Improved estimation of clay content from water content for soils rich in smectite and kaolinite.

Standard measurement methods for soil clay content, including the hydrometer and pipette methods, are laborious and difficult to repeat. This has motivated the application of proxy methods such as estimation of clay content from soil water vapor sorption isotherms. Previous studies have revealed that vapor sorption-based clay estimates are inaccurate for soils dominated by swelling or highly weathered non-swelling clays such as kaolinite. In this study, we evaluate an existing vapor sorption-based clay estimation model for soils dominated by either illite (IL), smectite (SM), or kaolinite (KA) clay minerals, and propose modifications to account for high SM and KA contents. Compared to the pipette method, the clay content obtained from the existing water sorption model at 50% relative humidity (RH) was accurate for the IL samples, but significantly overestimated (RMSE = 23.7%; ME = 19.7%) or underestimated (RMSE = 28.8%; ME = −17.0%) clay content for SM and KA rich soils, respectively. The proposed modification involves correcting the estimated clay content with a "slope factor", S for RH values ranging from 30 to 90% for both adsorption and desorption. For SM samples, S averaged 0.76 and 0.72 for adsorption and desorption, respectively, and for KA samples, S ranged from 2.26 to 1.19 and followed a polynomial relationship with RH. Comparison of the estimated clay content from the modified models showed markedly improved estimation of the measured clay content (for RH of 90%, RMSE = 8.0%; ME = 2.6% for SM samples and RMSE = 9.6%; ME = −5.7% for KA samples). • Model for estimating clay content from water content was evaluated for different soil types. • Model accurately estimated clay content for illite-rich samples. • Clay contents for samples rich in kaolinite and smectites were poorly estimated. • Model was improved by correcting for deviation of estimated clay content from measured data. • Improved model reduced average estimation errors by around a factor of two. [ABSTRACT FROM AUTHOR]