Proximal Soil Sensing Applications in Soil Fertility

Abstract

Soil samples were collected from two locations: Samawa and Rumetha from Al-Muthana governorate in the south of Iraq. The samples were divided into two sets to be used as a calibration and validation set. Vis-NIR reflectance (350–2500 nm) and GIS-Kriging were used in combination with Partial Least Square (PLS) to predict soil available Phosphorus and total N. For total N, only two regions reported higher determination coefficient R2 and lower Root Mean Square Error (RMSE) than the other wavelength regions. PLS calibration models yielded an R2 of 0.96 and 0.97 for Rumetha and 0.87 and 0.94 for Samawa location in bands at 500–600 and 800–1000 nm, respectively. The validation dataset from both study locations were used to predict the new unknown soil samples based on NIRS and GIS-Kriging models. The GIS-Kriging models were unfavorable predicted with an Q2 of 0.28 between laboratory-measured and predicted total N values for Rumetha and 0.43 for Samawa location. While NIRS- based validation models achieved highly predictive power with an R2v of 0.84 between laboratory-measured and predicted total N values for Rumetha and 0.85 for Samawa location. These results reveal an extreme decrease in model predictive ability when shifting from NIR Spectroscopy method to GIS-Kriging. According to the results of this study, three wavelength regions were reported as the main sensitive bands for soil available P. The best prediction ability was achieved for Rumetha location at 1400–1600 nm with an R2 of 0.85, lowest RMSE of 1.405, and lowest standard deviation of 1.577 and for Samawa location at 900–1000 nm with an R2 of 0.81, RMSE of 2.666 and lowest standard deviation of 2.879. At wavelength region 2100–2200 nm, both studied locations showed lowest R2, highest RMSE, and highest standard deviation. The capability of the NIRS- based and GIS-Kriging prediction models were evaluated by using cross-validation values Q2 and R2 between measured and predicted soil available P of each model. The selection principle parameters showed the best prediction by NIRS-models with an R2 of 0.79 for Rumetha and 0.75 for Samawa location. While the prediction ability of GIS-Kriging models were in worst with an Q2 of 0.17 for Samawa location and reasonable with an Q2 of 0.58 for Rumetha location. This empirical result is an evidence of the superiority of NIRS-based models for prediction soil available P over the GIS-Kriging models. To simulate soil fertilization conditions, 15 soil samples from each location have mixed with Urea fertilizer in the rates ranged from 0.1 to 0.8 g kg−1 soil (90–720 kg N ha−1) and Monopotassium Phosphate (MPP) fertilizer in the rates ranged from 160 to 700 kg P2O5 ha−1 (70–300 kg P ha−1). Based upon values of coefficients between lab-measured and absorbance spectra R2, four wavelength regions were highly correlated with total N. Among the four selected sensitive wavelengths for total N, the highest Q2 were observed at 2100–2200 nm with a Q2 of 0.97 for Rumetha and 0.95 for Samawa location. Three main bands for prediction soil available P are described in this study. The capability of NIRS-based models was best at 900–1000 nm with an Q2 of 0.75 for Samawa and 0.91 for the Rumetha location. The yielded quality parameter values were at best successfully models to predict total N and soil available P and well suited for a large variety of low to high concentrations. These results indicate that, NIR Spectroscopy is an effective tool for a rapid assessment of soil information under field conditions, through which decision on fertilizer requirements can be based (under the conditions of this study).