Global daily 1 km gapless XCO₂ (2003−2023) derived from multi-satellite observations and a spatiotemporal deep learning framework

Abstract

In recent years, atmospheric CO₂ concentrations have continued to rise, driving global warming and precipitating severe environmental and ecological crises. Spatiotemporal monitoring of column-averaged dry-air mole fractions of CO₂ (XCO₂) from satellite observations is indispensable for quantifying carbon sources and sinks, evaluating mitigation efforts, and elucidating the global carbon cycle. However, existing satellite XCO₂ products are severely limited by narrow swaths, adverse meteorological interference, and inconsistent spatiotemporal coverage across different platforms, resulting in sparse retrievals with substantial gaps. This study developed a spatiotemporal deep learning framework that couples ConvLSTM temporal modules with a U-Net spatial backbone, enhanced by residual connections and channel- and spatial-attention blocks. Using the column-averaged dry-air mole fraction of XCO₂ data of SCIAMACHY, GOSAT, and OCO-2, we derived the first gapless, daily global terrestrial XCO₂ dataset at 1 km resolution spanning 2003–2023. The results showed outstanding performance under three cross-validation strategies—sample-based, grid-based (spatial), and daily-based (temporal)—with R² = 0.996/0.963/0.952, RMSE = 0.46/0.62/0.73 ppm, and MAPE = 0.085 %/ 0.096 %/ 0.101 %, respectively. Moreover, independent validation against in situ TCCON measurements also confirms excellent agreement (R² = 0.988, RMSE = 1.10 ppm, MAPE = 0.216 %). Compared to previous efforts, this framework delivers significant improvements in both spatiotemporal resolution and predictive accuracy. The resulting full-coverage estimates reveal a global terrestrial XCO₂ increase of 2.22 ppm yr⁻¹ over the study period, with markedly larger seasonal amplitudes in the Northern Hemisphere. The high spatiotemporal fidelity captures rapid, fine-scale XCO₂ fluctuations—such as urban–suburban gradients and diurnal evolutions—that remain unresolved by previous coarser datasets. This seamless and high-quality dataset will provide a robust foundation for future global and regional “dual‑carbon” policy implementation and climate-change research. The dataset can be freely accessed at https://doi.org/10.11888/Atmos.tpdc.302399.