Estimation of Lower-Stratosphere-to-Troposphere Ozone Profile Using Long Short-Term Memory (LSTM)

High-spatial resolution ground-level ozone in Yunnan, China: A spatiotemporal estimation based on comparative analyses of machine learning models

Monitoring ground-level ozone concentrations is a critical aspect of atmospheric environmental studies. Given the existing limitations of satellite data products, especially the lack of ground-level ozone characterization, and the discontinuity of ground observations, there is a pressing need for high-precision models to simulate ground-level ozone to assess surface ozone pollution. In this study, we have compared several widely utilized ensemble learning and deep learning methods for ground-level ozone simulation. Furthermore, we have thoroughly contrasted the temporal and spatial generalization performances of the ensemble learning and deep learning models. The 3-Dimensional Convolutional Neural Network (3-D CNN) model has emerged as the optimal choice for evaluating the daily maximum 8-h average ozone in Yunnan Province. The model has good performance: a spatial resolution of 0.05° × 0.05° and strong predictive power, as indicated by a Coefficient of Determination (R2) of 0.83 and a Root Mean Square Error (RMSE) of 12.54 μg/m³ in sample-based 5-fold cross-validation (CV). In the final stage of our study, we applied the 3-D CNN model to generate a comprehensive daily maximum 8-h average ozone dataset for Yunnan Province for the year 2021. This application has furnished us with a crucial high-resolution and highly accurate dataset for further in-depth studies on the issue of ozone pollution in Yunnan Province.

Estimation of ground-level O3 concentration in the Yangtze River Delta region based on a high-performance spatiotemporal model MixNet

In recent years, the escalating ozone (O3) concentration has significantly damaged human health. The machine learning models are widely used to estimate ground-level O3 concentrations, but the spatial and temporal features in the data are less considered. To address the issue, this study proposed a novel framework named MixNet to estimate daily O3 concentration from 2020 to 2021 over the Yangtze River Delta. The MixNet utilized image convolution to extract the potential spatial information related to O3 fully. The temporal features were extracted by a Long Short-Term Memory (LSTM). A U-Net, a new jump connection method with an attention mechanism and residual blocks, facilitated a more comprehensive extraction of spatial features in the data. The extracted temporal and spatial features were fused to estimate ground-level O3. Meanwhile, a novel training method was proposed to enhance the accuracy of MixNet. The daily mean O3 maps have high validation results in comparison with ground-level O3 measurement, with R2 (RMSE) of 0.903 (14.511 μg/m3) for sample-based validation, 0.831 (19.036 μg/m3) for site-based validation, and 0.712 (25.108 μg/m3) for time-based validation. The season-average maps indicate that O3 concentration is summer > autumn > spring > winter. The highest value was 137.41 μg/m3 in the summer of 2021 over the Yangtze River Delta urban agglomeration, and the lowest value was 52.73 μg/m3 in winter 2020. The MixNet showed better performance compared with other models, and thus the "point-plane image thinking" will contribute to future studies in developing better methods to estimate atmospheric pollutants.

Estimation of surface ozone concentration over Jiangsu province using a high-performance deep learning model

Recently, the global background concentration of ozone (O₃) has demonstrated a rising trend. Among various methods, groun-based monitoring of O₃ concentrations is highly reliable for research analysis. To obtain information on the spatial characteristics of O₃ concentrations, it is necessary that the ground monitoring sites be constructed in sufficient density. In recent years, many researchers have used machine learning models to estimate surface O₃ concentrations, which cannot fully provide the spatial and temporal information contained in a sample dataset. To solve this problem, the current study utilized a deep learning model called the Residual connection Convolutional Long Short-Term Memory network (R-ConvLSTM) to estimate daily maximum 8-hr average (MDA8) O₃ over Jiangsu province, China during 2020. In this research, the R-ConvLSTM model not only provides the spatiotemporal information of MDA8 O₃, but also involves residual connection to avoid the problem of gradient explosion and gradient disappearance with the deepening of network layers. We utilized the TROPOMI total O₃ column retrieved from Sentinel-5 Precursor, ERA5 reanalysis meteorological data, and other supplementary data to build a pre-trained dataset. The R-ConvLSTM model achieved an overall sample-base cross-validation (CV) R² of 0.955 with root mean square error (RMSE) of 9.372 µg/m³. Model estimation also showed a city-based CV R² of 0.896 with RMSE of 14.029 µg/m³, the highest MDA8 O₃ in spring being 122.60 ± 31.60 µg/m³ and the lowest in winter being 69.93 ± 18.48 µg/m³.

Development of a high-performance machine learning model to predict ground ozone pollution in typical cities of China

High ozone concentrations have adverse effects on human health and ecosystems. In recent years, the ambient ozone concentration in China has shown an upward trend, and high-quality prediction of ozone concentrations has become critical to support effective policymaking. In this study, a novel hybrid model combining wavelet decomposition (WD), a gated recurrent unit (GRU) neural network and a support vector regression (SVR) model was developed to predict the daily maximum 8 h ozone. We used the ground ozone observation data in six representative megacities across China from Jan. 1, 2015 to Jun. 15, 2020 for model training, and we used data from Jun. 15 to Dec. 31, 2020 for model testing. The results show that the developed model performs very well for megacities; against observations, the model obtains an average cross-validated R² (coefficient of determination) ranging from 0.90 for Shanghai to 0.97 for Chengdu in the one-step predictions, thereby indicating that the model outperformed any single algorithm or other hybrid algorithms reported. The developed model can also capture high ozone pollution episodes with an average accuracy of 92% for the next five days in inland cities. This study will be useful for the environmental health community to prevent high ozone exposure more efficiently in megacities in China and shows great potential for accurate ozone prediction using machine learning approaches.

Large-Scale River Mapping Using Contrastive Learning and Multi-Source Satellite Imagery

River system is critical for the future sustainability of our planet but is always under the pressure of food, water and energy demands. Recent advances in machine learning bring a great potential for automatic river mapping using satellite imagery. Surface river mapping can provide accurate and timely water extent information that is highly valuable for solid policy and management decisions. However, accurate large-scale river mapping remains challenging given limited labels, spatial heterogeneity and noise in satellite imagery (e.g., clouds and aerosols). In this paper, we propose a new multi-source data-driven method for large-scale river mapping by combining multi-spectral imagery and synthetic aperture radar data. In particular, we build a multi-source data segmentation model, which uses contrastive learning to extract the common information between multiple data sources while also preserving distinct knowledge from each data source. Moreover, we create the first large-scale multi-source river imagery dataset based on Sentinel-1 and Sentinel-2 satellite data, along with 1013 handmade accurate river segmentation mask (which will be released to the public). In this dataset, our method has been shown to produce superior performance (F1-score is 91.53%) over multiple state-of-the-art segmentation algorithms. We also demonstrate the effectiveness of the proposed contrastive learning model in mapping river extent when we have limited and noisy data.