Use of pedotransfer function to expand soil bulk density data in Rio de Janeiro state, Brazil

Soil bulk density is an essential property for determining soil carbon stock and sequestration, influencing other soil properties and functions. It is a difficult attribute to measure, and therefore, many studies have used statistical and computational methods to model the variation of bulk density as a function of other soil properties, which is known as pedotransfer functions (PTFs). This study aims to: (a) derive a PTF to model soil bulk density in the state of Rio de Janeiro; and (b) apply the adjusted PTF to samples distributed across the state to build a statewide soil bulk density dataset. Soil data from 569 profiles (379 profiles compiled by Embrapa Solos and 190 profiles from the National Forest Inventory) were harmonized to a depth of 0–20 cm. Of these, only 103 profiles contained bulk density data. These samples were randomly split into 79 training and 24 validation subsets. Soil covariates without missing values – total organic carbon (TOC), clay content, pH in H₂O, exchangeable potassium, and sum of bases – were selected for the PTF training and validation. The PTF was adjusted by stepwise linear regression using backward variable elimination. The stepwise model selected only TOC and sum of bases (with TOC being the most significant) and had an R² of 0.69. The root mean square errors (RMSE) were of 0.12 g.cm⁻³ for training and 0.13 g.cm⁻³ for validation. For the 103 profiles, the model produced bulk density predictions (0.46 to 1.47 g.cm⁻³, with a mean of 1.29 g.cm⁻³) consistent with the observed values (0.46 to 1.58 g.cm⁻³, with a mean of 1.29 g.cm⁻³). For the full dataset (569 profiles), overall, bulk density predictions were reasonable; however, the PTF predicted six negative values in organic soils with high TOC content, which were replaced with the lowest positive predicted value of 0.10 g.cm⁻³, a feasible estimate for TOC-rich organic soils. The predicted bulk density of soils in the state of Rio de Janeiro ranges from 0.10 to 1.49 g.cm⁻³, with a mean of 1.27 g.cm⁻³. These predictions will be used for calculating soil carbon stocks at the 569 sites distributed across the state, capturing more adequate soil and landscape variation and ensuring proper spatial coverage for improving soil carbon stock assessment. This study can directly contribute to research related to soil carbon and soil quality, supporting the Sustainable Development Goals 13 (Climate Action) and 15 (Life on Land).