Abstract Soils are important components of ecosystem and their functioning is of great importance for human well-being. Describing, understanding, foreseeing, and controlling biological activities and functions in soil in the context of global change are major objectives for scientists. Modelling soil bioturbation by soil engineers is of great importance although it is faced with the difficulty to describe the soil physical environment. Creating a model of a soil leads to complexity problems: a soil is a multi-scale heterogeneous, three-dimensional and dynamic environment that cannot be modelled and simulated without defining a suitable and optimized representation and taking assumptions about the studied system. An approach based on fractal theory (often used in soil sciences) was chosen to model such a real complex environment; it was integrated into a multi-agent system (MAS). MAS allows to simulate situated agents (earthworms) in an virtual world. The originality of this present MAS is that it is based on a dynamic environment which builds itself, on demand, according to an abstract canvas tree and agent movements. The aim of this paper is to present this approach and its originality, and to describe the model and the simulator. A theoretical view of the approach is given and applied to a case study: the impact of earthworms on soil structure and organic matter dynamics.