Determination of Mehlich 3 Extractable Elements with Visible and Near Infrared Spectroscopy in a Mountainous Agricultural Land, the Caucasus Mountains

Soil spectroscopy is a promising alternative to evaluate and monitor soil water quality, particularly in mountainous agricultural lands characterized by intense degradation limited tests reports; few studies have evaluated the feasibility of VIS-NIR predict Mehlich 3 (M3) extractable nutrients. This study aimed (i) examine potential combination with partial least squares regression M3-extractable elements (Ca, K, Mg, P, Fe, Cd, Cu, Mn, Pb, Zn) basic properties (clay, silt, sand, CaCO3, pH, organic carbon-SOC), (ii) find optimal pre-processing techniques, (iii) determine primary prediction mechanisms for spectrally featureless properties. Topsoil samples were collected from representative area (114 525 ha) located region NW Azerbaijan. A series steps transformations applied spectral data, models calibrated based on coefficient determination (R2), root mean square error (RMSE), residual deviation (RPD). The leave-one-out cross-validated predictions showed that first derivative spectra produce higher accuracies (R2 = 0.51–0.91; RPD 1.20–2.29) most evaluation model performance techniques revealed both calibration validation considerable differences values associated sample size random partition or subsets. excellent very good (RPD > 2.0) SOC, Ca, fair (1.4 < clay, poor (1.0 1.4) Zn. Principal component correlation, stepwise analysis, variable importance projection procedures allowed elucidate underlying mechanisms. Unlike previous studies, estimations Cd concentrations linked their correlation CaCO3 rather than matter, whereas Mg P also connected Fe-oxides. particle sizes contributed predicting K concentration but confounded Zn concentration. weaker correlations Cu sizes, data yielded lower accuracy. major relied relations clay content mineralogy, exchangeable cations context association land use. results can be used mountain control effect management quality indices neutrality. Further are needed develop advantageous sampling schemes modeling.