Abstract:Air pollution is one of the most important environmental problems in China. As a major air pollutant, ozone (O3) will endanger human health and terrestrial ecosystems. It is of great practical significance to obtain a continuous full-coverage dataset of ozone with high spatio-temporal resolution to conduct mechanism research from its causes, development, diffusion, impact and other aspects. In this study, a 3-stage machine learning model was developed through multiple data fusion, and the LightGBM method is used to obtain the hourly spatio-temporal distribution dataset of the O3 concentration in China from 2013 to 2020, with a resolution of 0.25 °× 0.25°. We first revise the meteorological reanalysis data using ground observation and propose a data fusion algorithm to achieve the ground level distribution of ozone, which combines ground observation of pollutants, population data, revised reanalysis meteorological conditions, reanalysis of radiation, land and vegetation data, emission inventory and results of chemical transport model simulation. In addition, due to the common phenomenon that the previous prediction models underestimate the extreme value of the pollution periods,therefore, we redefined the heavy pollution event and assimilated it into the 0.25 grid by using the synthetic minimum oversampling technique (SMOTE) method to improve the model performance during the extreme pollution periods.Our model, with the 10-fold CV result of R2 = 71% and RMSE= 25.1μg·m-3, and our hourly O3 concentration results are spatially and spatially continuous with a similar distribution compare to the observation, which proves the reliability of our model. With higher time resolution, various exposure response indicators can be obtained. AOT40 calculated by high-resolution hourly ozone concentration further, which is far more accurate than it when directly predicted by daily indexs modeling.In addition, based on the distribution of AOT40, we assessed the agricultural damage and ecological damage caused by the change of surface ozone pollution during 2013-2020. Our estimation considered the planting area and phenological period of crops that the overestimation of crop RYL in the region can be avoided. The annual avrage production loss of wheat, rice and maize in China from 2013 to 2020 is 55.0, 57.4 and 23.6 Mt, respectively. Besides, The loss of gross primary productivity was also estimated. During 2013-2020, the ozone pollution in China caused an annual average loss of 2.1%, and the loss in the south was much higher than that in the north.

Spatio‐temporal Hourly and Daily Ozone Forecasting in China Using a Hybrid Machine Learning Model: Autoencoder and Generative Adversarial Networks

Predicting Air Quality Using a Multi-Scale Spatiotemporal Graph Attention Network

Long time series ozone prediction in China: A novel dynamic spatiotemporal deep learning approach

Development of a High-Performance Machine Learning Model to Predict Ground Ozone Pollution in Typical Cities of China.

Regional Prediction of Ground-Level Ozone Using a Hybrid Sequence-to-sequence Deep Learning Approach

A High-Performance Convolutional Neural Network for Ground-Level Ozone Estimation in Eastern China

Spatiotemporal distributions of surface ozone levels in China from 2005 to 2017: A machine learning approach

Air quality forecasting using a spatiotemporal hybrid deep learning model based on VMD-GAT-BiLSTM

A deep learning model integrating a wind direction-based dynamic graph network for ozone prediction

Machine Learning‐Driven Spatiotemporal Analysis of Ozone Exposure and Health Risks in China

Recurrent U-net: Deep learning to predict daily summertime ozone in the United States

Ozone concentration forecast method based on genetic algorithm optimized back propagation neural networks and support vector machine data classification

Combining physical mechanisms and deep learning models for hourly surface ozone retrieval in China

A hybrid CNN-Transformer model for ozone concentration prediction

CEEMD-subset-OASVR-GRNN for Ozone Forecasting: Xiamen and Harbin As Cases

A novel framework for daily forecasting of ozone mass concentrations based on cycle reservoir with regular jumps neural networks

Near Real-Time Distribution of Ozone in China from 2013 to 2020 and Agricultural Impacts

Daily air quality index forecasting with hybrid models: A case in China.

Hybrid machine learning model for hourly ozone concentrations prediction and exposure risk assessment

A hybrid deep learning model for regional O3 and NO2 concentrations prediction based on spatiotemporal dependencies in air quality monitoring network

Estimation of Near-Ground Ozone With High Spatio-Temporal Resolution in the Yangtze River Delta Region of China Based on a Temporally Ensemble Model