分布式属性相似空间插值算法及其对流域土壤属性插值的计算性能评估

DISTRIBUTED ATTRIBUTE-SIMILARITY SPATIAL NTERPOLATION ALGORITHM AND ITS PERFORMANCE EVALUATION ON SOIL ATTRIBUTE INTERPOLATION AT THE WATERSHED SCALE

[目的]如何准确、高效地实现点源数据向面源数据的空间拓展,是现代数字土壤制图技术实现由中小尺度制图转向大尺度、乃至全球尺度制图所需解决的关键问题之一。[方法]依托前期提出的局部属性相似性加权回归空间插值算法(LASWR),文章构建了一种基于云计算技术Hadoop的分布式空间插值算法(DLASWR),以应对大规模数字土壤制图的海量数据处理需要。DLASWR算法是基于Hadoop的MapReduce框架结构。算法的核心思想在于:(1)将待插值数据集分割成多个子数据集,由单个Map任务使用集中式LASWR算法对一个子数据集进行插值;(2)由Reduce任务归并所有Map任务的插值结果,并作为最终结果输出。[结果]对实际土壤样点属性的空间插值实验结果表明DLASWR算法具有良好的加速性能,与集中式LASWR算法相比显著提高了空间插值的计算效率。[结论]DLASWR算法可为数字土壤制图领域当前应用的空间插值方法由传统的集中式计算拓展成分布式计算提供技术参考。

Accurately upscaling data from point to spatial surface data with high efficiency is one key problem to be solved in the field of modern technologies for digital soil mapping that focuses on upscaling maps at small and medium scales to large,or even global scales.In order to meet the necessity of processing large amount of data by digital soil mapping,this study developed a distributed local attribute similarity weighted spatial interpolation model(DLASWR)based on the cloud computing technology of Hadoop and a local attribute similarity weighted spatial interpolation model(LASWR).The MapReduce framework in the Hadoop was adopted as the base for the DLASWR model.The core principle of the model included:(1)The database to be interpolated was first divided into multiple subsets,on which the centralized LASWR algorithm was implemented using a single Map mission.(2)The Reduce mission collected results from all Map mission and exported the outputs.Evaluation results by interpolating observed spatial soil properties indicated that the DLASWR model showed excellent accelerating capabilities,and the computing efficiency was significantly improved compared to those by the LASWR model.In summary,the DLASWR model provides technological reference for extending the current spatial interpolating method in digital soil mapping from centralized to distributing computing.