PROGRESS IN GEOGRAPHY ›› 2018, Vol. 37 ›› Issue (3): 427-437.doi: 10.18306/dlkxjz.2018.03.013

• Articles • Previous Articles     Next Articles

Comparison and application of NDVI time-series reconstruction methods at site scale on the Tibetan Plateau

Jianwen LIU1,2(), Yuke ZHOU3,*()   

  1. 1. Key Laboratory of Spatial Data Mining and Information Sharing of Ministry of Education, Fuzhou University, Fuzhou 350003, China
    2. Spatial Information Research Center of Fujian, Fuzhou 350003, China
    3. Key Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
  • Received:2017-04-12 Revised:2017-11-08 Online:2018-03-28 Published:2018-03-28
  • Contact: Yuke ZHOU;
  • Supported by:
    National Natural Science Foundation of China, No.41601478;National Key R&D Program of China, No. 2016YFC0500103;Science Technology Service Project of Chinese Academy of Sciences, No. KFJ-SW-STS-167


Vegetation index data based on satellite remote sensing have been widely applied in many fields such as vegetation monitoring and biomass estimation. However, the noise caused by the sensors, atmospheric conditions, and environmental factors affect the application of such data. Therefore, it is of practical significance to carry out research on the reconstruction of vegetation index. In this article, MODIS normalized differential vegetation index (NDVI) datasets of 106 meteorological stations on the Qinghai-Tibet Plateau with typical vegetation types, from 2000 to 2015, were reconstructed using three methods: cubic spline function (Spline), double logistic function (D-L), and singular spectrum analysis (SSA). Based on the results of phenological parameter extraction and NDVI time-series reconstruction, the ability of preserving the authenticity of the original data, detail fitting, and phenological character extraction of the three algorithms are compared and analyzed. The result suggests that there is no single method that performed the best for all types of vegetation, owing to the spatial heterogeneity of vegetation cover types and the varying denoising ability of the algorithms. D-L showed better performance for desert, shrubs, and woodland, which are heavily affected by random noise from snow and clouds, whereas for grassland and cropland, Spline had better performance. The fitting curves of SSA are below the D-L and Spline curves and the lower the NDVI value, the worse the reconstruction performance. With regard to maintaining the true values of the original data, Spline is superior to the D-L method for woodland, shrubs, and grassland, while the SSA is superior to the D-L function for woodland. The D-L method resulted in an earlier phenological period, Spline and SSA had a lagged outcome of the start of the growing season (SOS) and end of the growing season (EOS) respectively. Phenological index is unstable for shrubs and woodland and fluctuate irregularly for desert with all three methods, due to the low absolute value of NDVI. In addition, the phenology curve of spline is similar to that of SSA, and the SOS and EOS index derived from the D-L method is ahead of other methods for all vegetation cover types. The EOS extracted by the SSA method is larger than that by other methods for most of the vegetation cover types. This research could provide a reference for the selection of noise reduction methods for NDVI time-series data with different vegetation cover types in plateau vegetation.

Key words: time-series data reconstruction, vegetation phenology, double logistic fitting, cubic spline, singular spectrum analysis