Modelling soil water retention and water?holding capacity with visible–near?infrared spectra and machine learning

We need measurements of soil water retention (SWR) and available capacity (AWC) to assess model functions, but methods are time-consuming expensive. Our aim here was investigate the modelling AWC SWR with visible–near-infrared spectra (vis–NIR) machine-learning method cubist. used soils from 54 locations across Australian agricultural regions, three depths: 0–15 cm, 15–30 cm 30–60 cm. The volumetric content samples their vis–NIR were measured at seven matric potentials ?1 kPa ?1500 kPa. modelled following: (i) directly average different contents, (ii) contents field permanent wilting point calculated those estimates, (iii) air-dried soils, (iv) parameters Kosugi van Genuchten models, then reconstructed curves calculate AWC. compared estimates a local pedotransfer function (PTF) an established PTF. accuracy spectroscopic approaches varied generally better than PTFs. also more practical because they do not require additional properties for modelling. root-mean squared-error (RMSE) ranged 0.033 cm3 cm?3 0.059 cm?3. RMSEs PTFs 0.050 0.077 general Spectroscopy machine learning provides rapid versatile estimating characteristics diverse soils. Highlights Soil can be estimated vis-NIR specta. Parameters models spectra. spectroscopy performed functions. results apply range