Adaptive multi-scale self-supervised soil pore segmentation based on sequence similarity

One of the main challenges in soil science lies in the passage from heterogeneous soil structure to a quantified multi-scale 3D model. Here a new approach to quantify the microbial distribution relating to soil pore structure is presented. Characterising 3D microbial soil structural in digital porous media is not found and most soil process models tend to assume a homogenous spatial distribution of microbes. We measured the in situ spatial distribution of bacteria in arable soils across scales from sub-micrometers to metres and here we describe further progress to quantify and explicitly model the 3D microbial distributions, based on a stochastic Bayesian approach to predict spatial variation in the underlying bacterial intensity measure. A 3D higher order Multi-Markov chain model is introduced to model complex geometry of real soil structure and associated microbial distribution. In this study, Markov random fields are used to construct multiscale 3D Pore Architecture Models (PAM). The binary structure of PAM has been successfully used to predict multiphase flow behaviour in porous media such as hydrocarbon bearing reservoir rocks, we explore further to use such a new multi-components scheme in modelling pore structure incorporating with microbial spatial distribution, the multicomponent Markov chain model, which is a stationary multiple higher order Markov chain. The models parameterisation is based on high resolution SEM images of soil that have been prepared in a manner that preserves the microbial community information in situ. Based on the quantified 3D multiscale soil structure associated with microbial distribution components, the accurate reactive flow of microbial degradation can be simulated to predict environmental impact of microbial activates in the field. A variety of examples of structures and bacterial distribution created by the models are presented.