Forest expansion dominates China's land carbon sink since 1980

The study addresses long-standing uncertainties in China's land-use and cover-change (LUCC) history and its impact on the national terrestrial carbon (C) balance. Existing global LUCC reconstructions (e.g., LUH2 built from FAO reports and HYDE modeling) inadequately constrain forest dynamics and inherit biases from inconsistent national survey methods in China, leading to conflicting estimates of cropland and forest area changes and LUCC-induced C fluxes. Prior estimates of LUCC-related C emissions in China diverge by factors of 3–5 due mainly to differing LUCC data and simplified representation of forest dynamics that do not separate planted forests (PF) from natural forests (NF). The research aims to (1) reconstruct an improved, bias-corrected LUCC database for China by harmonizing multiple historical and remote sensing sources, (2) quantify China's terrestrial C stock changes since 1900 and particularly since 1980, and (3) attribute observed C changes to LUCC, climate, nitrogen (N) deposition, rising CO2, and forest management using a process-based model. The central hypothesis is that forest expansion since 1980 has dominated China's land C sink and that previous FAO/LUH2-based assessments underestimated LUCC contributions due to biased land-cover signals.

The standard LUCC forcing for carbon models, LUH2, derives national agricultural areas from FAO and spatializes with HYDE, but is not constrained by forest area and can misallocate cropland regionally. Historical databases for China have been inconsistent due to changing survey methods and policies influencing reported cropland areas (e.g., extrapolation methods in CAY before 1982; transitions to LRSY and NLRB leading to apparent abrupt cropland increases). Prior studies report widely varying LUCC-induced cumulative C emissions for China (e.g., 6.18–33 Pg C from 1700–2000) using similar bookkeeping models, indicating datasets are the dominant source of discrepancy. Forest area trajectories also vary across datasets, with some indicating net loss and others net gain over the 20th century. Additionally, most modeling studies did not distinguish PF from NF, despite their differing C dynamics under management. Multi-model intercomparisons (MsTMIP, TRENDY) have provided LUCC attribution frameworks but rely on LUH2-based forcing and thus may propagate the same LUCC biases for China.

Data development: The authors reconstructed historical, gridded LUCC (0.5°) for China from 1900–2019 by harmonizing multiple sources: high-resolution satellite products (e.g., GlobeLand30), vector maps (1980s), and tabular provincial statistics (1949–2018). Lakes, rivers, and barren areas from GlobeLand30 were assumed static since 1900. Impervious surface maps (1978–2017) were resampled, with 1900–1977 and 2018–2019 held at 1978 and 2017 levels, respectively. A top-down allocation model assigned annually determined provincial areas for cropland, forest, and wetland to grid cells, assimilating land conversion signals from reports, field surveys, and satellite images. Provincial forest and cropland areas were benchmarked to the most authoritative records and validated against National Forest Inventories (NFI), the National Forestry and Grassland Data Center, and other publications; spatial distributions were compared to HYDE and LUH2-GCB for 1980, 1990, and 2018. Forest detail: Forests were separated into natural (NF) and planted (PF). Annual provincial NF and PF areas (1949–2018) came from NFI and literature; earlier periods were interpolated/extrapolated from historical records. Spatial distributions of NF, PF, and total forest were reconstructed for base years (1900–2019), with linear interpolation between base years. Forest age and type maps for NF and PF were adopted from prior work. Harvesting was derived from LUH2 land-transition data (0.5°, 1900–2019), partitioned between NF and PF using time-varying harvesting ratios compiled from yearbooks and publications; where data were missing (pre-1949 and post-2004), assumptions reflected earliest available ratios and policy-driven shifts toward PF harvesting. Other drivers: Crop rotation maps (1980–2011) were used and held constant outside the range. Historical N fertilizer rates were compiled from FAO and literature; manure applications were from global gridded datasets. Daily climate (temperature and precipitation) was reconstructed from 839 stations and gridded products (1900–2019) with Anusplin interpolation for recent decades; shortwave radiation combined high-resolution datasets (2000–2015) and MSTMIP products (1901–1983). Atmospheric CO2 and N deposition were from IPCC and MsTMIP drivers, with updated N deposition (1996–2015) downscaled and used to adjust earlier years. Modeling and experiments: The Dynamic Land Ecosystem Model (DLEM), a process-based biogeochemical model used in MsTMIP and TRENDY, simulated carbon dynamics at 0.5° daily time steps (1900–2019), driven by the reconstructed LUCC, climate, atmospheric chemistry (CO2, N deposition), forest management, and agricultural management. Equilibrium initial conditions were obtained such that interannual net flux variations in C, N, and water met small-threshold criteria, followed by a 10-year spin-up and transient simulations. Three experiment groups quantified (1) historical LUCC impacts (1900–2019), and (2) direct and interactive effects of LUCC, climate, CO2, N deposition, N fertilizer/manure, and forest management for 1980–2019. Group-2 fixed one factor at 1980 levels while varying others; Group-3 varied one factor while fixing the rest at 1980 levels, enabling separation of direct versus interactive contributions. Additional DLEM simulations using LUH2-GCB forcing isolated the role of LUCC data. Uncertainty analyses varied parameters (e.g., instantaneous C emissions from LUCC), cropland residue return, and NF/PF management assumptions. Model calibration/validation used extensive field campaign data (2011–2015) for biomass and soil C.

- New LUCC reconstruction for China corrects biases in FAO-derived datasets (CAY, LRSY, NLRB transitions) that introduced spurious cropland increases in the 1980s–2010 period. Compared to LUH2-GCB, cropland area changes since 1980 diverge strongly (−14 Mha in this study vs +41 Mha in LUH2-GCB). - Forest area increased by 58 Mha from 1900 to 2019; other databases underestimated post-1980 forest expansion by 38–102 Mha. Spatially, widespread forest gains in the new dataset contrasted with forest declines in LUH2-GCB for 1980–2019. - China’s terrestrial ecosystems accumulated 8.9 ± 0.8 Pg C during 1980–2019 (model driven by new LUCC data). LUCC contributed nearly 44% of the national land C sink (abstract) and, in factorial attribution, accounted for 50.3% (3.96 Pg C) of total C increase with 18.1% (1.43 Pg C) from interactions. - LUCC impacts: 1900–2010s, LUCC caused a C loss of 5.1 ± 0.7 Pg C (this study), far lower than MsTMIP (13.8 ± 7.7 Pg C, 1900–2010) and TRENDY (9.4 ± 3.3 Pg C, 1900–2019). From 1980 onward, LUCC increased C storage by 4.3 ± 0.7 Pg C (dominated by biomass gains in southwest and northeast China), whereas MsTMIP and TRENDY simulated continued LUCC-induced C losses of 7.5 ± 1.6 and 5.3 ± 2.3 Pg C, respectively. - Using LUH2-GCB as LUCC forcing in DLEM reproduced larger C losses (11.4 ± 0.6 Pg C for 1900–2019; 6.5 ± 0.4 Pg C for 1980–2019), confirming LUCC forcing is the primary cause of model discrepancies. - Attribution since 1980: Climate change contributed a net 1.41 Pg C (biomass +1.63 Pg, soil −0.30 Pg) ≈ 18.0% of total increase; other factors (N deposition, CO2 fertilization, N fertilizer/manure, forest management) contributed smaller shares (each ~0.1–9.54%), while interactions accounted for 18.1% (1.43 Pg C). - Biome-specific LUCC effects (1980–2019): Forest expansion dominated net C accumulation; cropland, grassland, shrubland, and wetland changes were relatively small (−0.3 to +0.3 Pg C). Biomass increases accounted for ~76.3% of terrestrial C gains, with NF and PF contributing ~65% (2.9 Pg C) and ~35% (1.6 Pg C) of biomass accumulation, respectively. - The improved LUCC dataset narrows regional uncertainties and suggests previous studies underestimated LUCC’s contribution to China’s land C sink due to biased land-cover signals.

The findings demonstrate that corrected LUCC histories, particularly the magnitude and spatial distribution of forest expansion since 1980, fundamentally alter estimates of China’s land carbon sink and its drivers. Biases in FAO-reported cropland areas propagated into LUH2-GCB, causing models to infer spurious forest loss and to overestimate LUCC-induced carbon emissions. With the new LUCC forcing, DLEM indicates a strong post-1980 carbon sink consistent with extensive afforestation/reforestation and forest management, and shows LUCC as the dominant driver over climate, CO2, and N deposition. Comparisons with MsTMIP and TRENDY and DLEM experiments using LUH2-GCB confirm that discrepancies arise primarily from LUCC forcing, not model structure. The results resolve debates on whether LUCC contributed a sink or source in recent decades, providing evidence of a substantial LUCC-driven sink centered in forest biomass accumulation. These insights emphasize the centrality of accurate LUCC reconstructions for carbon budget accounting, attribution studies, and policy assessments, suggesting that afforestation and improved forest management are critical levers for climate mitigation in China.

This work delivers a bias-corrected, multi-source LUCC reconstruction for China (1900–2019) and shows that forest expansion since 1980 has dominated the national land carbon sink. Using the new LUCC forcing in a process-based model, the study quantifies a strong 1980–2019 sink (8.9 ± 0.8 Pg C), attributes roughly half of the increase to LUCC (with notable interactive effects), and highlights the leading role of forest biomass accumulation. The analysis explains discrepancies with MsTMIP and TRENDY as consequences of LUCC forcing biases and underscores the importance of reliable LUCC data in regional and global carbon budgets. Future research should continue to refine global LUCC datasets by integrating improved survey methods and remote sensing, represent NF and PF management more explicitly, reduce uncertainties in harvest and management histories, and extend attribution frameworks that cleanly separate LUCC from vegetation structural changes (e.g., LAI). Policy-wise, sustained reforestation and improved forest management offer effective pathways to advance China’s carbon neutrality goals with biodiversity co-benefits.

- LUCC reconstruction relies on assumptions where data were sparse: lakes/rivers/barren held constant since 1900; impervious surfaces assumed constant outside 1978–2017; cropland rotations fixed pre-1980 and post-2011; early (1900–1948) forest areas interpolated; PF areas before 1973 extrapolated from historical records. - Forest harvesting partition between NF and PF uses time-segmented ratios; pre-1949 ratios assumed equal to 1950–1962; post-2004 trends assumed (reduced NF, increased PF harvesting) due to policy—introducing uncertainty. - Forest management practices (type, intensity, spatial distribution) are poorly constrained; modeling assumed generic enhancements (e.g., +20% N uptake) and standardized fertilization/irrigation schedules for PF, which may not reflect local practices. - Attribution percentages differ depending on inclusion of interactive effects and experiment baselines; comparisons with MsTMIP/TRENDY are affected by differing experimental designs and forcings. - Despite extensive validation, uncertainties remain from model parameters (e.g., instantaneous emissions), cropland residue return, and driver datasets (e.g., N deposition reconstructions, climate forcing in early decades).