MODIS EVI时序数据重建方法及拟合分析
作者简介:王乾坤(1987-),河南焦作人,男,硕士生,研究方向为3S技术在植被遥感监测中的应用。E-mail: wqkone.2@163.com
收稿日期: 2014-10-22
要求修回日期: 2014-11-24
网络出版日期: 2015-06-10
基金资助
国家自然科学基金青年基金项目(41001279)
资源与环境信息系统国家重点实验室青年人才培养基金项目(O8R8B690PA)
中国科学院战略性先导科技专项“应对气候变化的碳收支认证及相关问题”(XDA05050102)
Comparison on Three Algorithms of Reconstructing Time-series MODIS EVI
Received date: 2014-10-22
Request revised date: 2014-11-24
Online published: 2015-06-10
Copyright
植被遥感监测中长时间序列数据择优的重建方法,已成为当今一个研究热点。本文以东北地区5种主要植被覆盖类型为例,在定性分析TIMESAT提供的3种常用重建方法对EVI(Enhanced Vegetation Index)时序曲线重建效果的基础上,定量对比研究了各方法,对原始高质量EVI点真实值的保真性,及对原始曲线整体特征的保持度。结果表明:S-G(Savitzky-Golay)滤波对原始曲线生长季的峰值及宽度重建效果较好,但容易因过度拟合保留过多噪声,特别是草地和灌丛类型;非对称性高斯函数(AG)和双Logistic曲线(DL)方法相似,对草地、灌丛和耕地的重建结果更接近真实值,但AG拟合对波峰处异常值的处理结果较差,重建后波峰表现低平。3种算法对原始EVI时序数据的保真性和对原始时序数据曲线特征的保持度,都表现出与植被类型分布相关的空间分布格局。分析结果表明,在东北地区,AG算法对草原和灌丛的重建效果最好,DL算法对耕地重建效果最优,S-G算法最适合对落叶阔叶林和落叶针叶林进行重建处理。
关键词: MODIS EVI; 非对称性高斯函数拟合; 双Logistic曲线拟合; Savitzky-Golay滤波; 中国东北
王乾坤 , 于信芳 , 舒清态 , 尚珂 , 文可戈 . MODIS EVI时序数据重建方法及拟合分析[J]. 地球信息科学学报, 2015 , 17(6) : 732 -741 . DOI: 10.3724/SP.J.1047.2015.00732
With the rapid development of remote sensing techniques, higher precisions of the vegetation remote sensing are required. Therefore, before using the time-series data, how to select the optimal algorithms to reconstruct it has been a hot research topic. Based on the five main land cover types in Northeast China, the reconstruction quality of three commonly used algorithms that included in TIMESAT tools has been qualitatively analyzed. Then, the fidelity performance and the capability to keep main characteristics of the three algorithms on EVI with respect to different land cover types were compared. The result shows that the S-G algorithm has a better performance in reconstructing the peak and the width of the EVI curves in the growing seasons, but it is prone to keep the noise data due to excessive fittings, especially common in land cover types of steppe and shrub. AG and DL algorithms generally present similar performances and the results are much closer to the true values for land cover types of steppe, shrub and arable land. But AG algorithm is easily influenced by noises for fitting the peak of the cures, which reduces the maximum EVI and causes the decline of vegetation growth. Spatial patterns of the fidelity performance and the capability to keep main characteristics of the three algorithms are all related to the distribution of vegetation types. Finally, we found that AG is a better algorithm to be used for the land cover types of steppe and shrub, DL is better for arable land, while S-G is better for the broadleaved deciduous forest and coniferous deciduous forest.
Key words: MODIS EVI; asymmetric Gaussian; double logistic; Savitzky-Golay; Northeast China
Fig. 1 Location of Northeast China图1 中国东北地区位置示意图 |
Tab. 1 Table of MODIS land quality assessment (MODLAND QA) of EVI表1 EVI植被指数总评质量评分表 |
QA可用性指数 | 质量描述 | 权重 |
---|---|---|
0 | 所有波段均为经校正的理想数据 | 1.0 |
1 | 经校正了的部分或全部波段为较理想数据 | 0.6 |
2 | 因云覆盖影响所有波段均未进行校正 | 0.0 |
3 | 因其他原因部分或全部波段未进行校正 | 0.1 |
Fig. 2 Spatial distribution of vegetation types in Northeast China图2 中国东北植被类型空间分布图 |
Fig. 3 The S-G curve of time series distribution with respect to 5 window settings图3 S-G滤波5种窗口设置方法时间序列分布曲线图 |
Fig. 4 Flowchart of comparative research of EVI series reconstruction techniques图4 时序EVI数据重建方法结果比较方法流程图 |
Fig. 5 Comparison of original and reconstructed MODIS EVI series regarding to different types of land cover from 2008 to 2010图5 不同土地覆被类型典型像元EVI时间序列重建前后曲线对比 |
Tab. 2 Average RMSE, RE and r values of different vegetation types from the three algorithms based on EVI series in Northeast China from 2008 to 2010表2 不同植被类型像元的平均RMSE、r和RE |
植被类型 | RMSE | r | RE | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
AG | DL | S-G | AG | DL | S-G | AG | DL | S-G | |||
落叶阔叶林 | 6.24 | 6.71 | 6.06 | 0.94 | 0.94 | 0.95 | -2.03 | -2.39 | -2.11 | ||
落叶针叶林 | 6.50 | 6.93 | 6.45 | 0.90 | 0.89 | 0.91 | -2.24 | -2.71 | -2.04 | ||
灌丛 | 4.98 | 5.22 | 5.81 | 0.93 | 0.92 | 0.90 | -1.59 | -1.82 | -1.67 | ||
耕地 | 4.72 | 4.68 | 6.28 | 0.95 | 0.95 | 0.92 | -1.34 | -1.67 | -1.44 | ||
草原 | 5.99 | 11.01 | 11.69 | 0.85 | 0.73 | 0.70 | -1.63 | -2.65 | -2.03 |
4.2.1 空间格局的保真性分析 |
Fig. 6 Spatial distribution of root mean square error (RMSE) between the origin and reconstructed high-quality EVI using the three algorithms in Northeast China (2008-2010)图6 东北地区3种方法重建前后高质量数据的RMSE空间格局(2008-2010年) |
Fig. 7 Fidelity performance statistics of the three RMSE based algorithms on EVI with respect to different land cover types (2008-2010)图7 基于RMSE的3种算法对不同植被类型EVI时序数据保真性分级统计(2008-2010年) |
Fig. 8 Spatial distribution of correlation coefficient (r) between the original and reconstructed EVI using the three algorithms in the Northeast China (2008-2010)图8 东北地区3种方法重建前后EVI时序数据相关系数的空间格局(2008-2010年) |
Fig. 9 Graded statistics of the capability to keep main characteristics of EVI using the three algorithms based on r with respect to different land cover types图9 基于r的3种算法对不同植被类型EVI时序数据主要特征保持度分级统计 |
Fig. 10 Spatial distribution of residual error (RE) between the origin and reconstructed EVI using the three algorithms in Northeast China (2008-2010)图10 东北地区3种方法重建前后时序EVI数据残差的空间格局(2008-2010年) |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
LPDAAC. MODIS User_Guide[EB/OL].https://lpdaac.usgs.gov/products/modis_products_table/mod09a1, 2011-02.
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
/
〈 | 〉 |