Prediction of soil carbon and nitrogen contents using visible and near infrared diffuse reflectance spectroscopy in varying salt-affected soils in Sine Saloum (Senegal)

• Soil C and N contents salinity were accurately predicted using VNIRS (RPD > 2) content was best PLSR by class on log-transformed values. locally weighted Salinity global without log-transformation. Spectrum pretreatment optimization depended the variable type. organic carbon (C) nitrogen (N) have an essential role in soil fertility, but they may be affected salinity, which is especially responsible for land degradation arid semiarid regions. The objective of this work to study ability visible near infrared diffuse reflectance spectroscopy (VNIRS) predict electrical conductivity (EC, a proxy salinity) variably salt-affected topsoils Sine Saloum region (Senegal). Different calibration procedures spectral pretreatments compared, log-transformation usefulness evaluated prediction optimization. Predictions involved three procedures: partial least squares regression (PLSR), used all samples similarly; (local) PLSR, with target individually giving higher weight closest spectra; per class, after discrimination these classes. performed possible spectrum (e.g., derivatization) decimal 311 topsoil (0–25 cm depth), either unsalted slightly salty (Salt-, EC ≤ 2 mS −1 ; 262 samples) or medium highly (Salt+, 49 samples). discriminated spectra: validation, 100% 95% Salt- Salt+ correctly assigned average, respectively. Best predictions achieved (R VAL = 0.87) local 0.77), respectively; 0.90). This suggested salinity; logC logN distributions almost symmetrical, hence usefulness, while logEC distribution very asymmetrical. No yielded systematically good predictions; nevertheless, first-order derivative 31-point gap often predictions, second-order derivatives poor results.