Optimizing sampling strategies for a regional three-dimensional soil organic carbon mapping in different landforms

• 3D SOC prediction accuracy was compared under different sample densities and landforms by 3448 profiles in Henan, China. • Hill-mountains showed higher SOC prediction accuracy than plains below 20×10-4 km2 sample density. • 20×10-4 and 40×10-4 km-2 were the optimal sample densities for 3D SOC prediction for hill-mountain and plain areas. Despite growing recognition of soil organic carbon (SOC) as a critical role in climate regulation and ecosystem services, a regional three-dimensional (3D) SOC mapping in different landforms remains largely unquantified. This study systematically evaluated the accuracy of 12 soil sample densities (10–200 × 10 −4 km 2 ) with profile depth functions to map 3D SOC distribution across contrasting topographies in Henan Province, China. Using 3,448 legacy soil profiles and 38 environmental covariates, we developed separated Random Forest models (RF) for plain and hill-mountain areas. The experimental results revealed a general decline in prediction accuracy for 3D SOC distribution with increasing soil depth. However, a notable exception was observed in hill-mountain areas, where the 60–100 cm layer demonstrated superior predictive performance (R 2 = 0.08) compared to the 30–60 cm layer (R 2 = 0.04) across all sampling densities. Prediction accuracy of SOC exhibited a positive correlation with sampling density across all landform types. At sampling density of < 20 × 10 −4 km 2 , hill-mountain areas showed significantly higher SOC prediction accuracy compared to plains. However, this relationship reversed at density of > 20 × 10 −4 km 2 , with plains achieving superior accuracy. This cross-over effect persisted consistently across all measured soil depths. The optimal sample densities for effective 3D SOC prediction were identified as 20 × 10 −4 km 2 for hill-mountain areas and 40 × 10 −4 km 2 for plain areas. Spatially, SOC content generally decreased from the western hill-mountain areas to the eastern plain areas. These results demonstrated that the precision of SOC mapping required landform-adapted sampling strategies and emphasized the critical role of pedoclimatic variables in low-relief regions. Our density-optimization framework provides a transferable paradigm for global SOC inventory efforts.