摘要:While the global carbon budget (GCB) is relatively well constrained over the last decades of the 20th century [1], observations and reconstructions of atmospheric CO2 growth rate present large discrepancies during the earlier periods [2]. The large uncertainty in GCB has been attributed to the land biosphere, although it is not clear whether the gaps between observations and reconstructions are mainly because land-surface models (LSMs) underestimate inter-annual to decadal variability in natural ecosystems, or due to inaccuracies in land-use change reconstructions. As Eurasia encompasses about 15% of the terrestrial surface, 20% of the global soil organic carbon pool and constitutes a large CO2 sink, we evaluate the potential contribution of natural and human-driven processes to induce large anomalies in the biospheric CO2 fluxes in the early 20th century. We use an LSM specifically developed for high-latitudes, that correctly simulates Eurasian C-stocks and fluxes from observational records [3], in order to evaluate the sensitivity of the Eurasian sink to the strong high-latitude warming occurring between 1930 and 1950. We show that the LSM with improved high-latitude phenology, hydrology and soil processes, contrary to the group of LSMs in [2], is able to represent enhanced vegetation growth linked to boreal spring warming, consistent with tree-ring time-series [4]. By compiling a dataset of annual agricultural area in the Former Soviet Union that better reflects changes in cropland area linked with socio-economic fluctuations during the early 20th century, we show that land-abadonment during periods of crisis and war may result in reduced CO2 emissions from land-use change (44%–78% lower) detectable at decadal time-scales. Our study points to key processes that may need to be improved in LSMs and LUC datasets in order to better represent decadal variability in the land CO2 sink, and to better constrain the GCB during the pre-observational record.