Regression Models Exploration to Measure Available Phosphorus in Soil Based on Middle Infrared Spectroscopy

The frequent monitoring of nutrients available to crops, combined with high sampling density, enhances the efficiency related to the management of fertilizers and other inputs linked to soil management. In this context, reflectance spectroscopy can be used to provide information about the physical and chemical properties of samples at low cost and high speed. However, the application of a reliable methodology to measure phosphorus availability in tropical soils using spectral data remains a significant challenge. This study evaluated the capability of estimating available phosphorus in the soil via spectrometry in the mid-IR range (4000-400 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−1</sup>). Samples have been collected by the Brazilian Soil Spectral Library (BraSpecS), which analyzed soils from different regions of the country. The spectral functions of the library were collected using a commercial FT-IR spectrometer. For data processing, multiple regression algorithms were explored (PLSR, SVR, Gradient Boosting, Random Forest, and Decision Tree). Regarding evaluation metrics, the coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), mean squared error (MSE), and the standard deviation of the error (RPD) were employed. The best result was obtained with the machine learning technique Decision Tree (RPD 2.23, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> 0.80).