Digital Soil Mapping of Topsoil Organic Carbon Content of Rio de Janeiro State, Brazil

Abstract

A database with 431 soil profiles of Rio de Janeiro State was used in a research project entitled Quantifying the magnitude, spatial distribution and organic carbon in soils of Rio de Janeiro State, using quantitative modeling, GIS and database technologies (Projeto CarbonoRJ, sponsored by FAPERJ Carlos Chagas Filho Foundation for Research Support in Rio de Janeiro State). These soil data were collected for other purposes and there were only limited soil bulk density data (103), which is needed for estimating soil organic carbon (SOC) stocks. Pedotransfer functions (PTFs) were estimated to be used in the modeling of organic soil carbon of topsoil (010cm), using the scorpan model. The following environmental correlates were used as predictor variables: satellite data (Landsat ETM+), lithology and soil maps, and a DEM and its derivatives. This dataset represents the best organized soil dataset in Brazil and is working as an educational trial for Digital Soil Mapping using a variety of methods for predicting soil classes and their properties. Multilinear analysis and regression-kriging were used to perform the modeling. Seven different models were built and compared through statistical methods. The main difference between the models was the set of predictor variables used to perform them. In general, all models performed well to predict the SOC stock. Nevertheless, model 6 was considered the best one since it presented the smallest AIC and RMSE as it used existing soil information (polygon soil maps) as a predictor variable, in addition to the variables used in the other models. The results obtained with this model were used to map topsoil carbon stock at a spatial resolution of 90 m.