The use of Pedotransfer functions and the estimation of carbon stock in the Central

Computer models have been used to assess soil organic carbon (SOC) stock change. Commonly, models require to determine soil bulk density (Db), a variable that is often lacking in soil data bases. To partly overcome this problem, pedotransfer functions (PTFs) are developed to estimate Db from other easily available soil properties. However, only a few studies have determined the accuracy of these functions and quantified their effects on the final quality of the spatial variability maps. In this context, the objectives of this study were: i) to develop one PTF to estimate Db in soils of the Brazilian Central Amazon region; ii) to compare the performance of PTFs generated with three other models generally used to estimate Db in soils of the Amazon region; and iii) to quantify the effect of applying these PTFs on the spatial variability maps of SOC stock. Using data from 96 soil profiles in the Urucu river basin in Brazil, a multiple linear regression model was generated to estimate Db using SOC, pH, sum of basic cations, aluminum (Al+3), and clay content. This model outperformed the three other PTFs published in the literature. The average estimation error of SOC stock using our model was 0.03 Mg C ha−1, which is markedly lower than the other PTFs (1.06 and 1.23 Mg C ha−1, or 15 % and 17 %, respectively). Thus, the application of a non-validated PTF to estimate Db can introduce an error that is large enough to skew the significant difference in soil carbon stock change.