Accurate estimation of carbon storage in terrestrial ecosystems and the scientific formulation of eco-environmental protection and land use policies are crucial for promoting regional low-carbon sustainable development and achieving the goal of carbon neutrality. Based on a large number of carbon density sample points, this study overlayed it with an ecological geographic zoning map and land use map, and used the Kriging interpolation method to obtain the spatial distribution dataset of carbon density in the Yellow River Basin (YRB). Then, the InVEST model was employed to measure the spatio-temporal evolution of terrestrial ecosystem carbon storage of YRB in 2000, 2010, and 2020, improving the accuracy of carbon storage estimation results. Pearson correlation analysis and multiscale geographically weighted regression model (MGWR) were used to analyze the impact of natural factors, socio-economic factors, and landscape pattern indexes on carbon storage per unit area at the level of county administrative units. The main findings are as follows: (1) The overall carbon density showed a spatial pattern of the west being higher than the east, with a decreasing trend from southeast to northwest in the east. (2) From 2000 to 2020, the terrestrial ecosystem carbon storage increased by 0.02% (7.011×10⁹-7.012×10⁹ t) in the YRB, and the spatial pattern of carbon storage was the same as that of carbon density, with significant spatial agglomeration characteristics. The "high and high aggregation" areas were mainly distributed in the Qinghai-Tibet Plateau region in the southwest part of the upper YRB, while the "low and low aggregation" areas were mainly distributed in the northern part of the upper YRB and most of the lower YRB. (3) Pearson correlation analysis showed that mean annual precipitation (Pr), normalized difference vegetation index (NDVI), and slope were positively correlated with carbon storage; mean annual temperature (TEM), Human Active Index (HAI), Shannon Diversity Index (SHDI), digital number value of nighttime light data (DN) and population density (PPOD) were negatively correlated with carbon storage. (4) The MGWR model showed that TEM, Pr, NDVI, and SHDI had strongly spatial heterogeneity in 2000, 2010, and 2020, and HAI showed strongly spatial heterogeneity after 2010. The slope had moderate spatial heterogeneity. DN and PPOD were global scale variables, and their spatial effects were stable. (5) The MGWR model showed that NDVI had the strongest effect on carbon storage per unit area at the level of county administrative units in the YRB. NDVI and slope had positive effects on carbon storage per unit area at the level of county administrative units; TEM, HAI, DN, and PPOD had negative effects; Pr and SHDI had both positive and negative effects.