Reconstruction of Forest and Grassland Cover Changes in the Hehuang Valley, Northeastern Margin of the Qinghai–Xizang Plateau, over the Past Two Millennia

Anthropogenic land use constitutes a key driver of land cover change, exerting profound impacts on terrestrial carbon/water cycles and ecosystems across global to regional scales. Consequently, long-term land cover datasets serve as fundamental data infrastructure for ecological effect assessment and paleoclimate modeling. Considering the dominant role of historical anthropogenic disturbances in vegetation change across the Hehuang Valley, we developed a 1 km × 1 km potential vegetation prior to land reclamation map through integrating remotely sensed land-use patterns with Random Forest-modeled vegetation predictions. Then, we derived changes in forest and grassland areas and their spatial patterns over the past two millennia by subtracting the spatially explicit cropland cover for six agricultural expansion stages. Finally, we compared our results with representative historical LUCC datasets. The main conclusions are as follows: (1) The potential vegetation of the Hehuang Valley was predominantly grassland, with forest occurring as a patchy and linear mosaic. Grassland and forest covered approximately 2.6 × 104 km2 and 0.7 × 104 km2, respectively. (2) Over the past two millennia, during the Han (202 BCE–220 CE), Tang (618–907 CE), Song (960–1279 CE), Ming (1368–1644 CE), and Qing (1644–1912 CE) dynasties, as well as the Republic of China period (1912–1949 CE), cropland reclamation reduced forest and grassland cover by 6% and 18%, respectively. Forest area decreased by 23.58–100.03 km2, while grassland area decreased by 310.72–1513.05 km2 across these periods. Grassland reduction was concentrated in valleys, whereas forest reduction was spatially scattered but locally intensive. These findings highlight the need to promote sustainable cropland development through technological innovation, improved management, and agricultural specialization rather than ecosystem conversion. (3) Compared with the HYDE 3.2 dataset and a national-scale reconstruction dataset, our framework better reflects the regional characteristics of the Hehuang Valley and provides a more detailed reconstruction of long-term forest and grassland changes.