9:15 AM - 9:30 AM
[HTT19-02] Estimation of aerosol concentration using image analysis techniques (SNAP-CII)
Keywords:Aerosol, Image analysis, Machine learning
Aerosols play an important role for the radiative forcing, such as scattering and absorption of sunlight and cloud condensation, and they, especially PM2.5, are also suggested to increase morbidity and mortality. In this study, we develop an algorithm, named SNAP-CII, to estimate aerosol concentration from sky image data. The SNAP-CII algorithm enables us to measure aerosol concentration with a commonly used camera even in places where it is difficult to install specialized equipment such as a sky radiometer.
We installed cameras on the rooftops of Fukuoka University (FU) and Kyushu Institute of Technology (KT) in December 2020. The captured image data were automatically sent via the Internet using the video transmission system developed by the National Institute of Information and Communications Technology and stored at 1 minute intervals. The image data from 9:00 to 18:00 were used for this study.
We used the data of concentration of suspended particle matter (SPM) measured by the Atmospheric Environmental Regional Observation System (AEROS) . The data measured at Sohara and Tobata monitoring station were used for analyzing the data taken by FU and KT, respectively. The solar radiation data observed by Himawari 8 satellite were used for selection of weather condition. In this study, we used the data with global solar irradiance (GSI) and diffuse solar irradiance (DSI) to select clear sunny conditions. The criteria were GSI > 550 W/m2 and DSI/GSI < 0.15.
A machine learning model was developed to classify SPM concentration into three classes (Low: 0 – 10, Middle: 10 – 25, High: > 25 [μg/m3]) using the sky image data. The reflectance (gray-scale, B, G, R) and pixel value ratio (B/G, G/R, R/B) were calculated to extract the difference in scattering by aerosols depending on wavelength. The reflectance was derived by dividing the luminance, which is the pixel value, by the illuminance. The illuminance was derived by applying a bilateral filter to the average image data from 11:00 – 14:00 on December 8 when the SPM averages were low (Sohara: 8.5 μg/m3, Tobata: 6.3 μg/m3) and the solar irradiance satisfied the criteria. The average values over eight image ranges (entire area, upper, middle and lower parts of sky, left, center and right of boundary parts of sky and buildings, and buildings area) were calculated.
We tested three types of classification models, K-nearest neighbor (KN), support vector machine (SVM) and random forest (RF). The data were randomly divided into training and test data in a ratio of 6:4. The accuracy, number of correct predictions divided by total number of samples, for each model was calculated. The RF had the highest accuracy of 70%, and the KN and SVM had 65% for the data taken by FU. The KT camera data also showed the highest accuracy by the RF model, and the accuracy was 66%.
In this study, we developed three-class classification model of aerosol concentration based on sky image data in order to measure aerosol concentration more conveniently. The classification algorithm, named SNAP-CII, was developed for clear sunny conditions in Fukuoka and Kitakyushu cities, and the accuracies were 66 – 70%.
We installed cameras on the rooftops of Fukuoka University (FU) and Kyushu Institute of Technology (KT) in December 2020. The captured image data were automatically sent via the Internet using the video transmission system developed by the National Institute of Information and Communications Technology and stored at 1 minute intervals. The image data from 9:00 to 18:00 were used for this study.
We used the data of concentration of suspended particle matter (SPM) measured by the Atmospheric Environmental Regional Observation System (AEROS) . The data measured at Sohara and Tobata monitoring station were used for analyzing the data taken by FU and KT, respectively. The solar radiation data observed by Himawari 8 satellite were used for selection of weather condition. In this study, we used the data with global solar irradiance (GSI) and diffuse solar irradiance (DSI) to select clear sunny conditions. The criteria were GSI > 550 W/m2 and DSI/GSI < 0.15.
A machine learning model was developed to classify SPM concentration into three classes (Low: 0 – 10, Middle: 10 – 25, High: > 25 [μg/m3]) using the sky image data. The reflectance (gray-scale, B, G, R) and pixel value ratio (B/G, G/R, R/B) were calculated to extract the difference in scattering by aerosols depending on wavelength. The reflectance was derived by dividing the luminance, which is the pixel value, by the illuminance. The illuminance was derived by applying a bilateral filter to the average image data from 11:00 – 14:00 on December 8 when the SPM averages were low (Sohara: 8.5 μg/m3, Tobata: 6.3 μg/m3) and the solar irradiance satisfied the criteria. The average values over eight image ranges (entire area, upper, middle and lower parts of sky, left, center and right of boundary parts of sky and buildings, and buildings area) were calculated.
We tested three types of classification models, K-nearest neighbor (KN), support vector machine (SVM) and random forest (RF). The data were randomly divided into training and test data in a ratio of 6:4. The accuracy, number of correct predictions divided by total number of samples, for each model was calculated. The RF had the highest accuracy of 70%, and the KN and SVM had 65% for the data taken by FU. The KT camera data also showed the highest accuracy by the RF model, and the accuracy was 66%.
In this study, we developed three-class classification model of aerosol concentration based on sky image data in order to measure aerosol concentration more conveniently. The classification algorithm, named SNAP-CII, was developed for clear sunny conditions in Fukuoka and Kitakyushu cities, and the accuracies were 66 – 70%.