outlier model removes outlier data that target concentration is low, but the model forecast is high. Under_outlier model removes outliers data that target concentration is high, but the model forecast is low. All_outlier model removes both Over_outlier and Under_outlier data. We compare 3 models with a conventional outlier removal model and non-removal model. Our outlier removal model shows better performance than the others.

• Journal of Korean Society for Atmospheric Environment
• /
• v.33 no.5
• /
• pp.497-514
• /
• 2017

This study quantitatively analyzes the effects of emission inventory choices on the simulated particulate matter (PM) concentrations and the domestic/foreign contributions in the Seoul Metropolitan Area (SMA) with an air quality forecasting system. The forecasting system is composed of Weather Research and Forecasting (WRF)-Sparse Matrix Operator Kernel Emissions (SMOKE)-Community Multi-Scale Air Quality (CMAQ). Different domestic and foreign emission inventories were selectively adopted to set up four sets of emissions inputs for air quality simulations in this study. All modeling cases showed that model performance statistics satisfied the criteria levels (correlation coefficient >0.7, fractional error <50%) suggested by previous studies. Notwithstanding the apparently good model performance of total PM concentrations by all emission cases, annual average concentrations of simulated total PM concentrations varied up to $20{\mu}g/m^3$ (160%) depending on the combination of emission inventories. In detail, the difference in simulated annual average concentrations of the primary PM coarse (PMC) was up to $25.2{\mu}g/m^3$ (6.5 times) compared with other cases. Furthermore, model performance analyses on PM species showed that the difference in the simulated primary PMC led to gross model overestimation in general, which indicates that the primary PMC emissions need to be improved. The contribution analysis using model direct outputs indicated that the domestic contributions to the annual average PM concentrations in the SMA vary from 44% to 67%. To account for the uncertainty of the simulated concentration, the contribution correction factor method proposed by Bae et al. (2017) was applied, which resulted in converged contributions(from 48% to 57%). We believe this study shows that it is necessary to improve the simulated concentrations of PM components in order to enhance the accuracy of the forecasting model. It is deemed that these improvements will provide more accurate contribution results.

• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.1122-1137
• /
• 2021

PM_2.5 concentration in Seoul could be predicted by deep neural network model. In this paper, the contribution of input factors to the model's prediction results is analyzed using the LRP(Layer-wise Relevance Propagation) technique. LRP analysis is performed by dividing the input data by time and PM concentration, respectively. As a result of the analysis by time, the contribution of the measurement factors is high in the forecast for the day, and those of the forecast factors are high in the forecast for the tomorrow and the day after tomorrow. In the case of the PM concentration analysis, the contribution of the weather factors is high in the low-concentration pattern, and that of the air quality factors is high in the high-concentration pattern. In addition, the date and the temperature factors contribute significantly regardless of time and concentration.