COS 81-9
A highly scalable modeling framework based on GPU technology for simulating radiative transport in complex urban and plant canopies

Wednesday, August 7, 2013: 4:00 PM
M100HC, Minneapolis Convention Center
Matthew C. Overby, University of Minnesota Duluth
Brian Bailey, University of Utah
Rob Stoll, University of Utah
Peter Willemsen, University of Minnesota Duluth
Eric Pardyjak, University of Utah
Background/Question/Methods

         Radiative transport plays a driving role in the exchange of energy and water in urban and forestry ecosystems. Because of its high level of complexity, fully modeling radiative exchange is often cost-prohibitive in terms of computational resources and time. A variety of approaches have been proposed to model radiative exchange and its impact on the energy and water budgets in urban and forestry systems. To date, these have all included some compromises in terms of physical complexity and computational cost. Models that are true to the physics of the problem generally can only include a low number buildings/trees before they become computationally unfeasible. Conversely, models that can include large physical domains generally make some assumptions of homogeneity and/or only use a simplified set of the relevant physical processes. Here we will present a new, physically realistic tree radiative transport model that is linked to spatially explicit energy and water budgets. Additionally, it is efficient enough that city-scale problems can be quickly solved while resolving both trees and buildings. The model includes solar irradiation and longwave radiation exchange between objects within the domain. Trees are treated as participating media and augment radiation through scattering, absorption, and emission. The tree crowns are composed of multiple cell volumes, each with their own physical properties and leaf energy and water budgets. 

Radiative fluxes can be calculated using ray-tracing algorithms, which are traditionally used in computer graphics to render scenes with complex light interactions. Since ray-tracing algorithms are inherently parallel, many computations can happen simultaneously, drastically reducing run time. Our model relies on the highly optimized NVIDIA® OptiX™ ray-tracing engine, which enables parallel computation on commodity class graphical processing units (GPUs). Because GPUs consist of a large array of streaming multiprocessors, complex scenes can now be performed efficiently on an inexpensive desktop computer instead of a supercomputing cluster.

Results/Conclusions

Tree radiation model results were validated using measurements obtained from an isolated tree. The model was able to predict how radiative fluxes on adjacent surfaces were augmented by vegetation, as well as the temperature distribution within the tree crown.  A full-scale simulation of a 5 km2 area of downtown Salt Lake City, Utah was performed which included over 1000 buildings/trees.  Execution times were on the order of a minute, and showed good scalability. Future studies using this model will examine ecosystem services of trees in the urban environment.