Forest expansion dominates China’s land carbon sink since 1980

Historical land use and cover changes

Existing databases differed significantly in representing historical LUCC in China (Fig. 1). Generally, datasets agree on the direction of change in cropland area until 1980 in Liu and Tian¹⁸, Ramankutty¹⁹, Houghton²⁰, and this study (Fig. 1b, c), while the magnitude of change varied greatly. Specifically, the total cropland expansion in China was comparable between our new data set and the LUH2-GCB from 1900 onwards (56 vs 60 Mha, Fig. 1b), but cropland area changes since 1980 diverged considerably (−14 vs 41 Mha, Fig. 1c). The differences were also evident across space and more distinct during the period of 1980 to 2019 (Fig. 2a–d), in which the cropland coverage was mainly declining in our reconstructed data but increasing in LUH2-GCB (Fig. 2b, d). We found that the distinct changes are derived from the abrupt cropland increases in the FAO data reported from China, upon which LUH2-GCB was based (see Supplementary Information 3).

The problems of cropland area expansion reported to FAO are likely caused by changes in the underlying database, in which the Chinese Agricultural Yearbook (CAY) was used prior to 1996, the China Land and Resources Statistical Yearbook (LRSY) from 1996 to 2007, and the National Land and Resources Bulletin (NLRB) after 2007 (Supplementary Information 3).

These three datasets are not consistent with each other because surveying methods were distinct. For example, cropland area in CAY before 1982 used an extrapolation method (i.e. “production-to-acreage” approach) due to limited field survey data¹¹. Specifically, the extrapolation method was widely adopted for convenience and for taxation purposes in the early period, such as in the framework of the first benchmark cropland survey conducted in 1953. Such methods assumed that low-productivity cropland occupied an area of 1/3–1/8 of a predetermined, “standard-productivity” cropland²¹, which greatly underestimates the acreages of low productivity cropland. Biases accumulated in this method persisted until the satellite era (1980s), while the 1953 surveying data were used as the baseline for CAY to update cropland area on an annual basis.

Besides the survey method, policies also contributed to a bias of reported cropland area. To tackle rising food demands, cropland expansion was highly encouraged by the government before the 1980s, implementing an incentive policy to allow new tax-free cropland without reporting to the government for the first 3–5 years^22,23. Even after the initial reporting free period, these newly cultivated croplands continued to be unreported due to political incentives to show increasing crop yield to the local authorities^23,24.

When the first comprehensive and systematic survey (i.e. the second national cropland survey conducted during 1985–1996) was completed, the cropland area was found to be larger than previously reported in CAY¹¹. Similarly, the shift from the use of LRSY to NLRB also introduced a spurious cropland area increment from 2007 to 2010 as small, fragmented croplands were identified by better technologies adopted in NLRB, which had remained undetected previously (Supplementary Fig. S10).

Thus, LUH2-GCB has inherited spurious temporal signals of abrupt cropland increment in FAO from the 1980s to 2010 (Fig. 1a and Supplementary Fig. S10). Therefore, if the areas of other land cover types (e.g. forest) are indirectly constrained from cropland area change, cropland area biases were mirrored in the area change of other land use types. This is the case for the LUH2-GCB and for Liu and Tian’s previous land use gridded datasets. Our new database, rebuilt from Yu et al.¹¹, corrected these problems in temporal dynamics by assimilating multiple data sources (Fig. 1a). More specifically, we retrospectively reconstructed information about cropland and forest areas year by year, using tabular data from official agencies (Supplementary Information 1 and Supplementary Data 1). To further reduce the aforementioned biases, we used the most recent and authoritative record of provincial cropland and forest areas available as the benchmark, and then spatialized the cropland and forest distributions using gridded maps as ancillary data (Supplementary Information 1). The area changes were also validated using inventory-based benchmark data (Fig. 1a, d, details were presented in Yu et al.¹¹ and Supplementary Information 1.2).

Changes in forest area in China also varied dramatically among databases. Based on Ramankutty and Foley¹⁹ and LUH2-GCB, a net forest loss was found from 1900 to the last available year, at 33–108 Mha whereas Liu and Tian¹⁸ and Houghton and Nassikas²⁰ reported a net increase of 15 Mha (1900–2005) and 70 Mha (1900–2015) in forest area, respectively (Fig. 1d, e).

By assimilating multiple source records, reports, and national surveys, however, our newly reconstructed and intensively validated database (Supplementary Figs. S4, S5, and S8) with corrected biases suggests that the forest area increased by 58 Mha from 1900 to 2019 (Fig. 1e). In particular, our data suggest that there is a surprisingly large underestimation of forest expansion in all other databases (38–102 Mha) after 1980 (Fig. 1f). We performed spatial analyses and show that widespread forest expansion in our reconstructed data was represented as a forest decline in LUH2-GCB during the period 1980–2019 (Fig. 2f, h). These existing biases in the dataset during the last four decades can be simply removed using recently available and spatially explicit forest products (Supplementary Table S2).

Bias in forest change might be explained by two reasons. First, gridded datasets inherited and transferred errors from the use of FAO-based cropland dataset in developing global land use databases such as HYDE and thus LUH2-GCB⁸. Second, the FAO forest area reported is an important reference data used in these databases. The FAO forest area is reported based on a “land use” definition, which underestimated gross “land cover” change signals between reported years (Supplementary Information 1.3). Specifically, the FAO forest area describes lands that have been forested and will continue to be used for forestry (e.g. cut-over area, fired-over area, unestablished afforestation land) (Supplementary Table S5). This approach overestimates forest area by including lands used for reforestation where no forest was yet created. Thus, for example, the FAO statistics reported a 157.2 Mha forest area in 1990 (Supplementary Fig. S7), which is ~30 Mha higher than officially released data.

More importantly, newly established forests were underestimated in such an accounting approach. The forest area expansion in China reported in the FAO statistics was 61 Mha from 1990 to 2019, which is 30 Mha lower than the officially released data¹⁶. Our reconstructed dataset, in agreement with officially released forest area, uses a “land cover” definition that characterizes the distribution of annually established forests. Therefore, the FAO statistics – a data set with definition specified to describe the area of land use – should be used with caution for constraining the temporal evolution of forest cover distribution in gridded data reconstruction, and the modeling community should be alerted to treat the LUCC data appropriately.

Nonetheless, the FAO and the related LUH2 products were the dominant LUCC forcing data used in multiple studies^3,25, including various process-model-based intercomparison projects (e.g. MsTMIP, LUMIP, NMIP, TRENDY), annually released Global Carbon Budget reports^2,26, and IPCC reports⁵, implying a potential bias of these assessments for the China region. In contrast, changes in forest area from our database were independently developed (Supplementary Information 1.2), intensively calibrated, and validated using officially released national forest inventories (NFIs, see Supplementary Figs. S4 and S5), which can help to reduce the potential bias of C balance assessment in China. More specifically, the total forest area and PF area in our database were compared with historical NFIs released by the National Forestry and Grassland Administration at provincial level since 1949 (Supplementary Figs. S4 and S5), which supports the reliability of our reconstructed data.

Historical carbon stock changes

To illustrate the bias in the C balance of China when using previous LUCC dataset, we performed simulations with the DLEM model for the period 1900–2019 at a resolution of 0.5 × 0.5 degree forced by our new LUCC dataset. We validated the distribution and changes of C stock using published studies and previously reported inventory-based estimations (Supplementary Information 6 and 7). The model could capture well C dynamics in China using inventory-based forest C stock changes at both provincial and national levels as the validation data set (Supplementary Fig. S14).

Our results show that the total C stock decreased by 6.9 ± 0.6 Pg from 1900 to 1980 and increased by 8.9 ± 0.8 Pg C from 1980 to 2019 (Fig. 3, derived from experiment S1 in Supplementary Table S10). Such a large C stock increment since the 1980s, which is dominated by vegetation biomass C accumulation, was not captured in the MsTMIP and TRENDY projects driven by different versions of the LUH2 data (Fig. 3). This is attributed to the fast expansion of forest area(s) that was not captured by this land use forcing (Fig. 1).

We found that the large-scale forest expansion in China alone has caused a substantial C accumulation since 1980 (0.21 ± 0.006 Pg C per year, Table 1). In contrast, the forest C sink of the TRENDY models is negligible (−0.02 ± 0.05 Pg C per year, Table 1). A moderate C source (0.10 ± 0.08 Pg C per year, Table 1) was even found in the MsTMIP models, since these models were driven by continuous forest area loss and cropland expansion since the 1980s (Supplementary Fig. S7).

A recent atmospheric inversion-based study reported that China’s land ecosystems were a large CO₂ sink of −1.11 ± 0.38 Pg C per year²⁷, which seems to be ecologically implausible and critically sensitive to the assimilation of the CO₂ record from one station²⁸. The compilation of previous studies from inventory- and satellite-based estimation, atmospheric inversion, and process-based models suggested that the Chinese C sink was much smaller (−0.18– −0.45 Pg C per year; Table 1). Our model-simulated terrestrial sink (~−0.28 ± 0.06 Pg C per year) was in this range (Table 1).

While our simulated C balance in different categories or biomes is close to previous estimations, three major differences are observed (Table 1). First, because the LUCC data used in previous global models suffered from biases as shown above, the national C sink was generally underestimated in these simulations (Table 1). Second, our estimation of the forest sink is around two to three times larger than the previous one during 1949–1998²⁹. This was mainly because forest area was underestimated by over 33% (53 Mha) in the previous study²⁹ compared to the national forest inventory (NFI)¹⁶. This underestimation may stem from exclusion of economic and bamboo forests. The third major difference is the role of grassland soils in C balance during the period 1980–2000. China’s grassland soils were previously reported as a minor sink of −0.007–−0.022 Pg C per year from the 1980s to the 2000s (Table 1), while our simulations suggest that grassland soils were a C source of 0.062–0.066 Pg C per year. This discrepancy lies in the approaches used and the accounting boundaries between studies (i.e. whether the transitions of grassland were considered), in which LUCC impacts were represented differently. For example, impervious surfaces (part of urbanized area) expanded into ~15 Mha of natural lands in China from 1978 to 2017³⁰, which further drove redistribution of cropland into marginal lands with the majority converted from grassland, causing wind erosion, habitat loss, and more water and fertilizer consumption³¹. Earlier studies using a static grassland map exclude the C stock loss in the land-use transition³². Thus, the distinct roles of grassland soils (i.e. sink vs source) derived from our simulations and earlier studies are not contradictory but are due to differences in accounting boundaries.

LUCC impacts on carbon stock changes

Our DLEM simulation indicates that LUCC induced a C loss of 5.1 ± 0.7 Pg C from 1900 to 2010s (Fig. 4a), which is substantially lower than that from MsTMIP (13.8 ± 7.7 Pg C, 1900–2010) and TRENDY (9.4 ± 3.3 Pg C, 1900–2019; Fig. 4e, f and Supplementary Fig. S18d, g). From 1980 onward, LUCC increased C storage by 4.3 ± 0.7 Pg C, with the major contribution from vegetation biomass C increment in the southwestern and northeastern regions (Fig. 4d and Supplementary Fig. S19a). Nonetheless, this C increase in biomass was not captured in MsTMIP and TRENDY models (Fig. 4e, f and Supplementary Fig. S19d, g), which simulated that LUCC continued to reduce C stock by 7.5 ± 1.6 and 5.3 ± 2.3 Pg C during the period 1980 to the 2010s, respectively (Fig. 4 and Supplementary Fig. S20).

To confirm that such discrepancy was induced by LUCC data but not the DLEM model, we set up additional DLEM simulations using the LUH2-GCB database (Supplementary Information 8). The simulated C losses induced by LUCC when DLEM was driven with LUH2-GCB were 6.5 ± 0.4 and 11.4 ± 0.6 Pg C during the periods of 1980–2019 and 1900–2019, which are close to MsTMIP and TRENDY simulations. These results confirm that the LUCC forcing database is the major contributor to the difference between our simulations and the MsTMIP and TRENDY projects. An earlier study reported that global LUCC-induced C emissions are substantially underestimated due to underrepresented tree harvesting and land clearing from shifting cultivation³³. Our simulation revealed that regional LUCC-induced C emission could also be overestimated in China due to a bias in the LUCC data.

There are also disputes over whether the LUCC induced a C sink in China since the 1990s or not (Supplementary Table S8). By using an updated LUCC database, our simulations revealed that LUCC was a strong C sink in China, and that its magnitude was larger than previous estimates since the 1990s (Supplementary Table S8). Our results using an improved LUCC forcing data can facilitate narrowing down the well-known, large uncertainty in LUCC-induced C change at regional scale.

Attributions of different factors on C stock changes since 1980

By using the DLEM model with factorial simulations (see Supplementary Information 8 for details), we examined the direct and interactive contributions of different drivers to terrestrial C stock change in China for the period 1980–2019, including LUCC, climate, forest management, N deposition, and CO₂ fertilization (see Methods, Fig. 5). Note that historical C stock change is not equivalent to the sum of factorial attributions as the baseline conditions differ (see Supplementary Information 8).

Overall, 81.9% (6.5 Pg C) of the terrestrial C sink during this period was attributed to direct impacts of all major factors, while the interactive effect contributed 18.1% (1.43 Pg C; Fig. 5c). Among all the factors examined, LUCC was the dominant driver accounting for 50.3% (3.96 Pg C) of the total C increment during the period 1980–2019 (Fig. 5c), which was largely attributed to biomass C accumulation (70.0%; Fig. 5a, c). Tian et al.¹³ reported that LUCC’s contribution to the sink in China was at 0.05 Pg C yr⁻¹ since the 1980s – an amount that is only about 30% of our simulations. The discrepancy is attributed to the different representation of forest expansion in model simulations, which was 65 Mha from 1980 to 2005 in our database but only ~14 Mha in Tian et al.¹³. Similarly, the increase in the global land sink during the recent period (1998–2012) was also mainly attributed to LUCC (i.e. decreased tropical forest area loss and increased afforestation in northern temperate regions), instead of CO₂ or climate change³⁴.

Climate change enhanced biomass C stocks by 1.63 Pg but caused a soil C loss of 0.30 Pg, thus contributing to land sink of 1.41 Pg C (18.0% of the total with all factors) since 1980 (Fig. 5). Other global change factors, such as N fertilizer application, atmospheric N deposition, and rising CO₂, had a relatively minor contribution (0.1–9.54%) to the terrestrial C sink. Therefore, conversely to previous studies^13,35,36,37, we showed that LUCC was the dominant driver of the recent land C sink in China, and other factors including climate change, rising CO₂, and N deposition, contributed much less (0.1–18.0%) to the C stock increment in China (Fig. 5c). Tian et al.¹³ pointed out that LUCC effects in China should not be ignored and that the CO₂ fertilization effect might be overestimated in Piao et al.³⁸.

Our simulations confirm these statements, and further show that LUCC was actually the largest contributor to land sink in China since 1980 (Fig. 5). In those studies which did not account for the influence of LUCC separately, the effects of other global change factors may have been overestimated by including LUCC impacts. For example, Chen et al.³⁹ and He et al.³⁷ attributed China’s C sink into different components including climate change, leaf area index (LAI) change, rising CO₂, and N deposition. Such partition inevitably masked the separate contribution from LUCC, because LAI changes are closely related to land-cover changes. Thus, the accurate representation of the LUCC should be prioritized in future modeling attribution studies.

Carbon stock changes in each land cover type since 1980

The contribution of the establishment of young and new forest plantations to C sink has received increasing attention^3,40,41,42. Our simulation (experiment S1, see Methods section) revealed that the increase in terrestrial C stock was dominantly contributed by biomass C accumulation (76.3%) (Fig. 5), in which the natural and planted forests accounted for 65% (2.9 Pg C) and 35% (1.6 Pg C) during the last four decades. We examined the LUCC effect (i.e. the largest contributor of C stock increment in Fig. 5) on the C stock of different biomes and confirmed that forest was the major contributor of the net C accumulation in China since 1980, while other biomes, including cropland, grassland, shrubland, and wetland, were relatively stable, varying from −0.3 to 0.3 Pg C during the same period (Fig. 6). A recent study documented that forest expansion was essential for a large C sink in southern China during 2002–2017, where newly-established and existing forests contributed to 32% and 34% of land C sink in the region⁴³. In comparison to the large biomass C increase since 1980 (3.0 Pg C, Fig. 6a), the SOC increase was much lower (0.7 Pg C) during the concurrent period, although SOC changes in each biome varied greatly (–3.4–8.6 Pg C; Fig. 6b) due to area change from land conversions. The biome-level analyses further revealed that the LUCC-induced C stock increment was dominantly contributed from forest and by area expansion, while C storage in grassland and shrubland was reduced by LUCC (Fig. 6).

This study highlights the dominant role of LUCC in determining the terrestrial C sink in China. Because of inaccurate representations of land cover change in China, previous estimates of the terrestrial C sink have been strongly underestimated. In contrast, forest expansion and cropland abandonment have been overestimated in the U.S., resulting in an underestimated C emission since 1980⁷. Hence, we highlighted that the global LUCC database should be further improved, which could potentially narrow down the C imbalance reported in global C budget accounting². In contrast to the previous studies, we showed that the contributions of factors including rising CO₂, N deposition, and climate change to the land C sink in China were much smaller than LUCC over the past four decades (1980-present time). Thus, reforestation projects could represent important climate change mitigation pathways, with co-benefits for biodiversity³³. To achieve the ‘C neutrality’ goal as the Chinese government declared, future climate policy should be directed to improve land management, especially forest ecosystems.

Implications for future LUCC data improvements

This study provides a novel reconstruction of recent land use change in China and assesses its implications in quantifying for terrestrial C storage dynamics. The improved dataset more accurately depicts the spatiotemporal dynamics of LUCC in China because the historically contradictory surveying records were identified, which helped to correct the biased temporal signals. Specifically, the improved surveying methods and the socioeconomic factors have greatly shaped the LUCC signals. We advocate that these impacts should be considered in the reconstruction of the national and global LUCC dataset, especially in the areas that have been intensively disturbed by human activities as is the case of China. These endeavours will be worthwhile, as demonstrated by the large impact that these bias corrections have on China’s C dynamic assessments since 1900. Thus, accurate delineation of LUCC forcing should be stressed in global simulations, including C budget accounting, biodiversity assessments, and ecosystem services evaluations.

Source: Ecology - nature.com