the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Aug 2026
HBPG-CO2: a global gridded monthly anthropogenic CO2 emission dataset at 0.05° resolution for 1970–2025
Abstract. Carbon dioxide emissions, as geospatial information, are necessary to analyze seasonal and regional variability, as well as the changing distribution of anthropogenic activity. However, the availability of continuous data with high spatial and temporal resolution is limited, both sectorial and global. Therefore, in this study, we present the Hierarchical Bayesian Physics-Guided Carbon Dioxide (HBPG-CO2) emission dataset, a global dataset of emissions from multiple sources for energy, industry, buildings, land transport, agriculture, waste, and fuel production. This dataset is based on multi-information harmonization and downscaling using geospatial information from multiple sources (nighttime lighting, roads, land use, population, settlements, meteorology, and point source data support temporal reconstruction and spatial refinement), various emissions inventories such as annual and monthly data from the Emissions Database for Global Atmospheric Research (EDGAR), as well as the Open source Data Inventory for Anthropogenic Carbon dioxide (ODIAC), Community Emissions Data System (CEDS), and daily estimates by country and sector. To integrate all the data, a hierarchical Bayesian component is applied to reconstruct monthly variability outside the directly observed period, while a physics-based component restricts the specific spatial allocation for each sector and preserves emissions data, while also allowing for area expansion based on decision trees. Furthermore, the integration of fossil fuel-related covariates that are underrepresented or absent in the original EDGAR fields helps reconstruct this dataset spatially and temporally. The resulting dataset contains 672 monthly layers per sector and global layers corresponding to the date range from 1970-01 to 2025-12. Comparisons with CEDS, ODIAC, EDGAR, and other studies show strong temporal and spatial consistency, while also preserving additional small-scale structure. Therefore, HBPG-CO2 provides a long-term, high-resolution, sector-specific emissions record for atmospheric inversion, Earth system modeling, carbon cycle analysis, and regional emissions assessment. The complete HBPG-CO₂ dataset is openly accessible through Zenodo at https://doi.org/10.5281/zenodo.21507026 (Antezana Lopez, 2026).
- Preprint
(19844 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2026-590', Anonymous Referee #1, 03 Sep 2026
-
AC1: 'Reply on RC1', Franz Pablo Antezana Lopez, 01 Oct 2026
1. Temporal reconstruction for 1970–1999.
We agree that relying mainly on post 2000 seasonality could limit the historical reconstruction. We therefore revised the temporal model to separate long term annual evolution from within year seasonality. Annual trends are constrained using multiple compatible inventories, while monthly variability is reconstructed only from datasets with genuine subannual information. In particular, MEIC and the daily CO₂ dataset of Li et al., covering 1970–2024 and aggregated to monthly resolution, provide additional information on seasonal variability where their geographic and sectoral coverage is compatible. An empirical hierarchical Bayesian formulation is then used to reconstruct monthly composition, while the pre 2000 period is explicitly identified as reconstructed rather than observed. We also added pseudo hindcast experiments to evaluate how well the seasonal reconstruction transfers across time.2. Independent validation.
We agree that comparisons with datasets contributing to construction should not be interpreted as independent validation. The revised manuscript now clearly separates internal consistency checks from external evaluation. Vulcan V4 is used as an independent high resolution benchmark over the United States and CAMS REG over Europe, with supplementary spatial correlations of approximately \(r=0.80\) and \(r=0.91\), respectively. The previous CEDS waste slope of 11.37 is now treated as evidence of differences in sector definitions, source coverage, and accounting rather than as a measure of model performance, and the corresponding sector crosswalk is discussed explicitly.3. “Physics guided” terminology.
We agree and have removed this term. The framework is now called MCR CO₂, Multisource Constraint based Reconstruction of CO₂ Emissions. The spatial component is described as constraint based and process informed, using sector specific spatial evidence, non negativity, bounded redistribution, eligibility constraints, and exact mass conservation. This terminology more accurately reflects the statistical and accounting nature of the reconstruction. moreover, we have done some modificaations to the proccedure to enhance results and reduce error.4. Improvement over simple interpolation.
The revised evaluation now compares MCR CO₂ with mass conserving equal allocation and interpolation baselines using the same 0.10° parent cell mass. Spatial correlation, RMSE or MAE, and relative improvement are used to quantify the added value of the multisource 0.05° reconstruction. In addition, masked support recovery experiments test whether previously unresolved emission locations can be recovered without excessive false activation. Importantly, the recurrent LSTM component is also used to identify plausible source environments in cells that are zero in EDGAR or other baseline emission fields, using sector specific covariate patterns learned from locations where emissions are already represented. Accepted locations receive emissions only through bounded redistribution, so the procedure increases spatial information without creating additional emission mass.Overall, the main contribution of MCR CO₂ is the integration of a long 1970–2025 emission record with multiple independent sources of temporal and spatial information within a single reconstruction framework. Long term trends, monthly seasonality, high resolution spatial allocation, and unresolved source support are treated separately and then reconciled under the same mass conserving accounting system. This allows the dataset not only to extend monthly anthropogenic CO₂ emissions further back in time, but also to recover spatial structures and plausible emission areas that are incompletely represented in existing global inventories. The revised manuscript includes these methodological changes, additional validation, updated figures, and supplementary comparisons. The complete revised PDF will be available within the next few days.
Citation: https://doi.org/10.5194/essd-2026-590-AC1
-
AC1: 'Reply on RC1', Franz Pablo Antezana Lopez, 01 Oct 2026
-
RC2: 'Comment on essd-2026-590', Anonymous Referee #2, 02 Oct 2026
The authors present HBPG-CO2, a global high-resolution monthly inventory of anthropogenic CO2 emissions for 1970–2025, developed through temporal reconstruction and spatial downscaling of existing inventories. However, its advantages over EDGAR and comparable products are not sufficiently demonstrated, and the accounting boundaries and assumptions underlying the temporal reconstruction and spatial allocation require clarification. Another major limitation is the insufficient independent validation and lack of quantitative uncertainty analysis, which limit confidence in the claimed improvements in accuracy. My main comments are as follows.
- The added value of HBPG-CO2 relative to its input inventories and other global emission products needs to be established more clearly. Annual totals and recent monthly profiles are largely derived from EDGAR, while the land-use, road, nighttime-light, and point-source information should be assessed against that already incorporated into existing inventories. Several comparable high-resolution CO2 inventories have already been published and incorporated extensive point-source information, such as MEIC-global-CO2 (1970-2021, Xu et al., 2024, DOI: 10.1007/s11430-023-1230-3), ODIAC2025 (1km, 2000-2024, DOI: 10.5194/essd-10-87-2018) and GID-based GRACED (Dou et al., 2022, DOI: 10.1016/j.xinn.2021.100182). The authors should clearly demonstrate the advantages of HBPG-CO2 over both EDGAR, on which it is largely based, and other comparable high-resolution CO2 emission inventories. Please distinguish improvements attributable to additional source information from those arising from the redistribution of existing estimates, and quantify their effects on seasonality, source locations, and subnational emission patterns. The introduction and results should emphasize these specific advances and reduce repetitive descriptions of global emission trends and input datasets.
- The accounting boundaries of anthropogenic CO2 and the mapping to the seven output sectors need clear clarification. The author should provide a crosswalk from EDGAR, CEDS, and the daily reference product to HBPG-CO2 using source categories, explicitly identifying excluded categories and those split across output sectors. For example, EDGAR’s monthly transport aggregate includes road, rail, shipping, and aviation, whereas HBPG-CO2 is variously described as representing road or ground transport, with aviation and shipping added separately when calculating total emissions. The allocation of industrial combustion and process emissions also needs to be specified. Besides, the author should also clarify whether CO2 emissions from open biofuel combustion and land-use-related processes are included in each output sector. For agriculture and waste, specify the CO2-emitting processes covered and distinguish their accounting boundaries from those of CH4 and N2O emissions and land carbon fluxes. These definitions are also essential for assessing the suitability of cropland or pasture as spatial allocation proxies in Figure 7, since the appropriate proxies depend on whether agricultural CO2 arises from liming, urea application, fuel combustion, residue burning, or other processes.
- EDGAR_2025_GHG already provides monthly country and sector totals for 1970–2024; limited coverage of monthly gridded products does not imply an absence of historical monthly information. Why were these temporal constraints not used, and how is extrapolating seasonal patterns learned from 2000 onward to 1970–1999 justified? The authors should specify the reference year for the 12-year extrapolation limit and clarify whether a fixed seasonal profile is applied beyond that limit.
- For the 2025 extension, the author describes how monthly shares are blended but does not explain how the annual emission totals required by Equation (6) are derived. Please specify the data and methods used to extrapolate annual emissions by country and sector and allocate these totals to source grid cells. Any assumptions that retain 2024 emission totals or spatial patterns should be stated explicitly. Regular updates are important for maintaining the long-term utility of high-resolution emission inventories. The authors should also indicate whether continued updates are planned for HBPG-CO2 and explain how new data will be incorporated into subsequent releases.
- Cropland, NDVI, precipitation, and albedo characterize agricultural conditions but do not directly determine emissions from liming, urea application, or fuel combustion; population also need not reflect the locations of waste treatment or incineration facilities. The authors should explain the process basis and weighting of the sectoral proxies and assess sensitivity to alternative combinations. For power and industry, how do the 0.1° GID data listed in Table 1 locate facilities within parent cells, and what advantage does the method offer over using facility coordinates directly? Historical allocation should also account for land-use changes and facility commissioning and retirement, rather than implicitly imposing recent spatial patterns on earlier years.
- The role of XCO2 in downscaling is ambiguous. Table 1 describes the CAMS EGG4 reanalysis as an atmospheric consistency covariate, whereas Section 2.2 states that satellite XCO2 supports spatial allocation. The author needs to identify the product actually used and explain how it enters the model. Column-averaged CO2 is influenced by atmospheric transport, background concentrations, and biospheric exchange, so it cannot be directly equated with local anthropogenic emission intensity. If XCO2 is used to determine allocation weights, the authors should explain how these influences are accounted for and assess its contribution through a sensitivity analysis comparing results with and without this covariate. If its role is only diagnostic, this should be stated consistently throughout the manuscript.
- The manuscript lacks a quantitative uncertainty assessment, which is essential for evaluating the reliability of an emission inventory and its suitability for different applications. Although Figure 4 refers to standard deviations across multiple runs, the ensemble design, number of realizations, and resulting uncertainty estimates are not reported. Uncertainties in the input annual inventory totals, monthly reconstruction, and spatial allocation should be quantified, with an explanation of how input errors and their correlations propagate through normalization and transfers between grid cells.
- The relationship between conservation at the original grid resolution and the activation of new emitting cells needs clarification. Holding each original 0.10° cell total fixed prevents within-cell disaggregation from correcting allocation errors between parent cells. Conversely, transferring emissions to an originally zero-valued cell changes the original cell totals. The authors should clarify the conservation constraints applied before and after redistribution. For example, if the grid cells involved lie in different countries, are emission transfers across national boundaries permitted?
- The allocation rules do not fully address zero emissions or missing information. Equation (1) is undefined when annual emissions are zero, while Equation (9) cannot be normalized when all sub cell baseline shares are zero. The authors should explain how these cases and missing covariates are handled, clearly distinguishing true zero emissions from missing data and absent spatial support. They should also clarify whether the numerical stabilization constant introduces small artificial positive emissions, which could be misinterpreted as newly identified emission sources.
- The claimed improvement in spatial accuracy needs to be supported by direct quantitative validation. Agreement in national totals, visual correspondence with allocation proxies, and residuals relative to EDGAR do not demonstrate more accurate source locations or grid-cell emissions. Comparisons with EDGAR, ODIAC, and uniform or sectoral-proxy disaggregation baselines should assess spatial correlation, hotspot displacement, and emission concentration across spatial scales. Besides, accuracy should be evaluated against facility records, regional inventories, or suitable observational constraints which were not used in model fitting or parameter selection.
- CEDS appears to be used for both reconstruction/calibration (Table 1 and Section 3.3) and external comparison (Section 4.2). Its role in each step and the independence of the evaluation need clarification. Also, figure 9 also reports symmetric mean absolute percentage errors of 104.2% for agriculture and 140.6% for waste, with a regression slope of 11.37 for waste. Simply attributing these discrepancies to differences in sector definitions is insufficient.
- Figure 7 should include emission color scales and units, with identical ranges for HBPG-CO2 and EDGAR and clear distinctions between zero, missing, and newly activated cells. Figure 11a labels emissions as annual means but uses monthly units.
Citation: https://doi.org/10.5194/essd-2026-590-RC2
Data sets
HBPG-CO2: a global gridded monthly anthropogenic CO₂ emission dataset at 0.05° resolution for 1970_01–2025_12 Franz Pablo Antezana Lopez https://zenodo.org/records/21507026
Model code and software
Code and data sample of HBPG-CO2: a global gridded monthly anthropogenic CO₂ emission dataset at 0.05° resolution for 1970_01–2025_12 Franz Pablo Antezana Lopez https://zenodo.org/records/21523272
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 482 | 244 | 172 | 898 | 167 | 133 |
- HTML: 482
- PDF: 244
- XML: 172
- Total: 898
- BibTeX: 167
- EndNote: 133
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents HBPG-CO2, a global monthly CO2 emission dataset at 0.05° for 1970–2025 across seven sectors, constructed by reconstructing monthly shares from EDGAR using a hierarchical Bayesian formulation and downscaling the 0.10° EDGAR grid to 0.05° with covariate-weighted allocation under mass conservation. My main concern is that the dataset is a statistical reprocessing of established inventories with unconstrained uncertainties. It does not introduce new activity data, emission factors, or independently produced regional inventories.
Specific comments