Large-scale ecosystem restoration is one of the few conservation interventions with multi-decade data behind it — China's Loess Plateau is the best-documented case of what sustained, government-funded restoration can achieve, and its limits.
China's Grain for Green Program, launched in 1999 on the Loess Plateau, is among the longest-running large-scale ecological restoration efforts in the world, converting sloped cropland to forest and grassland to combat severe soil erosion.
Over more than two decades of implementation, satellite data show a Normalized Difference Vegetation Index (NDVI) increase across 84.61% of the Loess Plateau, with the restoration program itself estimated to account for 23.8–35.8% of the observed vegetation greening — the remainder attributed to natural environmental factors.
Source: ScienceDirect, 2024; peer-reviewed remote sensing studiesThe program's carbon outcomes are more mixed than its vegetation-cover success: land restored to forest is projected to keep increasing as a carbon sink through 2060, while land restored to grassland showed carbon gains from 2000–2020 that are projected to reverse in coming decades — a reminder that "restoration" is not a single uniform outcome.
The Loess Plateau program predates, but now sits alongside, a formal international restoration target: Target 2 of the 2022 Kunming-Montreal Global Biodiversity Framework, under which 196 countries committed to bringing at least 30% of degraded ecosystems under effective restoration by 2030.
The Loess Plateau is frequently cited as a restoration success story, and the vegetation data supports that framing. But the mixed carbon-sink findings for grassland conversion illustrate a pattern we see across restoration projects generally: headline "vegetation cover" metrics can look strong while more complex outcomes (carbon storage, biodiversity value, hydrological function) tell a more qualified story.
Attributing vegetation change to the restoration program specifically, versus natural climate variability, requires modeling assumptions that introduce real uncertainty — the cited 23.8–35.8% program-attribution range reflects this. Comparable multi-decade satellite monitoring is not available for most other large-scale restoration projects worldwide, making the Loess Plateau an unusually well-documented but potentially unrepresentative case.