We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Research on Relationship Model Between AOD and PM<sub>2.5</sub> Based on Optimized Neural Network.
- Authors
GUO Hengliang; GE Qixu; DAI Wenhao; QIAO Baojin
- Abstract
In order to solve the problem of low accuracy of PM2.5 inversion by BP neural network, based on the MODIS aerosol optical depth (AOD) and PM2.5 concentration data of Henan Province from 2017 to 2019, scaled conjugate gradient (SCG) was used instead of the LM (Levenberg-Marquardt) algorithm used by the traditional BP neural network, which has fast convergence speed and does not require other parameters. In order to verify the optimization algorithm, the research data is classified according to the seasons, and the relationship model is established using the daily average PM2.5 near-surface value and AOD. For the lack of MODIS AOD in space and time, the DDV (Dense Dark Vegetation) algorithm is used to retrieve MODIS remote sensing image as a supplement. Among them, 70% of the data set to establish training set, 20% of the data set to establish verification set, 10% of the data set to establish test set. The experimental results show that there is a significant positive correlation between PM2.5 and AOD. The correlation is 0.7 in spring, 0.84 in summer, 0.68 in autumn, and 0.56 in winter. Compared with BP neural network model, the R value of the optimized model is improved in four seasons, from 0.54 to 0.62 in spring, from 0.82 to 0.86 in summer, from 0.72 to 0.79 in autumn, and from 0.53 to 0.64 in winter. RMSE decreased in four seasons. It can be seen from the results that there is a significant correlation between PM2.5 and AOD, and AOD can be used to invert PM2.5. The optimized neural network model has a significant improvement in accuracy. The optimized neural network model can be used to estimate the high-precision PM2.5 concentration.
- Subjects
HENAN Sheng (China); ARTIFICIAL neural networks; REMOTE sensing
- Publication
Environmental Science & Technology (10036504), 2021, Vol 44, Issue 12, p46
- ISSN
1003-6504
- Publication type
Article
- DOI
10.19672/j.cnki.1003-6504.1410.21.338