Prediction of Tropospheric NO<sub>2</sub> Profile Using CNN-SVR-Based MAX-DOAS

Yifeng Pan; Xin Tian; Pinhua Xie; Ang Li; Jin Xu; Bo Ren; Xiaohui Huang; Wei Tian; Zijie Wang

doi:10.3788/AOS202242.2401001

Acta Optica Sinica, Volume. 42, Issue 24, 2401001(2022)

Prediction of Tropospheric NO₂ Profile Using CNN-SVR-Based MAX-DOAS

Yifeng Pan¹, Xin Tian^1,2, Pinhua Xie^2,3,4、*, Ang Li², Jin Xu², Bo Ren^2,4, Xiaohui Huang¹, Wei Tian¹, and Zijie Wang¹

¹Information Materials and Intelligent Sensing Laboratory of Anhui Province, Institutes of Physical Science and Information Technology, Anhui University, Hefei 230601, Anhui , China

²Key Laboratory of Environmental Optics and Technology, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei 230031, Anhui , China

³CAS Center for Excellence in Regional Atmospheric Environment, Xiamen 361021, Fujian , China

⁴School of Environmental Science and Optoelectronic Technology, University of Science and Technology of China, Hefei 230026, Anhui , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Flow chart of MIV method

Download full size

Fig. 2. Percentage of MIV value of different input variables in the sum of all input variables

Download full size

Fig. 3. Frame of CNN-SVR hybrid model

Download full size

Fig. 4. Comparison of NO₂ profiles predicted by CNN-SVR model and retrieved by MAX-DOAS

Download full size

Fig. 5. Monthly average profile of NO₂ from July 2019 to June 2020

Download full size

Fig. 6. Prediction error of CNN-SVR model under different AOD ranges

Download full size

Table 1. RMSE values under different input variable systems

View table

Table 1. RMSE values under different input variable systems

System	Input variable system	RMSE
F1	x₁，x₄，x₅，x₆，x₇，x₃，x₂，x₈	0.512
F2	x₁，x₄，x₅，x₆，x₇，x₃，x₂	0.487
F3	x₁，x₄，x₅，x₆，x₇，x₃	0.452
F4	x₁，x₄，x₅，x₆，x₇	0.415
F5	x₁，x₄，x₅，x₆	0.322
F6	x₁，x₄，x₅	0.401
F7	x₁，x₄	0.498
F8	x₁	0.586

Table 2. Parameter setting of CNN

View table

Table 2. Parameter setting of CNN

Network layer	Model parameter setting
Input layer	685×11 spectral data matrix
Convolution layer 1	64 1×1 convolution kernels；kernel_size：5
Convolution layer 2	128 1×1 convolution kernels；kernel_size：5
Pool layer 1	MaxPool；kernel_size：1；stride：2
Convolution layer 3	128 1×1 convolution kernels；kernel_size：5
Pool layer 2	MaxPool；kernel_size：1；stride：2
Convolution layer 4	256 1×1 convolution kernels；kernel_size：5
Pool layer 3	MaxPool；kernel_size：1；stride：2
Convolution layer 5	512 1×1 convolution kernels；kernel_size：5
Pool layer 4（adaptive pooling layer）	Output one-dimensional vector
Full connection layer	Output 21 neurons

Table 3. Prediction results of CNN-SVR model test set under different training set samples
View table
Table 3. Prediction results of CNN-SVR model test set under different training set samples
Ratio of training set SMAPE /% MAPE /%
0.5 21.97 23.41
0.6 18.11 19.33
0.7 15.13 16.24
0.8 11.55 12.41
0.9 8.52 9.14

Table 4. MAPE and SMAPE under four prediction models
View table
Table 4. MAPE and SMAPE under four prediction models
Model MAPE /% SMAPE /%
Training error Test error Training error Test error
CNN 12.59 17.63 11.61 16.11
SVR 11.82 15.14 10.92 14.28
BP 35.68 41.42 33.63 39.21
CNN-SVR 7.93 9.14 7.25 8.52

Table 5. R² under four prediction models
View table
Table 5. R² under four prediction models
Model R² in modeling set R² in testing set
CNN 0.86 0.83
SVR 0.89 0.87
BP 0.81 0.77
CNN-SVR 0.95 0.93

Table 6. Prediction error range of CNN-SVR model under different AOD ranges

View table

Table 6. Prediction error range of CNN-SVR model under different AOD ranges

AOD	Prediction MAPE range /%
（0，0.5］	（3.17，7.63）
（0.5，1.0］	（5.13，8.21）
（1.0，1.5］	（6.73，9.16）
（1.5，2.0］	（6.15，11.19）
（2.0，2.5］	（7.59，11.80）
（2.5，3.0］	（7.21，13.17）
>3.0	（9.71，16.23）

Tools

Get Citation

Copy Citation Text

Yifeng Pan, Xin Tian, Pinhua Xie, Ang Li, Jin Xu, Bo Ren, Xiaohui Huang, Wei Tian, Zijie Wang. Prediction of Tropospheric NO₂ Profile Using CNN-SVR-Based MAX-DOAS[J]. Acta Optica Sinica, 2022, 42(24): 2401001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites