Multi-source Sentinel-5P Pixel-Based Dataset for Supervised Regression of XCO2 and XCH4 Data set

Dataset Description Overview This dataset integrates multi-source satellite observations, ERA5 meteorological reanalysis variables, and ground-based TCCON reference measurements for machine learning modelling of atmospheric CO₂ and CH₄. All variables have been spatially collocated and temporally aligned to ensure consistency across data sources. The primary objective of the dataset is to support supervised regression of column-averaged dry-air mole fractions of carbon dioxide (XCO₂) and methane (XCH₄). Ground-based measurements from the Total Carbon Column Observing Network (TCCON) are used as target variables for model training and validation. Users should cite the TCCON network as described in Wunch et al. (2011) and refer to the specific station data DOI as appropriate. Satellite and environmental predictors are spatially and temporally collocated with TCCON observations. Target Variables (Training Data) Total Carbon Column Observing Network (TCCON) TCCON is a global network of ground-based Fourier Transform Infrared (FTIR) spectrometers that retrieve high-precision column-averaged greenhouse gas concentrations from solar absorption spectra. The following target variables are included: XCO₂ Column-averaged dry-air mole fraction of carbon dioxide (ppm). \[ $XCO_2 = \frac{\text{total column } CO_2}{\text{total column dry air}}$ \] Typical precision: ~0.25 ppmTypical accuracy: ~0.4 ppm (after calibration to WMO scale) XCH₄ Column-averaged dry-air mole fraction of methane (ppb). $XCH_4 = \frac{\text{total column } CH_4}{\text{total column dry air}}$ Typical precision: 2-5 ppbTypical accuracy: ~5-8 ppb These measurements are widely used to validate satellite missions, including OCO-2 and Sentinel-5P. In this dataset, TCCON XCO₂ and XCH₄ serve as the supervised learning targets. TCCON Stations Included TCCON Sites in the Dataset The dataset includes observations from the following TCCON sites: Pasadena, USA (Wennberg et al., 2022) Edwards / California Desert, USA (Iraci et al., 2022) Lamont, Oklahoma, USA (Wennberg et al., 2025) Xianghe, China (Zhou et al., 2022) Park Falls, Wisconsin, USA (Wennberg et al., 2022b) Garmisch, Germany (Sussmann et al., 2025) Orléans, France (Warneke et al., 2024) Paris (Jussieu), France (Té et al., 2022) Karlsruhe, Germany (Hase et al., 2024) Harwell, United Kingdom (Weidmann et al., 2023) Bremen, Germany (Notholt et al., 2022) East Trout Lake, Canada (Wunch et al., 2022) These stations cover diverse climatic and surface environments across North America, Europe, and East Asia. Predictor Variables Sentinel-5P (TROPOMI) CH4_column_volume_mixing_ratio_dry_air CH4_column_volume_mixing_ratio_dry_air_bias_corrected CH4_column_volume_mixing_ratio_dry_air_uncertainty CO_column_number_density These variables represent column-averaged trace gas retrievals derived from hyperspectral shortwave infrared observations. MODIS Aerosol and Vegetation Products Optical_Depth_047 Optical_Depth_055 leaf_area_index_high_vegetation leaf_area_index_low_vegetation LC_Type1 (IGBP land cover classification) EVI Aerosol optical depth (AOD) and vegetation structure variables are included to characterise atmospheric loading and surface properties. Sentinel-2 Level 2A Derived Surface Indicators NDVI (Normalised Difference Vegetation Index) S2_SCL (Scene Classification Layer) Vegetation indices and surface classification masks are derived from Sentinel-2 reflectance data. ERA5 Meteorological Reanalysis (ECMWF) surface_net_solar_radiation_sum surface_pressure temperature_2m total_evaporation_sum u_component_of_wind_10m v_component_of_wind_10m volumetric_soil_water_layer_1 volumetric_soil_water_layer_2 Meteorological and soil moisture variables are included to capture atmospheric transport, boundary-layer dynamics, and land–atmosphere interactions. Processing and Harmonisation All satellite and reanalysis predictors were spatially collocated with TCCON station coordinates and temporally matched to observation dates. Quality filtering procedures were applied where relevant (e.g., cloud masking using Sentinel-2 SCL). Vegetation indices were derived from calibrated surface reflectance data. All variables were harmonised to a common spatial and temporal framework to enable supervised machine learning. Intended Use This dataset is designed for: Supervised regression of XCO₂ and XCH₄ Satellite product evaluation and bias assessment Data fusion studies combining satellite and ground-based observations Machine learning benchmarking for atmospheric trace gas modelling Refrences Wunch, D., Toon, G. C., Blavier, J. F. L., Washenfelder, R. A., Notholt, J., Connor, B. J., ... & Wennberg, P. O. (2011). The total carbon column observing network. Philosophical transactions of the Royal Society A: Mathematical, physical and engineering sciences, 369(1943), 2087-2112. Iraci, L. T., Podolske, J. R., Roehl, C., Wennberg, P. O., Blavier, J.-F., Allen, N., Wunch, D., & Osterman, G. B. (2022). TCCON data from Edwards (US), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.edwards01.R0 Hase, F., Herkommer, B., Groß, J., Blumenstock, T., Kiel, M. ä ., & Dohe, S. (2024). TCCON data from Karlsruhe (DE), Release GGG2020.R2 (Version R2) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.karlsruhe01.R2 Notholt, J., Petri, C., Warneke, T., & Buschmann, M. (2022). TCCON data from Bremen (DE), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.bremen01.R0 Sussmann, R., Rettinger, M., & Mostafavi Pak, N. (2025). TCCON data from Garmisch (DE), Release GGG2020.R1 (Version R1) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.garmisch01.R1 Té, Y., Jeseck, P., & Janssen, C. (2022). TCCON data from Paris (FR), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.paris01.R0 Warneke, T., Petri, C., Notholt, J., & Buschmann, M. (2024). TCCON data from Orléans (FR), Release GGG2020.R1 (Version R1) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.orleans01.R1 Weidmann, D., Brownsword, R., & Doniki, S. (2023). TCCON data from Harwell, Oxfordshire (UK), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.harwell01.R0 Wennberg, P. O., Roehl, C. M., Wunch, D., Blavier, J.-F., Toon, G. C., Allen, N. T., Treffers, R., & Laughner, J. (2022). TCCON data from Caltech (US), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.pasadena01.R0 Wennberg, P. O., Wunch, D., Roehl, C. M., Blavier, J.-F., Toon, G. C., & Allen, N. T. (2025). TCCON data from Lamont (US), Release GGG2020.R1 (Version R1) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.lamont01.R1 Wennberg, P. O., Roehl, C. M., Wunch, D., Toon, G. C., Blavier, J.-F., Washenfelder, R., Keppel-Aleks, G., & Allen, N. T. (2022b). TCCON data from Park Falls (US), Release GGG2020.R1 (Version R1) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.parkfalls01.R1 Wunch, D., Mendonca, J., Colebatch, O., Allen, N. T., Blavier, J.-F., Kunz, K., Roche, S., Hedelius, J., Neufeld, G., Springett, S., Worthy, D., Kessler, R., & Strong, K. (2022). TCCON data from East Trout Lake, SK (CA), Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.easttroutlake01.R0 Zhou, M., Wang, P., Kumps, N., Hermans, C., & Nan, W. (2022). TCCON data from Xianghe, China, Release GGG2020.R0 (Version R0) [Data set]. CaltechDATA. https://doi.org/10.14291/tccon.ggg2020.xianghe01.R0