Visible near infrared reflectance spectroscopy to predict soil phosphorus pools in chernozems of Saskatchewan, Canada

Citation

Abdi, D., Cade-Menun, B.J., Ziadi, N., Tremblay, G.F., Parent, L.É. (2016). Visible near infrared reflectance spectroscopy to predict soil phosphorus pools in chernozems of Saskatchewan, Canada. Geoderma Regional, [online] 7(2), 93-101. http://dx.doi.org/10.1016/j.geodrs.2016.02.004

Plain language summary

Methods to study soil phosphorus (P) pools, including soil organic P, can be complicated, time-consuming and expensive. We tested a rapid, inexpensive method, visible near infrared reflectance spectroscopy (VNIRS), to see if it could be used to accurately predict a number of soil P pools, including total P, organic P and soil test P on soils from Saskatchewan that had been under long-term (31 years) or short-term (8 years) no-tillage management to grow field pea and spring wheat. The VNIRS method could be used successfully to predict some soil P pools such as soil test P and organic P, but was not as accurate for soil total P and some other soil properties.

Abstract

To date, there are few direct or rapid methods to quantify the concentration of many soil phosphorus (P) pools, including organic P (OP). Visible near infrared reflectance spectroscopy (VNIRS) is a rapid, inexpensive, and accurate technique for analyzing a wide variety of organic materials and is increasingly used in soil science. The aim of this study was to examine the potential of VNIRS to predict the concentrations of a number of soil P pools, including OP, total P analyzed by digestion (TP), available P using the Olsen (POls) and Mehlich-3 (PM3) methods, and other related soil chemical properties [organic matter (SOM) and Mehlich-3 extractable Al, Fe, Ca, Mg, and Mn] that could influence the soil P cycle. Soil samples (n = 360) were taken from experimental sites near Indian Head, SK, Canada, from short-term (8 yr) and long-term (31 yr) no-tillage plots of a field pea (Pisum sativum L.)-spring wheat (Triticum aetivum L.) rotation receiving five P fertilizer application rates annually. We randomly selected 80% of each sample set for calibration, while the remaining 20% was used for validation. Prediction models were developed using modified partial least squares regression with cross-validation. The VNIRS models were evaluated using the coefficient of determination of validation (Rv2) and the ratio of standard error of prediction to standard deviation (RPD). The VNIRS predictions for OP were moderately successful for the total soil sample set and for the short-term set (0.80 ≤ Rv2 < 0.90 and 2.25 ≤ RPD < 3.00), and were moderately useful for the long-term set (0.70 ≤ Rv2 < 0.80 and 1.75 ≤ RPD < 2.25). The VNIRS predictions were successful for soil CaM3 and SOM, moderately successful for MgM3, moderately useful for POls, PM3, and AlM3, and less reliable for TP, FeM3, and MnM3. This study demonstrated that VNIRS is a promising analysis technique for soil OP, POls, PM3, CaM3, AlM3, MgM3, and SOM. Success in predictions of OP, POls, and PM3 may be attributed to their relationship to SOM and/or to CaM3 or MgM3.