遥感植被指数时间序列数据集,已广泛应用于陆地生态环境变化监测与模拟、植被覆盖动态变化分析、植被物候特征识别与信息提取等多方面的研究。但其因受遥感器采集与传输过程、大气条件、地面状况等诸多因素的影响,时序NDVI数据包含各种噪声,因此研究者们发展了一系列时间序列曲线重建方法。本文对近年来提出或改进的重建算法原理、优缺点进行阐述;然后,选择当前最为常用的3种方法,即非对称高斯函数(AG)拟合、双Logistic曲线(D-L)拟合和Savitzky-Golay(S-G)滤波法,以藏北地区不同土地覆被类型样点像元NDVI时间序列为实例,对算法的去噪效果、保真性能、生长峰值及细节处理效果等方面进行比较研究。结果表明,AG与D-L拟合两种算法具有较好的一致性,但对生长峰值模拟有所差异;3种方法对荒漠、荒漠草原、草原、灌丛、作物用地及林地等不同覆被类型各具优势,表现出区域、覆被类型和应用目的差异性。最后,基于AG拟合算法对整个藏北地区2007-2009年MODIS NDVI时间序列进行重建,处理后像元NDVI空间格局异质性减弱。

Remote sensing vegetation index time-series datasets have been applied widely in various fields, such as monitoring and simulation of terrestrial ecological environment system on global and regional scales, analysis of vegetation cover dynamics, recognizing features and extracting information of vegetation phenology and so on. However, due to errors from information collection and transmission process of sensors, different atmospheric conditions and random ground situations etc, various residual noise exists in NDVI time-series data. Therefore, a series of NDVI time-series data reconstruction methods have been developed to solve the problem. In this paper, various newly proposed and modified reconstruction methods are summarized and evaluated firstly. Then the three primary methods, i.e. asymmetric Guassian (AG) function fitting, double logistic (D-L) fitting method and Savitzky-Golay (S-G) filteration, are introduced in terms of basic principles and features. Based on the TIMESAT program, by taking some NDVI time-series of pixel samples of different land cover types as instance, the suitability and defects of the three approaches are analyzed and discussed. The result shows that the former two methods generally present similar performances, but on fittings of peak of these time-series curves there are minor differences. Besides, the three methods have inconsistence effects in different study areas, land cover and research purposes. In detail, the S-G method has a more excellent performance in reconstructing NDVI curves in alpine desert and forest-covered area with much noise, while the D-L and AG function fitting methods could conduct more accurate reconstructed time-series dataset in alpine shrub or marshy grassland regions, especially for fitting the start and end period of vegetation growing season. For crops-planted land cover, the double logistic fitting method can't fit the correct curve of NDVI time-series, and the asymmetric Guassian function fitting has the best performance. Finally, based on AG function fitting method, the NDVI time-series of 2007-2009 in North Tibet are reconstructed, and the processed NDVI values are clearly more homogeneous than those in raw data.