[en] Soil water retention curves (SWRCs) are key inputs to feed Richards’s equation-based hydrological models. Knowing that these models play a role in a wide range of societal issues, they must be based on reliable data. SWRCs are usually obtained in laboratory on soil samples using one/some of the available methods. Although some studies show that different non-harmonized elements of the procedures for the determination of SWRCs in laboratories can significantly influence the measurement of retention properties, to date, these procedures are not harmonized. The impact of these non-harmonized procedures on the legacy SWRCs data and on the hydrological models they feed remains to be investigated.
The challenge was to carry out an interlaboratory comparison using an artificial constructed porous reference sample set with controlled retention properties that can be transferred safely between laboratories. The reference sample was composed by a mix of glass beads and cement. The inter-laboratory comparison involved 14 European laboratories with 3 successive rounds of measurements of four retention points (10, 50, 100 & 300 hPa) on 84 reference samples. The samples followed specific inter-laboratory exchange schemes designed to assess both the intra-/inter-laboratory variability and the effect of sample transfer. The random effect related to the laboratories, samples and transport between laboratories on the SWRCs were determined based on a Bayesian linear mixed model programmed in the “Stan” language.
A simple bulk density analysis showed that the reference samples were not uniform, with bulk densities ranging from 1.575 to 1.835 g/cm³. Nevertheless, the linear mixed model shows that the variance explained by the differences between laboratories is more important than the variance explained by the intrinsic differences between samples. This underlines the fact that differences in SWRC measurements, on a same sample, from one laboratory to another can be substantial. However, the dry mass of the samples increased significantly between the first and last series of measurements, despite the fact that some material losses were reported. Although not considerable, the transfer of samples between laboratories seems to significantly influence the retention curves. This indicates that the retention characteristic of the reference samples could change over time. In this case, the methodology used to analyze sources of variability can be biased and could lead to inaccuracies in the estimation of the variability attributed to the laboratories or samples.
These results shows us that the uncertainty associated with the determination of the retention curve in the laboratory can be substantial and should be a concern. However, a more appropriate porous reference sample is needed to refine the investigation and gain insight into the underlying causes of this uncertainty.
