Laboratory analysis of soil's chemical, physical, and biological properties has been costly and time-consuming. These methods require extensive sample preparation and produce toxic byproducts. Globally, scientists want faster, cheaper soil analysis methods. Soil spectroscopy is gaining popularity because it is fast, nondestructive, and environmentally friendly. This study developed chemometric models to assess important soil properties pertaining to soil health. The developed models were subsequently employed to quantitatively predict these properties for soils in Michigan. Prediction models were developed using partial least square regression and random forest from two individual libraries and a combination of the two libraries). . The samples were scanned in the laboratory in the mid-infrared (MIR) range (6000-400cm-1), and preprocessed with first derivative to improve predictions The evaluation of these models was conducting using an independent test set obtained from each library. Additionally, we investigated the factors driving the model performance Soil properties measured were: total carbon (TC), soil organic matter (OM), pH, base cations (calcium [Ca2+], magnesium [Mg2+], potassium [K+]), cation exchange capacity (CEC), total nitrogen (TN), and extractable phosphorus (P-).. OM predicted the highest (R2=0.93, RMSE = 2.14 %) and P and K were the lowest (R2 = 0.26 and R2 = 0.17), Additionally, we deployed the models to predict soil properties that have not yet been measured on a long-term ecological research site (LTER). The impact of these findings demonstrates the potential of soil spectroscopy to enhance global soil carbon monitoring and expand our understanding of soil health by establishing relationship with soil properties.
Read
