Estimation of PM<sub>2.5</sub> Concentration Using Deep Bayesian Model Considering Spatial Multiscale
Directly establishing the relationship between satellite data and PM<sub>2.5</sub> concentration through deep learning methods for PM<sub>2.5</sub> concentration estimation is an important means for estimating regional PM<sub>2.5</sub> concentration. However, due...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
MDPI AG
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/897952685233496aaaf89e6fa4e8cde3 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Directly establishing the relationship between satellite data and PM<sub>2.5</sub> concentration through deep learning methods for PM<sub>2.5</sub> concentration estimation is an important means for estimating regional PM<sub>2.5</sub> concentration. However, due to the lack of consideration of uncertainty in deep learning methods, methods based on deep learning have certain overfitting problems in the process of PM<sub>2.5</sub> estimation. In response to this problem, this paper designs a deep Bayesian PM<sub>2.5</sub> estimation model that takes into account multiple scales. The model uses a Bayesian neural network to describe key parameters a priori, provide regularization effects to the neural network, perform posterior inference through parameters, and take into account the characteristics of data uncertainty, which is used to alleviate the problem of model overfitting and to improve the generalization ability of the model. In addition, different-scale Moderate-Resolution Imaging Spectroradiometer (MODIS) satellite data and ERA5 reanalysis data were used as input to the model to strengthen the model’s perception of different-scale features of the atmosphere, as well as to further enhance the model’s PM<sub>2.5</sub> estimation accuracy and generalization ability. Experiments with Anhui Province as the research area showed that the <i>R</i><sup>2</sup> of this method on the independent test set was 0.78, which was higher than that of the DNN, random forest, and BNN models that do not consider the impact of the surrounding environment; moreover, the RMSE was 19.45 μg·m<sup>−3</sup>, which was also lower than the three compared models. In the experiment of different seasons in 2019, compared with the other three models, the estimation accuracy was significantly reduced; however, the <i>R</i><sup>2</sup> of the model in this paper could still reach 0.66 or more. Thus, the model in this paper has a higher accuracy and better generalization ability. |
|---|