基于Google Earth Engine的重庆市植被指数长时间序列S-G滤波方法的改进与实现  

Reconstruction of Chongqing’s Long Time-series NDVI through an Improved S-G Filter based on Google Earth Engine

作　　者：吴川虎 陶于祥 罗小波[1,2] Wu Chuanhu;Tao Yuxiang;Luo Xiaobo(Department of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Spatial Big Data Research Center,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065 [2]重庆邮电大学空间大数据研究中心,重庆4000065

出　　处：《遥感技术与应用》2021年第5期1189-1198,共10页Remote Sensing Technology and Application

基　　金：国家自然科学基金项目(41871226)。

摘　　要：由于云、大气等因素,归一化植被指数(Normalized Difference Vegetation Index,NDVI)时间序列数据存在一定的误差。Savitzky-Golay(S-G)滤波方法能够在一定程度上减小这种误差,抑制突降的低质量像元值,但对于低质量像元高值的抑制和高质量像元值的保护有所欠缺,而且不能很好的运用于不同时间间隔的时间序列影像重建中。基于Google Earth Engine(GEE)云平台,综合利用空间插值和时间滤波以及像元质量分析对重庆2014年春至2018年冬250 m分辨率的MOD13Q1长时间序列数据集进行重建研究。同时使用皮尔逊相关系数(Pearson)、新提出的平滑度指数以及NDVI变化差值,对样本点和整幅影像定量对比重建结果。研究表明:相同参数下,新方法的重建影像与原始影像的相关性高于S-G方法;模拟噪声实验中,其与两幅模拟噪声影像的相关性分别为0.87和0.94,而SG方法的相关性仅为0.65和0.61。Due to factors such as clouds and atmosphere,there are certain errors in the Normalized Difference Vegetation Index(NDVI)time series data set.The Savitzky-Golay(S-G)filtering method can reduce this er⁃ror to a certain extent and suppress the sudden drop of low-quality pixel values,but it is lacking in the suppres⁃sion of high-value low-quality pixels and the protection of high-quality pixel values.,And cannot be used well in time series image reconstruction at different time intervals.Based on the Google Earth Engine(GEE)cloud platform,a comprehensive use of spatial interpolation,temporal filtering,and pixel quality analysis to recon⁃struct the 250m resolution MOD13Q1 long-term data set in Chongqing from spring 2014 to winter 2018.At the same time,the Pearson correlation coefficient(Pearson),the newly proposed smoothness index and the differ⁃ence of NDVI change are used to quantitatively compare the reconstruction results of a single sample point and a single image.Research shows that under the same parameters,the correlation between the time series recon⁃structed by the new method and the original image is higher than that of the S-G method;in the simulated noise experiment,the correlation between it and the two simulated noise images are 0.87 and 0.94,respectively,while the correlation of the S-G method is only 0.65 and 0.61.

关 键 词：S-G滤波 植被指数 时间序列重建 Google Earth Engine(GEE) 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置] TP79[自动化与计算机技术—控制科学与工程]