Mapping Global Terrestrial Spatiotemporal Dynamics of Atmospheric CO₂ Carbon Isoscapes for Ecological Studies

The isotopic composition of atmospheric carbon dioxide (δ¹³C_atm) provides insights into the terrestrial carbon cycle. However, long-term global δ¹³C_atm maps with both high spatial resolution and continuous temporal coverage remain scarce. Here, we present a new global terrestrial dataset of monthly δ¹³C_atm isoscapes from 2001 to 2020 at 0.05° spatial resolution, developed by integrating in situ observations with optimized 4D CO₂ concentration fields from inversion outputs, reanalysis data, and geographic information using machine learning. Among four tested models, the Gradient Boosting Machine demonstrated the highest predictive performance under random validation (R² = 0.80, RMSE = 0.12‰) and maintained robust performance across three spatially independent validation frameworks (R² = 0.56-0.67, RMSE = 0.16‰-0.19‰). Key predictors were air temperature and atmospheric CO₂ concentration. The resulting global terrestrial isoscapes reveal strong spatial and temporal heterogeneity. Model predictions closely align with National Oceanic and Atmospheric Administration (NOAA) marine boundary layer (MBL) observations in terms of trend magnitude, seasonal amplitude (< 0.13‰ deviation), and latitude gradient (< 0.2‰). In our study, δ¹³C_atm seasonal amplitude varies from 0.06‰ in Southern Hemisphere mid-latitudes to 0.6‰ in Northern Hemisphere high latitudes, indicating strong hemispheric asymmetry. Moreover, over 99.8% of global terrestrial grid cells show negative trends in all seasons, with a global terrestrial average annual depletion rate of -0.030‰ ± 0.0006‰ year^-1. The trend shows stronger depletion during summer and autumn, reaching its peak in August (-0.035‰ ± 0.0012‰ year^-1), while spring and winter seasons remain comparatively stable. This study delivers a long-term, high-resolution global terrestrial δ¹³C_atm isoscape dataset, offering a valuable tracer for carbon cycle research and, importantly, robust data support for large-scale investigations of carbon-water coupling in terrestrial ecosystems.