BOOK OF ABSTRACTS 147 IHSS&IWA26 / BRNO / CZECHIA / 23–28 August 2026 Thursday, 27 August 2026 / Hall C Soil Organic Matter, Caustobiolites, and Biochar SL72 Evaluating Spectroscopy and Chemometrics as Tools to Improve Agricultural Soil Fertility Testing Gbenga Daniel Adejumo* and Derek Peak Department of Soil Science, University of Saskatchewan, Saskatoon, SK S7N 5A8, Canada Soil researchers continue to explore spectral sensing techniques as a potential solution for testing soil organic carbon (SOC), an important soil fertility indicator. Here, we provide practical guidance for stakeholders on using spectral sensing as a tool for soil fertility testing across diverse Saskatchewan (SK) agricultural soils and beyond. To achieve this, we first reviewed previous and current research studies to evaluate how different modelling approaches influence soil testing, and to identify remaining research gaps. The review highlighted the importance of large datasets in model training. Then, we collected relatively large soil dataset (n = 2205; GPS-referenced) across six SK agricultural regions and five soil zones. Spectral data were acquired alongside laboratory analysis of SOC and total nitrogen (TN), with TN serving as a benchmark for SOC model performance due its established predictive reliability. As a modelling exercise, we explored multiple spectra pre-processing and modelling approaches and found that combining continuous wavelet transform pre-processing with cubist modelling achieved highest performance. Model performance was primarily influenced by the coefficient of variation (Kendall’s tau = 0.69), followed by spectral variability (0.35), and sample size (0.23). Additionally, soils from different agricultural sites and zones responded differently to predictive modelling, with severe limitations in carbonate-rich regions. We further examined strategies to enhance soil fertility modelling in calcareous soils and found that reference methodology affected model performance. Finally, we explored approaches to optimize models across diverse SK soils. We observed that incorporating locally relevant spectral data (20 – 30% of testing data) improved model precision, while smaller investments (approximately 10% of the testing data) still yielded reliable predictions. We concluded that soil sensing holds strong potential for soil fertility testing, provided stakeholders balance the cost of including locally relevant data with the level of accuracy required for specific applications.
RkJQdWJsaXNoZXIy NDA4Mjc=