A discrete element model for describing coupled hydro-mechanical processes during drying of soils with coated worm burrows

Deep burrowing earthworms produce exudates that coat the biopore wall with compacted finer-textured and organic matter-rich. The coating exhibit high spatial heterogeneity that, although connected to the inter- and intra-aggregate pore network in structured soils, can limit the flow exchange between the macropore and the soil matrix during preferential flow. Such flow exchange can be dynamically quantified if known the complex hydro-mechanical interrelations between biopore structure and soil matrix affecting the stress-strain behaviour at macroscopic scale. Our hypothesis was that the hydro-mechanical interrelations may be described with the discrete element method (DEM) coupled with the pore finite volume (PFV) approach if the model reproduces the pore network between coating and soil aggregates. Therefore, the objective was to develop a coupled DEM-PFV model together with a parameterization procedure based on machine learning algorithm to find the dependency between macroscopic mechanical and hydraulic soil properties obtained from drainage experiments of biopore samples to calibrate micro parameters of the model. The solid phase of the soil matrix was created using DEM inside a cube of about 5 cm edge, randomly filled with two aggregate sizes of 1 mm diameter (constituted by particles of 0.052 mm in diameter) and 0.4 mm diameter (constituted by particles of 0.03 mm in diameter). The pack of aggregates was compressed until the porosity reached the experimental value. The coating surface was created with a thickness of 0.25 mm and particles of 0.015 mm in diameter and compressed to reproduce the experimental porosity. The DEM models were coupled with a two-phase PFV model (2PFV) to simulate hydro mechanical effects during drainage. A total of 500 drainage simulations were performed for matrix and coated sample by randomly varying particle Young's modulus and bond strength. Saturation and strain along with the pressure head were measured to train the machine learning algorithm. The drainage experiments were designed to promote the movement of water from the soil matrix across the coated burrow surface. Thus, the samples were placed in the sandbox with the coated burrow in contact with the sand layer. An optical-laser sensor together with a tensiometer were used to quantify the pressure-head and sample shrinkage while the pressure was reduced at a rate of approximately 50 Pa s^-1. In total, 40 samples of each treatment were used in these measurements. The poly-dispersed DEM-2PFV model was able to reproduce the pore network of coating material and the inter- and intra-aggregate pore network of the matrix that changed dynamically with the increment of pressure head. The machine learning model revealed that the bond strength among particles within aggregates governed the shrinkage of soil matrix, while the particle stiffness of the coating material reduced the susceptibility of aggregate breakage producing a more stable inter-aggregated pore network during the drainage process. This study confirmed that coating material present in biopore surface increases the horizontal soil hydro structural stability. The microscale hydro-mechanic modelling can be useful for finding flow exchange parameters inputs for upscaled models and correlating pore-scale parameters to experimentally determined stress-strain macro parameters.