Understanding of the processes governing soil organic carbon turnover is confounded by the fact that C feedbacks driven by soil erosion have not yet been fully explored at large scale. However, in a changing climate, variation in rainfall erosivity (and hence soil erosion) may change the amount of C displacement, hence inducing feedbacks onto the land C cycle. Using a consistent biogeochemistry-erosion model framework to quantify the impact of future climate on the C cycle, we show that C input increases were offset by higher heterotrophic respiration under climate change. Taking into account all the additional feedbacks and C fluxes due to displacement by erosion, we estimated a net source of 0.92 to 10.1 Tg C year−1 from agricultural soils in the European Union to the atmosphere over the period 2016–2100. These ranges represented a weaker and stronger C source compared to a simulation without erosion (1.8 Tg C year−1), respectively, and were dependent on the erosion-driven C loss parameterization, which is still very uncertain. However, when setting a baseline with current erosion rates, the accelerated erosion scenario resulted in 35% more eroded C, but its feedback on the C cycle was marginal. Our results challenge the idea that higher erosion driven by climate will lead to a C sink in the near future.

The latest projections of soil organic carbon (SOC) changes from different Earth system models (ESMs) show a very wide range of values, adding uncertainty on land-atmospheric carbon (C) feedback under climate change (1). Because the soil is the largest terrestrial C pool, different representations of SOC turnover may amplify the magnitude and even the sign of the C balance (2). Recent studies suggest that the general soil C sink simulated by 2100 is too optimistic, as ESMs underestimate the mean age of SOC (3). Moreover, the level of confidence of current projections and the possibility of improving process descriptions of SOC dynamics may be further limited by the exclusion of relevant processes, such as soil erosion (4). This erosional process not only redistributes SOC across the landscape but also generates different feedbacks (5, 6). In eroding landscape positions, the C that is laterally displaced can be partially replaced with new photosynthates stabilized by the less C-saturated subsoil exposed by erosion, generating a continuous C sink (that is, dynamic replacement) (7).

Our modeling framework has some limitations. Soil erosion by water is a selective process that redistributes different size particles across the landscape, in turn potentially changing the environmental conditions affecting plant growth and SOC decomposition (33). In addition, the displaced C is associated to different mineral/particle fractions (for example, C bound to mineral particles, C in aggregates, and particulate C) (34), and the turnover of these different fractions is not yet implemented into the large-scale models, which are still currently based on the compartment approach of “conceptual” C pools (35). We estimate that the development of full sediment delivery models coupled with new generations of biogeochemical models based on measurable C fractions will take another decade. To do this, we will need more information and increased process understanding to transfer microscale processes, such as aggregate breakdown, to millions of hectares. Meanwhile, we could account for lateral C fluxes with feasible approaches, as we have described here. Future land C projections rely on complex Earth system and biogeochemical models that do not incorporate the erosion process, lowering the level of confidence of our predictions. We have attempted to include the main lateral and vertical C fluxes as driven by a climate change projection to increase the confidence of future projections of soil C erosion on the C cycle.