Vis/NIR spectroscopic measurement of selected fertility parameters of Cuban agricultural soils

Abstract Visible (Vis) and Near Infrared Reflectance (NIR) spectroscopy has been recognised as a rapid-response analytical tool to predict soil fertility parameters. In this respect, the conventional methods frequently used in Cuba to determine some fertility parameters important for sugarcane production, such as organic matter (OM), available phosphorus (P) and potassium (K2O), are difficult, costly, and time-consuming for practical use. Also, it has been demonstrated that there is a gap at present between the applied fertiliser rates based on present recommendations and the real requirements of the crop, which implies a significant loss in sugarcane production. Therefore, this study was undertaken to test the accuracy of calibration models of soil fertility parameters obtained in laboratory conditions from air dried samples of different fields of Villa Clara province. The correlation coefficients of P and K2O with the OM were taken into account to achieve this purpose. The parameters P and K2O, which are not spectrally active in the Vis/NIR range are better predicted when are highly correlated with OM. Also, the wavelength intervals to simplify this methodology were selected.The soil samples were collected from Cambisol and Vertisol groups of 10 Agroindustrial complexes from the plough layer (0 20 cm). The samples were split into two datasets, one for calibration on the landscape context and the other for validation on an independent field. For samples used in calibration set a sampling scheme proposed by the Fertilisers and Amendments Recommendations Service (SERFE, in its Spanish acronym) from Cuba was used. The soil samples for the independent validation sets were selected from two municipalities from Villa Clara province (Santa Clara and Sagüa la Grande, for Cambisol and Vertisol respectively). The reflectance spectra were acquired in laboratory conditions for all the soil samples by using a portable Vis/NIR spectrophotometer in reflectance mode in the wavelength range of 399-1697 nm. The regression models were built in Matlab 7.9 (Mathworks, 2009) by means of Partial Least Squares (PLS) regression, Locally Weighted Regression (LWR) and Support Vector Machine (SVM). These regression methods enabled the relating of near infrared reflectance spectra to measured values of OM, K2O and P in the soil. The pre-processing method included Log (1/R), Smoothing and Mean Centre. This method was used while developing the regression models for handling the possible interferences which do not carry chemical information. The PLS, LWR and SVM analyses were evaluated in Venetian blinds cross validation to optimize the model complexity for reliable prediction of these soil fertility parameters.Results indicate a significant spatial variability of all essential soil fertility parameters at landscape level and at field scale; however there was less variation in the OM content than in the P and K2O content. Also, all the regression models(PLS, LWR and SVM) provided good correlations between soil spectra and OM. The best accuracy corresponded to the nonlinear regression models for Cambisol and Vertisol at landscape level and within a field scale. For the prediction of the average soil fertility parameters at landscape level on Cambisol soil the best results were obtained for OM (0.90≤R2≤0.93; 0.12≤RMSEP≤0.14), followed by K2O (0.77≤R2≤0.79; 3.47≤RMSEP≤3.62), Olsen P (0.69≤R2≤0.81; 0.27≤RMSEP≤0.35) and Oniani P (0.64≤R2≤0.65; 3.31≤RMSEP≤3.54). The prediction accuracy at landscape level on Vertisol soil were similar for OM (0.81≤R2≤0.87; 0.16≤RMSEP≤0.22) and for K2O (0.83≤R2≤0.87; 2.09≤RMSEP≤2.40). However, these results were better than for Olsen P (0.76≤R2≤0.80; 0.55≤RMSEP≤0.67) and Oniani P (0.74≤R2≤0.81; 2.86≤RMSEP≤3.10). The results for the prediction of variation in soil fertility parameters within a field on Cambisol soil showed successful correlations between soil spectra and OM (R2=0.92; RMSEP=0.14). Also, in this type of soil the non-linear regression models gave the best results for K2O (0.61≤R2≤0.63; 5.13≤RMSEP≤5.88), Olsen P (0.68≤R2≤0.83; 0.27≤RMSEP≤0.34) and Oniani P (0.70≤R2≤0.72; 2.32≤RMSEP≤2.52). Within a field on Vertisol soil the results were lower than those obtained in Cambisol. The results obtained for OM (0.79≤R2≤0.80; 0.21≤RMSEP≤0.24) were higher than K2O (0.60≤R2≤0.61; 3.03≤RMSEP≤3.37), Olsen P (0.51≤R2≤0.58; 0.59≤RMSEP≤0.78) and Oniani P (0.56≤R2≤0.58; 2.91≤RMSEP≤4.23). These results promoted the basic knowledge for applying a strategy of precision fertilisation in Villa Clara province.status: publishe