Carbon dioxide (CO ₂ ) is the most important atmospheric contributor for global warming. Satellite remote sensing is a commonly used method for high-precision CO ₂ detection, but it often suffers from striping issues, which hinders its ability to achieve full coverage. Therefore, the limited availability of data poses challenges for global carbon accounting. In this paper, we generate a global seamless and high-resolution dataset of column-averaged dry air CO ₂ mole fractions (XCO ₂ ) from 2017 to 2020 by integrating multiple data sources using deep learning techniques Feedforward Neural Network (FNN). The data sources primarily include satellite, ground-based, and reanalysis XCO ₂ products, satellite vegetation index data, and meteorological data. The spatial resolution of the dataset is 0.1°, and the temporal resolution is one day. Moreover, this paper also investigated the importance of features in deep learning models and examined the spatiotemporal variations of global XCO ₂ . The results demonstrate that the FNN approach with time series features yields an improved dataset (R=0.98, RMSE=0.82 ppm) compared to the other methods. In an FNN, the known model data (CT XCO ₂ ) are considered as the most important features. We find that the global XCO ₂ has increased by approximately 7.5 ppm from 2017 to 2020. This seamless and fine-scale dataset provides valuable support for understanding global carbon cycling and formulating carbon emission reduction policies.

CO₂ is one of the primary greenhouse gases impacting global climate change, making it crucial to understand the spatiotemporal variations of CO₂. Currently, commonly used satellites serve as the primary means of CO₂ observation, but they often suffer from striping issues and fail to achieve complete coverage. This paper proposes a method for constructing a comprehensive high-spatiotemporal-resolution XCO₂ dataset based on multiple auxiliary data sources and satellite observations, utilizing multiple simple deep neural network (DNN) models. Global validation results against ground-based TCCON data demonstrate the excellent accuracy of the constructed XCO₂ dataset (R is 0.94, RMSE is 0.98 ppm). Using this method, we analyze the spatiotemporal variations of CO₂ in China and its surroundings (region: 0°–60° N, 70°–140° E) from 2019 to 2020. The gapless and fine-scale CO₂ generation method enhances people’s understanding of CO₂ spatiotemporal variations, supporting carbon-related research.

Atmospheric correction is crucial for quantitative remote sensing. However, an insufficient radiative transfer model under polluted conditions limits the accuracy of high-resolution satellite data and hinders practical applications. We propose a novel atmospheric correction framework, Regional Aerosol Clustering-assisted Neural Networks for Atmospheric Correction (RACNN), which combines a radiative transfer model with a multi-layer neural network (MLNN). Ground-based aerosol data from three AERONET sites in Beijing were used to cluster typical aerosol conditions, including regional aerosol pollution, and the 6S model simulated spectral data under diverse atmospheric scenarios. The MLNN was trained to learn the mapping between top-of-atmosphere reflectance and surface reflectance, incorporating geometric and spectral information. Model performance was evaluated using both simulated and field-measured data. The results demonstrate that the proposed method achieves high prediction accuracy across various spectral bands, with an overall root mean square error (RMSE) of approximately 0.03 sr^-1 and a correlation coefficient (R) exceeding 0.94. Among the different seasons, the highest prediction accuracy was observed in summer and fall, while lower accuracy was noted in winter and spring. These seasonal differences may be attributed to the varying aerosol properties and vegetation status. Specifically, the model exhibited robust performance on land cover types such as forests and croplands, whereas relatively higher prediction errors were found in grasslands. The proposed RACNN method outperforms traditional corrections in both accuracy and efficiency under clear skies and in the presence of aerosol pollution, demonstrating the potential of physics-informed machine learning for scalable, high-resolution atmospheric correction.