A High-Resolution Canopy Radiation and Microclimate Model Based on GPU Technology

Microclimate is important in nearly all biological processes associated with plant canopies (e.g., photosynthesis, evapotranspiration, plant growth, pest and disease spread, and insect and microbial community composition) and an understanding of the interactions across scales is needed to improve our ability to predict various plant and ecosystem processes. Canopy microclimate is largely a function of how solar irradiation interacts with and is absorbed by canopy elements. The energy from intercepted radiation is re-distributed by conduction, convection, and latent cooling. Modeling these processes is difficult, particularly in complex canopies with a high degree of heterogeneity. Radiative energy exchange is computationally intensive, and has traditionally imposed a limitation on the number of canopy elements (e.g., trees) that can be included in a given simulated domain. Convection and latent heat exchange are especially challenging in terms of accurately describing meteorological forcing and biological responses. Because of the large computational expense associated with modeling these processes, current models must make compromises between physical complexity and domain size. There is a need for a modeling tool that can predict microclimate over whole-canopy scales, but also retain all of the important physical processes for individual canopy elements. With this goal in mind, we present a canopy microclimate model that resolves inter- and intra-crown variability, while being efficient enough to include whole-canopy scales. The model solves energy budgets for ground surfaces and sub-volumes of individual tree crowns. Important physics in the radiation model component are included that are commonly neglected in other models such as anisotropic emission and scattering. The associated computational cost was afforded by performing computations in parallel on graphics processing units (GPUs). This included using efficient ray tracing packages optimized for GPU architectures. Efficiency was also improved by using a novel scattering algorithm that reduces the number of required ray traces. Ultimately, this means that domains with tens of thousands of trees with sub-meter resolution can be simulated on a desktop workstation in a few minutes. The model compares well with experimental data obtained from a relatively isolated urban tree and a grape vineyard, which includes direct measurement of leaf surface temperatures, boundary-layer and stomatal conductances, whole-plant water use, ambient turbulence, and the local radiation budget.