This paper presents a localised data assimilation framework for forecasting the evolution of marine oil spills. The framework consists of an advection-diffusion model together with data assimilation and adaptive meshing to improve the accuracy and precision of forecasts, respectively. To provide high parallel scalability, all computation is localised to individual subdomains with the solution being synchronized between direct neighbours at the end of each timestep. No global communication is enforced during computation. The scheme is developed within a novel programming environment aimed at facilitating efficient code development by leveraging advanced 'separation of responsibilities' principles. The front-end API provides the developer with a simple C++ development environment and a suite of parallel constructs that denote tasks to be operated concurrently. Tasks related to the machine and system level are managed by computer scientists at the core-level. We present parallel scalability compared to a benchmark MPI implementation.