ABSTRACT
This chapter aims to predict soil organic carbon content of tropical soils using soil pedon data and satellite imagery with a neural network approach. Soil pedon data represents the most comprehensive and detailed available set of laboratory measurements and descriptive properties of soils as they occur across the landscape. The soil pedon data used to train the neural networks were from the United States Department of Agriculture Natural Resources Conservation Service and the Projecto Radambrasil from areas in western and central Brazil. Neural networks are a unique tool for modeling soil carbon from soil pedon data because of their ability to interpret complex nonlinear relationships in large data sets and either classify or predict information in a usable form. Visual comparison of the soil organic carbon for the tropics in South America with the maps created by predictions of the neural network show similarities in patterns of carbon.
