A CNN+LSTM (Convolutional Neural Network + Long Short-Term Memory) based deep hybrid neural network was established for the citywide daily PM_2.5 prediction in South Korea. The structural hyperparameters of the CNN+LSTM model were determined through comprehensive sensitivity tests. The input features were obtained from the ground observations and GFS forecast. The performance of CNN+LSTM was evaluated by comparison with PM_2.5 observations and with the 3-D CTM (three-dimensional chemistry transport model)-predicted PM_2.5. The newly developed hybrid model estimated more accurate ambient levels of PM_2.5 compared to the 3-D CTM. For example, the error and bias of the CNN+LSTM prediction were 1.51 and 6.46 times smaller than those by 3D-CTM simulation. In addition, based on IOA (Index of Agreement), the accuracy of CNN+LSTM prediction was 1.10–1.18 times higher than the 3-D CTM-based prediction. The importance of input features was indirectly investigated by sequential perturbing input variables. The most important meteorological and atmospheric environmental features were geopotential height and previous day PM_2.5. The obstacles of the current CNN+LSTM-based PM_2.5 prediction were also discussed. The promising result of this study indicates that DNN-based models can be utilized as an effective tool for air quality prediction.

Concentrations of ambient particulate matter (such as PM_2.5 and PM₁₀) have come to represent a serious environmental problem worldwide, causing many deaths and economic losses. Because of the detrimental effects of PM_2.5 on human health, many countries and international organizations have developed and operated regional and global short-term PM_2.5 prediction systems. The short-term predictability of PM_2.5 (and PM₁₀) is determined by two main factors: the performance of the air quality model and the precision of the initial states. While specifically focusing on the latter factor, this study attempts to demonstrate how information from classical ground observation networks, a state-of-the-art geostationary (GEO) satellite sensor, and an advanced air quality modeling system can be synergistically combined to improve short-term PM_2.5 predictability over South Korea. Such a synergistic combination of information can effectively overcome the major obstacle of scarcity of information, which frequently occurs in PM_2.5 prediction systems using low Earth orbit (LEO) satellite-borne observations. This study first presents that the scarcity of information is mainly associated with cloud masking, sun-glint effect, and ill-location of satellite-borne data, and it then demonstrates that an advanced air quality modeling system equipped with synergistically-combined information can achieve substantially improved performances, producing enhancements of approximately 10%, 19%, 29%, and 10% in the predictability of PM_2.5 over South Korea in terms of index of agreement (IOA), correlation coefficient (R), mean biases (MB), and hit rate (HR), respectively, compared to PM_2.5 prediction systems using only LEO satellite-derived observations.