Contrasting soil carbon stock changes across European agricultural ICOS ecosystems

Measuring soil organic carbon (SOC) stock changes across space and time is crucial for comprehending ecosystem responses to climate variability, land-use change, and agricultural management. However, long-term comparisons remain challenging due to differences in sampling designs, analytical methods, and data structures.

From an integrative approach combining soil depth harmonization, equivalent soil mass method, and statistical analyses, we quantified soil organic carbon (SOC) stock changes across contrasting ecosystems within the Integrated Carbon Observation System infrastructure (ICOS). Historical and labeled? ICOS soil data were harmonized using several complementary approaches. Soil depth limits were harmonized using cubic spline interpolation and a non-model approach based on a 1-mm depth discretization and soil organic carbon (SOC) stocks were compared using equivalent soil mass approach. The comparisons between sampling campaigns were further constrained by accounting for the spatial occurrence of sampling points within similar soil types, defined by comparable coarse fragment contents and similar soil texture. The design-based approach was applied to compare SOC stocks between sampling campaigns. Differences in SOC stocks were assessed using Welch’s t-test, which does not assume equal variances.

Results show that cumulative soil organic carbon (SOC) stock changes in the ~0-60 cm layer are strongly site-dependent, with no consistent trend observed across sites sharing the same land-use type. At the Grignon station (FR-Gri), SOC stocks decreased significantly by 950 ± 40 g C m^-2 over 13.2 years (2005-2019), and at the Estrées-Mons A28 station (FR-EM2), a significant SOC loss of 318 ± 145 g C m^-2 was observed over 6 years (2015-2021). In contrast, no detectable SOC stocks changes were observed at the Lamasquère station (FR-Lam, 2015-2020), the Lonzée station (BE-Lon, 2007-2017), and the Klingenberg station (DE-Kli, 2008-2019), although SOC stock changes at the latter site showed a non-significant tendency towards an increase (p = 0.07). Minimum detectable difference analysis demonstrates that only SOC losses at FR-Gri can be robustly detected with the current sampling design, while changes at other sites remain below detection limits, underscoring the importance of accounting for methodological sensitivity in long-term SOC assessments.

These differences among agricultural sites may be not related to land use per se, but rather to site-specific management practices, particularly the balance between organic carbon imports and exports, nitrogen fertilization, and soil type. At the Grignon station, observed SOC losses were consistent with simulations from the AMG model. Similarly, at the Lonzée site, the differences between soil inventories fell within the range simulated by RothC, indicating agreement between measured and modeled SOC stocks and supporting near-equilibrium conditions.

Overall, SOC stock changes were highly site-specific, with SOC losses detected only at Grignon, while changes at other sites were close to or below detection limits despite harmonizing the datasets obtained with different sampling methods. These results highlight the need for uncertainty-aware interpretations and monitoring designs optimized for detecting long-term SOC changes.