基于FY3/MERSI数据的辽宁省植被指数重建及时空变化分析  

Reconstruction and spatial-temporal variation analysis of the vegetation indices in Liaoning province based on FY3/MERSI data

作　　者：冯锐[1,2] 纪瑞鹏 武晋雯[1,2] 于文颖 刘丹[3] 陈妮娜 王莹[4] 张玉书 FENG Rui;JI Rui-peng;WU Jin-wen;YU Wen-ying;LIU Dan;CHEN Ni-na;WANG Ying;ZHANG Yu-shu(Institute of Atmospheric Environment,China Meteorological Administration,Shenyang 110016,China;Key Laboratory of Agrometeorological Disasters,Liaoning Province,Shenyang 110166,China;Heilongjiang Province Institute of Meteorological Sciences,Harbin 150030,China;Ecological Meteorology and Satellite Remote Sensing Center of Liaoning Province,Shenyang 110016,China)

机构地区：[1]中国气象局沈阳大气环境研究所,辽宁沈阳110166 [2]辽宁省农业气象灾害重点实验室,辽宁沈阳110166 [3]黑龙江省气象科学研究所,黑龙江哈尔滨150030 [4]辽宁省生态气象和卫星遥感中心,辽宁沈阳110166

出　　处：《气象与环境学报》2022年第1期65-73,共9页Journal of Meteorology and Environment

基　　金：风云卫星应用先行计划(FY-APP-2021.0302);辽宁省民生科技计划项目(2021JH2/10200024);辽宁省自然基金指导计划(2019-ZD-0857);辽宁省重点研发计划项目(2019JH2/10200018);辽宁省农业攻关及产业化项目(2018108004)共同资助。

摘　　要：为建立中国风云三系列气象卫星长时间序列归一化植被指数数据集,选用滤波和函数拟合方法,针对林地、湿地、水稻、玉米、大豆、城市和水体7类地物开展数据重建效果定量分析,确定最佳数据重建方法,并在辽宁省开展时空变化分析。结果表明:非对称高斯函数拟合法(Asymmetric Gaussians,AG)、Savitzky-Golay滤波法(SG)、双Logistic函数拟合法(Double Logistic,DL)和时间序列谐波分析法(Harmonic Analysis of Time Series,HANTS)四种方法均表现出相对较好的去噪能力。SG方法对噪声比较敏感,HANTS方法在低值区受噪声影响大。AG和DL方法平滑效果较好,DL方法的峰值更接近于原始峰值。在高植被覆盖区和季节性作物区,SG方法相关系数最高(>0.93)、均方根误差最低(<0.1);在城市和水体低植被指数区,HANTS方法相关系数最高,为0.87,但四种方法的均方根误差均在0.06左右,差别不大。综合考虑曲线和定量分析结果,选取SG方法进行辽宁省植被指数数据集数据重建。辽宁省植被指数数值高低的空间分布与下垫面植被类型相符合,东部山区林地植被指数最高,达到0.75以上。2009—2020年,辽宁省NDVI年均值存在波动,不同地物植被指数变化存在差别,水体和城市植被指数变化相对较小,旱田作物(玉米、大豆)的植被指数受干旱年的影响植被指数变化稍大。辽宁省主要粮食作物植被指数年内均呈单峰分布,与一年一熟型吻合,均在8月上旬达到最大值。To establish a long-term normalized difference vegetation index(NDVI)data set with the FY-3 series of Chinese meteorological satellites,four filtering and function fitting methods were used to quantitatively analyze the results from reconstructed data on seven types of ground features including forest land,wetland,waterbody,urban,rice,soybean,and corn.Determination for the best data reconstruction method and spatial-temporal variation analysis on the vegetation indices in Liaoning Province was further conducted.The four methods including Asymmetric Gaussian function(AG),Savitzky-Golay filtering(SG),Double Logistic function(DL),and Harmonic Analysis of Time Series(HANTS)show more effective denoising abilities.The SG method was more sensitive to noise overall,whereas the HANTS method was highly affected by noise in the low-value areas.The AG and DL methods had better smoothing effects,while the peak value of the DL method was closer to the original peak value.In areas with high vegetation coverage and seasonal crops,the SG method had the highest correlation coefficients(>0.93)and the lowest root mean square errors(<0.1).In areas with a low vegetation index,such as cities and water bodies,the HANTS method had the highest correlation coefficient of 0.87,but the root mean square errors of all four methods were around 0.06 with little discrepancies.Considering the curve and quantitative analysis comprehensively,the SG method was selected to reconstruct the vegetation index data set of Liaoning Province.The spatial variations of vegetation indices were consistent with the vegetation types of the underlying surface.The vegetation indices of forest land in the eastern mountainous areas were the highest with values of over 0.75.During 2009-2020,the annual average NDVI values in Liaoning Province experience fluctuation,and there were differences among the variations of vegetation indices for different ground features.The variations for water bodies and cities were relatively small,whereas those of the dry field crops(e.g.ma

关 键 词：FY3/MERSI 归一化植被指数 数据重建 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]