大规模森林多精度生长仿真模型及其计算加速算法

Multi-resolution Growth Model for Large-scale Forests and Its Computational Acceleration

既能高效快速模拟大规模森林的生长,又能根据资源量来精确模拟局部树木的生长,是森林生长仿真计算中的重要问题,但以往的方法很难同时达到这两个要求.提出一种适用于GPU并行加速计算的,同时兼顾大规模森林场景仿真和局部样地树木生长仿真的计算模型.在基于SORTIE模型快速计算大规模森林场景的生长演化情况的基础上,引入了改进的植物影响圈模型(field of neighborhood,FON)模型对森林局部区域进行精细仿真,从而实现一种大规模森林的多精度仿真模型.本文进一步利用了通用图形处理器(graphic processing unit,GPU)并行计算特性设计了上述模型适用于统一计算架构(compute unified device architecture,CUDA)的计算加速算法.应用上述模型和算法,本文针对斜坡样地上不同树种的生长进行生长仿真模拟.计算结果表明模型和加速算法是非常有效的.

How to achieve the efficient simulation of the changes of a large-scale forest, as well as the fine-scale simulation of individual tree growth based on the allocation of resources is an important issue in the computation of forest dynamics. It is difficult to achieve both requirements simultaneously by previous methods. In this paper,we propose the forest multi-resolution growth model which combines the refined tree growth model in local plots and the simulation of a global forest scenario. The evolution of the forest is efficiently simulated based on the SORTIE model on GPU for the global scale. For the local plot of interest, we adopt the improved FON model to compute the growth of trees more precisely by including the influences of resources. To further accelerate the computation of the tree growth of lo- cal plots, we implement the improved FON growth model on a Graphics Processing Unit { GPU } by exploiting the parallel computation in the model. The acceleration is totally implemented with CUDA, a parallel computing platform and programming model invented by NVIDIA. To verify the effectiveness of our method, we perform the experiment on trees of different species on a ramp-like plot. The resuits show that our model and acceleration algorithm achieve good performance in both accuracy and efficiency.