基于MapReduce的空间敏感性分析并行算法设计

近年来，随着遥感空间数据广泛应用于生态系统，推动了区域尺度生态遥感参数模型的发展。敏感性分析对识别模型关键参数，降低模型不确定性和完善模型具有重要作用。区域尺度的生态遥感参数模型，在进行模型参数敏感性分析时，由于涉及到空间数据的复杂运算，单机环境无法满足快速分析的要求。为了提高生态遥感参数模型空间敏感性分析效率，本文以青藏高原为研究区域，利用植被光合模型VPM（Vegetation Photosynthesis Model）和开源云计算平台Hadoop，设计和实现了基于Sobol′的生态遥感参数模型空间敏感性分析并行算法，并在实验室集群环境下进行算法分析，验证了算法的有效性和适用性。该算法的核心是利用MapReduce并行编程技术，对空间敏感性分析中的地图抽样和模型迭代过程进行任务分割，将分割后的子任务分配至不同的计算节点进行并行计算。实验表明，本文提出的并行策略，能有效缩短地图抽样和模型迭代计算时间，相比于单机算法，并行算法的运行速度提高了14倍左右。

Research on Spatial Sensitivity Analysis Using Parallel Algorithm Based on MapReduce

In recent years, with the rapid development of remote sensing technology, the spatial data represented by remote sensing images is widely used in ecosystem modeling, which promoted the development of ecological remote sensing parametric model in the regional scale. Sensitivity analysis is a key step for ecosystem model un-certainty quantification. It can identify the dominant parameters, reduce the model calibration uncertainty, and en-hance the model optimization efficiency. Due to the intensive computation of spatial data during the sensitivity analysis, the traditional stand-alone environment cannot meet the requirements of rapid analysis for the regional scale remote sensing parametric model. This study designed and realized a parallel algorithm of Sobol′ spatial sensitivity analysis utilizing Hadoop, which is an open source cloud computing platform, based on VPM （Vegeta-tion Photosynthesis Model）. In order to verify the efficiency of the algorithm, we designed a comparison experi-ment to compare the efficiency differences of the traditional serial algorithm and the parallel algorithm. The par-allel programming technology we used in this research was MapReduce, which divided the processes of map sampling and the iterative calculation during the spatial sensitivity analysis into subtasks, and assigned them to multiple computing nodes for parallel computing. The numerical experiment showed that the parallel strategy proposed in this study effectively shortened the time of model iterative calculations and significantly improved the efficiency of spatial sensitivity analysis for ecological remote sensing parametric model. Compared with the serial algorithm, the computing efficiency of the parallel algorithm was enhanced by 14 times.