Predicting and Mapping of Soil Carbon and Nitrogen Stocks by Diffuse Reflectance Spectroscopy and Magnetic Susceptibility in Western Plateau of São Paulo

With the advancement of agriculture, the use of proximal sensors is one of the most innovative solutions in the search for optimizing time and inputs maximize profitability and reducing environmental impact. In addition, the development of fast, accurate and low-cost methods to quantify soil attributes is of paramount importance to enable detailed mapping. In this context, the aims of this work were to evaluate the prediction capacity of soil C and N stocks by diffuse reflectance spectroscopy (DRS) and magnetic susceptibility (MS) at a sandstone-basaltic transition, and evaluate the spatial variability of this attributes. Soil samples (0.00–0.25 m depth) were collected at 446 sites, air-dried and passed through a 2-mm sieve and analyzed in the laboratory. Were used partial least squares regression (PLSR) for DRS data and linear regression for MS. Good prediction accuracy parameters were obtained between the C stock with DRS (r = 0.87; RMSE = 1.24) and MS (r = 0.88; RMSE = 1.98), as well as the N stock with DRS (r = 0.82; RMSE = 2.10) and MS (r = 0.78; RMSE = 2.41), revealing that these are good predictors of important soils properties. However, the carbon stock maps shows error 28% for DRS and 46% for MS, overestimating C stocks in 5.6 t ha−1 and 9.2 t ha−1, respectively for DRS and MS. The error of the nitrogen stock maps was 6% for DRS, overestimating by 0.2 t ha−1, and 12% for DM, overestimating by 0.4 t ha−1. These errors are in line with current carbon credit certification protocols. However, as a perspective, it will be essential to include predictive improvements in these methods so that they can be fully used in certified soil carbon sequestration offset protocols.