The Abnormal Segments Reconstruction And Phenology Detection Of The HJ Satellite Normalized Difference Vegetation Index Time-series

HJ-1A/B small satellite constellation has a2days repetition cycle and30m spatial resolution (multispectral remote sense image). This means that HJ-1A/B can provide hyper-temporal Normalized Difference Vegetation Index (NDVI) time-series with a medium-high spatial resolution. However, the quality of the HJ NDVI time-series can be abnormally low. The low quality series even appear in succession, which is referred to as the abnormal segment. The composing method or the quality flags may not solve this problem satisfactorily and therefore a large amount of noise and long abnormally low periods often remains in the HJ NDVI time-series. Also, the actual resolution of the HJ NDVI time-series on time dimension is not clear. It is lack of application and evaluation. To these, the thesis completed the following studies:(1) This thesis introduces the build process of HJ NDVI time-series, compares the quality between HJ and MODIS NDVI time-series and analyzes the causes of abnormal value in HJ NDVI time-series. Research has shown that for the pure pixels, the MODIS and HJ NDVI time-series description about the ground feature change are similar, their correlation coefficient is higher than0.7. For the mixed pixels, the HJ NDVI time-series can describe the difference of the features in mixed pixels. This shows that the HJ NDVI time-series have higher spatial resolution but the similar time dimension resolution comparing with MODIS NDVI.(2) Reconstruct HJ NDVI time-series by the Savitzky-Golay filter, Asymmetric Gaussian function fitting method and Harmonic Analysis of Time-Series. The reconstruction results show that these three methods can complete the reconstruction of HJ NDVI time-series. Each reconstruction method has its own characteristic and can obtained relatively good results in good meteorological condition area. But in many cases, especially in the rainy season or areas with serious atmospheric pollution, the low quality series often appear in succession, which is referred to as the HJ NDVI abnormal segment. This abnormal phenomenon cannot be removed by the method introduced above. Therefore we need to propose a specific method aimed to HJ NDVI time-series abnormal segments, to improve the series quality.(3) Present a method to reconstruct the abnormal segments in the HJ NDVI time-series with the assistance of the MODIS NDVI time-series. The co-integration test was adopted to decide whether the MODIS can be used for the reconstruction of NDVI time-series for the corresponding HJ image pixels. Statistical quality control methods were used to single out the abnormal segments in the HJ NDVI time-series and establish an error correction model (ECM) that combines MODIS and HJ NDVI time-series to perform the reconstruction. The results show that abnormal segment in the HJ NDVI time-series can be corrected through the proposed method.(4) Present a vegetation phenology detection method using HJ NDVI time-series and analyze the time accuracy of the detection results. The phenology detection parameter calibration process is proposed in the detection method. This process is used to find the correspondence between the HJ NDVI time-series curve characteristics and the phenology time. The vegetation phenology are be detected by the feature extraction of HJ NDVI corrected and reconstructed time-series and its first/second order derivative series. The results show that using HJ NDVI time-series can distinguish the12key phenological dates of winter wheat and rice from seedling emergence to the harvest date. There are5phenological dates estimating results are less than10d against the statistical data, the rest are16d. The results demonstrate HJ NDVI time-series and MODIS NDVI time-series have equal phenology detection ability, indicating they have similar time dimension resolution.