Unintended consequences of combating desertification in China

China’s desertification-prone region (DPR) extends from Central Asia to northeastern China over more than 1.2 million km2, with over 60% managed by traditional pastoral and agricultural systems affecting the livelihoods of about 47.9 million people. Concerned about desertification’s impacts on ecology and food security, China has implemented major countermeasures since the early 2000s, notably the widespread grain-for-green and grazing exclusion practices supported by legal frameworks such as the Grassland Law. Investments since 2002 in these programmes have exceeded 780 billion RMB and may increase further. Yet the benefits of these ambitious practices for combating desertification and their broader sustainability impacts remain unclear. This study quantifies vegetation responses to climate variability, isolates the effects of intervention practices on vegetation trends, and evaluates impacts on agriculture, livestock production, and local incomes, thereby assessing the environmental and economic outcomes of these programmes in the DPR.

Global vegetation greening since the 1980s has been attributed mainly to climate change, agricultural advances, and CO2 fertilization, suggesting that natural and management drivers can outweigh direct ecological interventions. Prior national ecological restoration projects in China have reported carbon sequestration benefits, and debates persist about the effectiveness of measures like protective afforestation and grazing exclusion. International literature also highlights potential trade-offs between conservation measures and local poverty, underscoring the need to evaluate socio-ecological outcomes. Against this backdrop, the study positions China’s desertification-control programmes within the context of global greening drivers and prior Chinese restoration efforts to assess their incremental contributions and unintended socioeconomic consequences.

Study area delineation: The DPR boundary was compiled from CAREERI CAS products and the Desert Distribution Map of China, excluding the Tibet District, scattered southern areas, Gobi deserts, and salinized lands. The DPR totals ~122.3×10^3 km2 and was subdivided into nine subregions. Land use: 30 m CNLUCC datasets (1980s–2020) were used to detect land-use changes, identify permanent farmlands and grasslands, and delineate areas involved in grain-for-green (farmlands converted after 2000 to forest/grassland/unused) and grazing exclusion (provincial records post-2010 refined to DPR). Economic data: County-level population, disposable income, GDP, grain and livestock production were compiled from national databases; outliers/missing values were handled via linear regression, with 2020 values extrapolated from 1982–2019 trends. Livestock were converted to standard sheep units; downscaling to DPR used CNLUCC areas. Government subsidies were calculated from official reports by practice area. Vegetation data: A 250 m fractional vegetation cover (CD FVC) dataset (1982–2018) was reconstructed from a 250 m NDVI product using an improved pixel bipartite model with NDVImin/NDVImax determined from 5th/95th percentiles across years. Quality issues in 2019–2020 led to restricting analyses to 1982–2018. Climate drivers: TerraClimate monthly data (1/24°) for Tmax, Tmin, precipitation, solar radiation, and wind speed were aggregated to growing-season (Apr–Oct) annual series. Monthly 0.5° CO2 (1982–2018) from CMIP6 inputs were similarly aggregated. Future climate: Outputs from 21 CMIP6 GCMs were used for historical (1980–2014) and SSP-RCP scenarios (2015–2050: SSP1-2.6, SSP2-4.5, SSP5-8.5). Model outputs were bias-corrected by removing differences to observations during 1982–2018. Attribution framework: For grain-for-green areas, a 500 m buffer was used to pair restored land (RL) pixels with adjacent permanent land (pmnt) pixels with similar environment. The contribution of grain-for-green to FVC trend was computed as the relative difference between post-implementation linear trends of RL and pmnt derived from satellite time series. Only permanent farmlands/forests/unused lands were used as pmnt to avoid grazing impacts. For grazing exclusion, only permanent grasslands were analyzed. The period was split into before/after exclusion (province-specific breakpoints). A per-pixel stepwise multiple linear regression (SMLR; precipitation mandatory; other predictors selected by F-test at 95%) was fit between FVC and environmental variables (precipitation, mean temperature, solar radiation, wind speed, CO2) in the pre-exclusion period, then used to simulate the natural post-exclusion FVC trend. The grazing exclusion effect was estimated as the difference between observed and predicted post-exclusion trends, normalized by the total change. Factor contributions to climate-driven change were computed from regression coefficients and factor changes. Model performance was assessed using R^2, MAE, and RMSE; average R^2 was 0.53 with low errors across land-use types. Projections: Using CMIP6 scenarios and the SMLR, FVC trends were projected to 2050, with stippling for significant trends (Mann-Kendall, 95%). Food and income impacts: Grain yield and income were derived from production and price statistics; livestock and meat production were derived from sheep units and assumed ~20 kg meat per sheep unit. Post-2011 grazing metrics excluded areas under grazing exclusion. Grain production decreases were estimated directly from cultivated area reductions; losses in grain and meat production due to practices were computed relative to expected outputs based on baseline farmland (2000) and grassland (2010) areas.

• Vegetation drivers and policy contributions: Despite large-scale interventions since 2000, FVC had been increasing since 1982, indicating substantial natural recovery. After isolating climate and CO2 effects, only 63% of grain-for-green areas contributed positively to FVC; 14% showed negative effects, yielding an average contribution of −1.06% to FVC across the DPR. For grazing exclusion, only 20.45% of implemented grasslands exhibited net FVC increases attributable to exclusion; the average contribution to vegetation restoration was 13.40%. The combined contribution of both practices to FVC increases was 13.07%.
• Natural factors dominate: Warming and wetting trends (precipitation +2.43 mm/decade; temperature +0.37 °C/decade over ~40 years), together with CO2 fertilization, were the principal drivers of FVC increases in grasslands and farmlands. Factor attribution maps show significant contributions from precipitation, temperature, solar radiation, wind speed, and CO2.
• Food production trade-offs: Using baseline areas (farmland in 2000; grassland in 2010), expected 2020 outputs were estimated at 28.2 million tonnes of grain and 106.0 million sheep (~2.1 million tonnes of meat). Due to land-use restrictions from the two practices, average foregone production during 2001–2020 was 13.4% for grain and 24.2% for meat. In 2020, available land (3.2 million ha farmland; 48.3 million ha grassland) yielded 24 million tonnes of grain and 1.6 million tonnes of meat. Based on per-capita requirements (400 kg grain, 21 kg meat), current outputs could support 59.9 million people, below the expected 70.6 million.
• Income impacts and subsidies: The DPR’s 2020 GDP was ~1,092 billion RMB (~153 billion USD), with 8.6% from farming and grazing. About 45.4% of household disposable income for farmers and herders derives directly from these activities. Direct compensation for ecological practices totaled ~5.70 billion RMB (~0.80 billion USD), only ~6.10% of direct income. Direct income of farmers and herders averaged ~75.1 billion RMB/yr (10.5 billion USD/yr), ~15% lower than the expected ~88.4 billion RMB/yr (12.4 billion USD/yr) due to the practices.
• Future projections: CMIP6-based SMLR projections indicate continued warming-wetting to 2050, with expected FVC improvements across 68.12% of farmlands, 65.06% of forests, and 56.29% of grasslands and an overall FVC increase of ~8.17% for the DPR.

The study shows that much of the DPR’s vegetation recovery is explained by climate variability and CO2 fertilization rather than by intervention practices. The marginal or negative additional effect of grain-for-green in many pixels, coupled with modest gains from grazing exclusion, suggests that current programmes are over-applied or poorly targeted. Socioeconomic analyses reveal significant trade-offs: reduced cultivated and grazing areas have diminished grain and meat production and lowered farmers’ and herders’ direct incomes, while subsidies have not offset lost earnings. These findings address the core question by demonstrating that the unintended consequences of the two major programmes include food security risks and increased economic hardship, with limited incremental ecological benefits given strong natural recovery signals. Looking ahead, projected climate trajectories favor further greening, implying opportunities to recalibrate policies. The authors recommend adapting programmes to local conditions: easing blanket land-use restrictions where natural recovery is strong; adopting measures such as farmland shelterbelts, water-saving agricultural practices, and rotational grazing to enhance vegetation while maintaining production; and improving subsidy mechanisms and transparency to support livelihoods, recognizing fiscal constraints. Policy design should target positive socio-ecological synergies—raising rural incomes, maintaining food security, and combating land degradation—rather than prioritizing uniform land-use exclusion.

China’s grain-for-green and grazing exclusion practices in the DPR have yielded limited additional vegetation restoration relative to strong climate- and CO2-driven greening, while imposing substantial costs on agricultural and pastoral production and rural incomes. With future climate likely to further enhance vegetation growth, uniform restrictions are nonurgent in many areas and may be counterproductive. The study contributes a spatially explicit attribution framework combining remote sensing, land-use histories, and climate regressions to quantify practice-specific impacts alongside socioeconomic outcomes. Policy implications include tailoring interventions to local contexts, shifting toward management practices (shelterbelts, water conservation, rotational grazing) that balance ecological goals with productivity, and reforming compensation schemes to better offset income losses. Future research should integrate biodiversity and wildlife movement impacts (e.g., fencing), assess long-term soil and carbon dynamics under alternative management regimes, refine socioeconomic downscaling and price elasticities, and explore optimization frameworks that jointly maximize ecological restoration, food production, and livelihoods under evolving climate scenarios.

Vegetation analyses used CD FVC data through 2018 due to quality issues in 2019–2020. The attribution relies on SMLR with an average R2 of 0.53; unmodeled residual drivers (e.g., nitrogen deposition, natural disturbance recovery, management nuances) were assumed stationary between pre- and post-intervention periods. Programmes launched before 2000 (e.g., Great Green Wall) were not evaluated for ecological–economic trade-offs. Economic data involved county-level imputations and downscaling, introducing uncertainty; assumed constant meat yield per sheep unit (~20 kg) simplifies heterogeneity. CMIP6 projections carry scenario and model uncertainties despite bias correction. Spatial pairing and buffering for grain-for-green and province-level timing for grazing exclusion may misclassify some local interventions.