Near Infrared Reflectance Spectroscopy Prediction of Soil Nitrogen Supply in Humid Temperate Regions of Canada

Abstract

Near infrared reflectance spectroscopy (NIRS) holds promise for rapid assessment of soil total N and organic C contents, but its ability to predict soil N supply in soils with contrasting physio-chemical properties needs to be evaluated. The objectives of this study were to: (i) evaluate NIRS predictions of measured soil parameters (total N, organic C, and C/N ratio) and as an indirect indicator of soil N supply, namely corn (Zea mays L.) N uptake and (ii) assess the effect of sample set heterogeneity on NIRS predictions. Soil samples (n = 282 in the total set) were collected between 2000 and 2009 from 52 sites across four Canadian provinces and were allocated to fine-(>350 g clay kg-\ n = 101) and coarse-textured subsets (<350 g clay kg~n = 181). Prediction models were developed using modified partial least squares regression. Prediction precision was assessed using the coefficient of determination {R^) and ratio of prediction to deviation (RPD = ratio of standard error of prediction corrected for bias to standard deviation of the reference data used in the validation). For the total set, predictions were reliable for total N, organic C and C/N ratio (0.7 < R^ < 0.9, 1.75 < RPD < 3), and less reliable for soil N supply {R^ < 0.7, RPD < 1.75). Prediction precision for total N, organic C, C/N ratio, and soil N supply increased with set homogeneity (i.e., better in the texture subsets than the total set), resulting in reliable predictions for soil N supply (RPD > 2.00) in the texture subsets. This study demonstrated the possibility of developing reliable NIRS predictive models for total N, organic C, and C/N ratio for soils with contrasting physiochemical properties, as well as reliable NIRS predictive models for soil N supply within homogeneous texture sets.Abbreviations: CV, coefficient of variation; Math treat, math treatment; n^ number of samples in calibration; n^, number of samples invalidation; NIRS, near infrared reflectance spectroscopy; organic C, organic carbon; R^, coefficient of determination; RPD, ratio of prediction to deviation; SD, standard deviation; SEC, standard error of calibration; SECV, standard error of cross-validation; SEP, standard error of prediction; SEP(C), standard error of prediction corrected for bias; total N, total nitrogen; SNVD, standard normal variate and detrending; T-Outliers, outliers discarded during calibration.

Infrared spectroscopy prediction of organic carbon and total nitrogen in soil and particulate organic matter from diverse Canadian agricultural regions

“…Previous studies at field scale in Canada reported excellent estimation by NIR-PLSr for SOC (R 2 = 0.86-0.98; RPD = 2.9-3.6) and TN (R 2 = 0.83-0.93; RPD = 3.4-4.8) (Nduwamungu et al 2009), or reliable estimation for SOC (R 2 = 0.875; RPD = 2.16) and poor estimation for soil organic N (R 2 = 0.326; RPD = 1.22) (Martin et al 2002). Reliable estimation by NIR-PLSr for SOC (R 2 = 0.86; RPD = 2.65) and TN (R 2 = 0.85; RPD = 2.56) was found for soils from the eastern Canada (St. Luce et al 2012), and excellent estimation by VNIR-PLSr for SOC (R 2 = 0.92; RPD = 3.70) and TN (R 2 = 0.93; RPD = 3.83) were found for the soils from the western Canada (St. Luce et al 2014). A previous study carried out at the same site (Woodslee), including the 50 samples used in this study and 245 more samples from other field trials at the same experimental farm (CV for SOC = 60%; CV for TN = 40%; ranging 3.4-36.5 g C kg −1 , 0.6-2.8 g N kg −1 ), provided excellent predictions for SOC and TN (R 2 = 0.96-0.99 for MIR and R 2 = 0.95-0.97 for NIR, the RPD values > 4) (Xie et al 2011).…”

“…Luce et al 2014) at different scales (local, regional, national, or global). In Canada, MIR and NIR spectroscopies have been successfully used to estimate C and N in bulk soils (Martin et al 2002;Nduwamungu et al 2009;Xie et al 2011;St. Luce et al 2012;St.…”

Predicting Soil Phosphorus‐Related Properties Using Near‐Infrared Reflectance Spectroscopy

“…The modified partial least squares regression method (PLSR) of the WinlSI III software was used to develop calibration equations for the soil and crop properties. To maximize the probability of developing a robust calibration equation for each property, a maximal number of soil samples, corresponding to 80% of each soil sample set, were randomly selected by the software to be used for the calibration set, and the remainder samples were used for the validation set (Ludwig et al, 2002;Brunet et al, 2007;St. Luce et al, 2012).…”

“…Predicting crop P-uptake and P-budget from soil spectra would eliminate the need to establish relationships between soil test P and crop response to P fertilization. Börjesson et al (1999), Terhoeven-Urselmans et al (2008), and more recently St. Luce et al (2012) link NIRS soil spectra to winter cereal N-uptake and report good predictions (i?^^ > 0.70), but to our knowledge, the prediction of crop P-uptake and annual P-budget by NIRS has not been documented.…”

Evaluation of Two Portable Hyperspectral-Sensor-Based Instruments to Predict Key Soil Properties in Canadian Soils

“…A prototype mid-IR spectrophotometer is a further addition to the ISE which will allow for the integration of sensors based on different measurement principles [ 44 , 45 ]. However, applicability of this instrument to predict SOC and texture in comparison with the earlier mentioned P4000 unit needs to be assessed on many diverse soil samples [ 46 ].…”