This paper has presented three different applications of DEVS for the modelling of environmental systems. The choice has been made not to focus on actual implementation but on the overall presentation of the models, showing the benefits added by the DEVS formalism to the different paradigms used in these simulations. Depending on what is to be observed, the actual model can be very different, concentrating on a cell, a particle or a front, but nevertheless all are thought as DEVS automaton, with inputs, outputs, update and a time advance function that handles the asynchronous nature of the systems and the time granularity naturally. In the first application, a cellular Mediterranean fruit fly model was developed to estimate concentration of insects in mature fruits. Applying DEVS on this problem made the modelling task easier as it helped defining the asynchronous nature of the problem with a well defined structure. In the second application, DEVS is used to define models of algae advected by a prescribed flow. Defining the model as DEVS helped in this case to directly cope with the large diversity of current speed in the sea, as DEVS models position is not integrated over time, but for along specific events, here defined as the time to travel a given distance: the quantum distance. Moreover, the algae model can be run in inverse simulation just by using a negative quantum distance, therefore enabling the model to answer other questions. The last application extended the particle model by linking particles together to form a front. The method helped to directly focus the computation on the very front line of the fire, providing simulation times under a minute for a large fire that run for a whole day with a resolution of ten meters. To illustrate that it is possible to couple DEVS models with discrete time simulation, the fire spread simulator is later coupled with a meso-scale atmospheric model, enabling to simulate fire plume, and fire induced winds. In all these simulation, the use of DEVS or DEVS variants proved to offer a well defined framework that removed the constraints on the way models are thought and offers the frame on which models are built.