Understanding the impacts of land use change and rapid urban expansion on regional carbon sequestration, and clarifying the mechanisms driving spatiotemporal variations in carbon storage, are crucial for ecosystem scientific management and sustainable, high-quality development. This study integrates land use data with the intensity map (IM) model, landscape expansion index (LEI) , integrated valuation of ecosystem services and tradeoffs (InVEST) model, and (XGBoost-SHAP) interpretable machine learning model to reveal the spatiotemporal dynamics and driving mechanisms of carbon storage under rapid urbanization. Results indicate that from 1980 to 2020, built-up land increased by 2,914.02 km(2) while cropland, forest, and grassland decreased by 2,398.52 km(2), 273.96 km(2), and 174.15 km(2), respectively. Systematic conversion patterns emerged: preferential conversion from cropland to built-up land contrasted with suppressed conversions (forest to unused land, water to forest, and built-up land to forest/grassland). Urban expansion peaked during 2000-2010 (1,310.91 km(2) expansion, 13.56% annual rate). Edge-expansion dominated urban growth, increasing from 690.91 km(2) (1980-2000) to 3,392.67 km(2) (2000-2020). The Cropland Protection (CP) scenario exhibits the maximum carbon storage capacity by 2030, demonstrating its efficacy in reversing the declining trend of carbon stocks induced by urban sprawl. Carbon storage in the alluvial/sedimentary sand land in the ancient course of the Yellow River declined by 1.46 x 10(8) t over four decades. XGBoost-SHAP analysis identified fractional vegetation cover (FVC) as the primary determinant of carbon storage, with natural factors' influence diminishing while human activity-related factors became increasingly dominant. These findings provide scientific basis for land use planning and carbon management in the acient course of Yellow River, offering reference for similar rapidly urbanizing regions.