Estimation of daily ground-level PM2.5 concentrations over the Pearl River Delta using 1 km resolution MODIS AOD based on multi-feature BiLSTM

Predicting the Impact of Change in Air Quality Patterns Due to COVID-19 Lockdown Policies in Multiple Urban Cities of Henan: A Deep Learning Approach

“…However, the lack of access to data on air pollution in emerging countries, where public and private transport systems are high, can cause errors or alterations in the real information on the state of air quality. Another study [73] aimed to build a deep learning time series model using the Bi-directional Long Short-Term Memory (Bi-LSTM) network, combining various factors such as AOD, meteorology, and socio-economic factors. However, our study is an extended work which focused on multiple cities and the analysis of the behavior of air quality patterns.…”

Scientometric and multidimensional contents analysis of PM2.5 concentration prediction

“…It is used for the prediction of spatial and temporal variation of PM 2.5 concentration in recent years. A drawback of AOD data is the possibility of loss due to meteorological factors, however, their utility lies in the ability of multiple sources to complement each other, and these data can be verified and calibrated with ground-based measurements of PM 2.5 concentrations [ 43 , 67 ].…”

“…(1) Socio-economic indicators: these resources are the fundamental sources of PM 2.5 generation, such as energy consumption, energy structure, industrial structure, population density, urbanization level, and economic development status. Data on land use, landscape index, traffic information, population, regional pollution emissions, three-dimensional building morphology, food service distribution, bus stop density, intersection density, and shortest distance to roads are also included in regression prediction models [ 3 , 67 ]. (2) Natural condition indicators, topographic condition data include: Enhanced Vegetation Index (EVI), Digital elevation model (DEM), and Normalized Vegetation Index (NDVI); Meteorological conditions include: relative humidity (RHU), mean pressure (PRS), mean temperature (TEM), gradient Surface Temperature (GST), mean wind speed (WIN–S), wind direction (WIN-D), visibility (VIS), planetary boundary layer height (PBL), dew point temperature (DPT), atmospheric stability etc.…”

“…From the site level, the local prediction at the site level and the large-scale scope at the city level and above are not matched spatially, lacking an integrated multi-temporal scale comprehensive prediction technology system. In the future, PM 2.5 concentration data from ground monitoring stations can be used to match with regional aerosol optical thickness and meteorological data in time and space for satellite-model-ground multi-source data to form an input dataset and build a complete PM 2.5 multi-source database [ 67 ]. Combining the advantages of different prediction methods, a “One-Click” PM 2.5 prediction system can be further developed to facilitate accurate prediction of PM 2.5 concentrations and PM 2.5 pollution warnings for environmental departments, as well as to provide a scientific basis for regional industrial planning and environmental target setting, which can effectively improve the practical application of prediction results.…”

“…A single model often has certain shortcomings in dealing with PM 2.5 concentration time series, and the prediction accuracy cannot meet the needs. Therefore, many scholars have started to experiment with combined prediction models for PM 2.5 concentrations [ 7 , 18 , 67 ], which can solve the shortcomings of a single model by combining several models to complement each other’s strengths. For example, AI prediction models can be integrated with various data pre-processing and optimization techniques, such as decomposition methods, outlier detection and correction, clustering algorithms, and intelligent optimization algorithms, to enhance their performance.…”