A software framework for process flow execution of stochastic multi-scale integrated models

Abstract

Dynamic environmental models use a state transition function, external inputs and parameters to simulate the change of real-world processes over time. Modellers specify the state transition function and the external inputs required in the process calculation of each time step in a component model, a self-contained numerical module representing an individual spatio-temporal process. Depending on the application case of a component model - such as standalone execution or in an integrated model - the source of the external input needs to be specified. The required external inputs can thereby be obtained by a file operation in case of a standalone execution. Alternatively, required inputs can be obtained from other component models, in case the component model is part of an integrated model. Using different notations to specify these input requirements, however, requires a modification of the state transition function per application case and therefore would reduce the generic applicability of a component model. To address this problem, we propose the function object notation as a means to specify the input requirements of a component model. This function object notation provides modellers with a uniform syntax to express the input requirements within the state transition function. During component initialisation, the function objects can be parameterised with different external sources. In addition to a uniform syntax, the function object notation allows a modeller to specify a request-reply execution flow of the coupled models (i.e. a component requests data needed for its own progress from another component). We extend the request-reply execution approach to Monte Carlo simulations and implement a software framework prototype. Using this prototype, we build an exemplary integrated model by coupling components for land use change, hydrology and Eucalyptus tree growth at different temporal discretisations to obtain the probability for bioenergy plant growing in a hypothetical catchment. The presented approach allows modellers to specify input requirements in the state transition function independently from the source of external inputs and therefore increases the reusability of these component models.