A Scalable Algorithm for Light Computation

Models of forest ecosystems are needed to understand how climate and land-use change impact carbon storage and biodiversity. Light is the primary resource that controls the germination, growth and survival of trees in a forest. However, existing models cannot capture the details of forest light without large sacrifices in efficiency, thus limiting their ability to explore large spatial and temporal domains. In this paper we describe a forest light model based on a novel graphics-processor algorithm that reduces runtime by two orders of magnitude over conventional ray tracing algorithms without sacrificing fine-scale detail in understory light conditions. These algorithms allow us to simulate large landscapes for long periods of time without loss of accuracy. We also present experimental results that compares the uncertainty in our light predictions to the (much larger) inherent variability in the ecological processes driven by understory light, seed germination and seedling growth. Finally, we compare our results to previous attempts to speed up conventional ray tracing models, which are based on reducing the range of zenith angles considered, and find that such approaches consistantly overpredict light in forest gaps and edges, precisely the places where 3D ray tracing models are required over simpler 1D models.