ABSTRACT
The development of mathematical or statistical models of relationship between environmental variables and soil organic C (SOC) at observation sites and their application to environmental data sets of entire study areas to generate predictive SOC maps is known as predictive mapping of SOC. This process has progressed dramatically due to computational advances made over the past few decades. For instance, advances in geographic information systems (GIS), remote sensing, and digital terrain modelings have created tremendous improvement over the way SOC maps have been produced. Prediction approaches have varied from statistical approaches of linear regression to more complex machine learning techniques (MLT) such as random forest and articial neural networks. The research in SOC prediction in terrestrial ecosystems is driven by the important role these models will play in better understanding and managing the global C cycle. This chapter demonstrates a geographically weighted regression (GWR) approach to map the SOC pool down to 0.5 m depth for the state of Ohio in the United States as a case study. In this approach, a varying relationship is considered between the environmental variables and the SOC pool over the study area. Topographic attributes, land use map, climate data, normalized difference vegetation index (NDVI), and bedrock geology map were used to predict the SOC pool and predicted SOC was validated using independent samples.
