Soil Quality Analysis Using Modern Statistics and NIR Spectroscopy Procedure

In this paper, we combine the recent development on mathematical statistics with modern chemistry in order to provide a new approach for soil quality analysis. Precisely, from modern chemistry we use near-infrared reflectance (NIR) spectroscopy procedure as a fast, accurate and inexpensive tool to evaluate chemical properties. Based on the collected data, the relationship between soil quality variables is modeled by using the functional statistics, which allows for analyzing a data as a curve or an image. The used predictor models are functional classical regression (FCR), functional local linear regression (FLLR), functional relative error regression (FRER) and functional robust regression (FRR). We prove that the performance of these models is closely linked to the homogeneity of the data. Considering the Abisko soil data, we show that FNR and FLLR are suitable for soil organic matter data, while for the Ergosterol concentration data the use of FRER and FRR are adequate. Furthermore, the proposed functional approach, permits us to avoid many drawbacks of the classical approach as principal component regression (P.C.R.).