A lightweight dataflow-based software framework for building forest simulators

Abstract

Simulators are critical tools for decision-making in forest management. We propose a dataflow-based model linking approach to increase the flexibility and modularity of forest simulator software while maintaining high computational efficiency. Our approach dynamically constructs a data model and model chains for simulation based on a model library, enabled treatments, and requested output variables. Models are framework- and language-independent pure functions, described through metadata in a domain-specific language. A case study with three model libraries demonstrates the applicability and efficiency of our approach. We observed a 96% speedup compared to an unoptimized real-world model implementation, while 95% of our code (measured by lines) was framework-independent and reusable. We observed a 15% slowdown compared to an optimized hand-written C implementation of a simpler model. We conclude that dataflow-based model linking can be used to build flexible and modular simulation software with a small runtime overhead.