A scalable plant-resolving radiative transfer model based on optimized GPU ray tracing

A new model for radiative transfer in participating media and its application to complex plant canopies is presented. The goal was to be able to efficiently solve complex canopy-scale radiative transfer problems while also representing sub-plant heterogeneity. In the model, individual leaf surfaces are not resolved, but rather vegetation is aggregated into isothermal volumes. Using the leaf angle distribution and leaf area density functions, the volumes realistically augment the radiation field through absorption and anisotropic scattering and re-emission. The volumes are grouped to form individual plants, and individual plants are grouped to form entire canopies. The model increases efficiency by performing ray tracing calculations on graphics processing units (GPUs) using the NVIDIA^® OptiX™ and CUDA™ frameworks, and through efficient algorithms for radiation reflection, scattering, and emission. This efficiency allows for realistic representation of heterogeneity, while also allowing for the solution of problems with very large domains (~10⁵ trees) quickly on an inexpensive desktop workstation. Problem execution time scaled nearly linearly with the number of discrete elements in the domain. Model results are compared with experimental data collected from an array of radiation sensors within and above a grapevine canopy and an isolated tree. Agreement between simulated and measured values of shortwave and longwave radiation were very good, with model predictions generally within the expected measurement accuracy.