Estimating soil chemical properties by diffuse reflectance spectroscopy: Promise versus reality

Abstract Diffuse reflectance spectroscopy in the visible and near‐infrared (VNIR) spectral range has increasingly been utilised in recent years to measure soil properties, including organic carbon and macronutrient content, cation exchange capacity, pH, and even trace metal concentration. Proponents have even suggested that conventional laboratory soil testing methods can ultimately be replaced by these less laborious and more rapid spectroscopic methods. However, to this point in time, reflectance spectroscopy has not demonstrated accuracy comparable to that of conventional physical–chemical testing methods. This review explains why correlations between VNIR spectral features and specific chemical properties within selected sets of soil samples are indirect at best and therefore cannot be trusted more universally to predict soil chemical properties in larger and less homogeneous sample sets. Because mathematical manipulation of the raw spectral data has become increasingly complex and obscure to the users, particularly when machine learning is employed, the possibility exists that the correlations found and the predictions made are based on chance relationships resulting from overfitting the spectral data to include random noise in addition to any real trends. It is concluded that, for real‐world soil testing for agronomic and environmental purposes (e.g. available nutrient and trace metal levels), spectral reflectance methods are not sufficiently reliable to replace conventional testing. Conversely, these reflectance methods show utility and convenience in measuring general trends across landscapes in levels of soil organic matter, carbonates, soluble salts and specific clay minerals. Highlights VNIR reflectance spectroscopy is reviewed as a method to measure soil properties. Estimating most soil chemical properties by VNIR relies on indirect or proxy correlations. Complex regression models used in predicting soil properties may overstate accuracy. Reflectance spectroscopy has lower reliability and extrapolability than conventional soil testing.