GPU-CA model for large-scale land-use change simulation

Dan Li,Shaoying Li,Yihan Zhang,Yizhong Zheng,Jigang Qiao,Kai Liu,Chunhua Lao,Xia Li,Xiaoping Liu,Yimin Chen

doi:10.1007/s11434-012-5085-3

Abstract

Land-use change simulation for large-scale regions (i.e. provincial regions or countries) is very useful for many global studies. Such simulation, however, is affected by computational capability of general computers. This paper proposes a method to implement cellular automata (CA) for land use change simulation based on graphics processing units (GPUs). This method can be applied to large-scale land-use change simulations by combining the latest GPU high-performance computing technology and CA. We carried out the experiments by simulating land-use change processes at a provincial scale. This involves a lot of sophisticated techniques, such as model mapping, and computational procedure of GPU-CA model. This proposed model has been validated by land-use change simulation in Guangdong Province, China. The comparison indicates that the GPU-CA model is faster than traditional CA by 30 times. Such improvement is crucial for land-use change simulations in provincial regions and countries. The outputs of the simulation can be further used to provide information to other global change models.

Highlights

Land-use change simulation for large-scale regions is very useful for many global studies
The results indicate that the overall accuracy of graphics processing units (GPUs)-cellular automata (CA) simulation is 82.9%, indicating that the GPU-CA model is very effective and can be applied to large-scale land-use change simulations
In comparison with other parallel computing models, the GPU general-purpose model has advantages of low computational cost, high computing density, easy configuration and installation, computational speedup increasing with the increased computational scale

Summary

GPU-CA computing model framework

The GPU-CA computing model is a CA model for simulating land-use changes combined with the GPU-based general-purpose computing technique. The model uses the high-performance computing capability of the latter technique. It improves the computational efficiency of CA simulation, especially in solving computing bottleneck problems within large-scale and/or high-resolution land-use change simulation. The GPU-CA model framework includes three aspects, namely, mapping the CA model, addressing key technical problems, and designing computing procedures. The GPU-CA land-use change model can use any mature CA model, such as logistic regression-based CA [18], artificial neural-network CA [2], decision tree-based CA [19], and others. We focus on the non-urban to urban land conversion process, we use the logistic regression-based CA model and the data parallel computing pattern

Model mapping

GPU-CA computing procedure

Case application and analysis

GPU-CA model performance analysis

Findings

Conclusions