Abstract
There is a serious lack of detailed and accurate three‐dimensional soil distribution information worldwide. This study examined the effectiveness of combining radial basis function (RBF) neural networks and profile depth functions to map the three‐dimensional distribution of soil organic matter (SOM) in a subtropical hilly landscape in southern Anhui Province, China. The RBF networks were used to predict the lateral distribution of SOM based on its relations with terrain attributes and land uses, while the depth functions were used to fit its vertical distribution based on sparse measurements of SOM in soil genetic horizons. Compared with power and logarithmic functions, the equal‐area quadratic splines had smaller bias, higher accuracy, and more stable performance in fitting the vertical SOM distribution. The prediction accuracy of the whole three‐dimensional mapping method decreased with depth within the upper 60 cm, while the best accuracy occurred below 60 cm. In the upper 30 cm, areas with high elevation tended to have high predicted SOM content and vice versa. There were local deviations from this pattern in areas where toeslopes and ravines had higher predicted SOM content than backslopes, even though the latter are at higher elevations. Multiple regressions with dummy variables showed that the influence of terrain conditions on SOM content was strong in the upper 60 cm and weak below 60 cm, while that of land use was strong in the upper 30 cm and weak below 30 cm. Both influences were the strongest in the upper 15‐cm soil layer. Under the same terrain conditions, agricultural cultivation is associated with SOM accumulation in the upper 30 cm.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have