Research On Land Cover Change Detection Methods Using Time Series NDVI

Along with the rapid development of science and technology, the land use intensity by people is strengthening, which results in the rapid change of land surface biochemical cycle, hydrograph and landscape dynamical state, and becomes an important factor affecting global change. At the same time, land cover change is the direct reflect of land use. Therefore, the research on land cover change and its spatial temporal regulation has become the hot spot of global change research. As remote sensing has the advantage of observing large scale earth surface rapidly and accurately, it has become the main tool of data acquiring for land cover change detection.Land surface change is a long, dynamical and complex process, so researches on few scenes can’t reflect the real change information and regulation, and long term persistent observing data is needed. With the development of remote sensing technology, the multi-scale high dynamical and persistent observation of the earth is achieved, and a large amount of data is accumulated, which has detailed recorded the process and track of various changes on the earth surface. However, there are not only changes caused by vegetation phenology, climate condition changes recorded in the remote sensing images, but also changes caused by variations of atmosphere condition and the imaging sensor. Thus, it is necessary to remove the changes caused by interfering information to derive the real land cover change information. Therefore, a lot of difficulties occur in land cover change technique based on time series remotely sensed images.The current research of multi temporal remote sensing images mainly reference the two scence image change detection theory. As to the change detection technology indirectly from the multi temporal remote sensing images, the theory foundation is too weak relatively, although a lot of work has been done from different aspect. NDVI is an important remote sensing parameter, which can sensitively reflect the change of vegetation growth condition, biochemical property, and the ecosystem parameter. It can also reflect the composite condition of the land cover type according to the pixel on the image. Therefore, vegetation indices such as NDVI are used for land cover change research instead of land cover change information. In this paper, time series of NDVI images are used for land cover change detection.According to the abnormal essence of land cover change in long time series, land cover change is linked to NDVI anormality based on the analysis of character of land cover in time series. Based on the response characteristic of NDVI to land cover type, NDVI time series and anormaly detection are used to study land cover change. The main work and innovations are as follows.1) A solution for long time series NDVI dataset reconstruction is proposed based on spatial temporal correlation. On account to the issue that there is much noise in the long time series NDVI dataset, a scheme for time series NDVI dataset restruction basd on temporal filter and spatial filter is proposed. First, the inner atmosphere noise is filtered by modeling the short term evolution of NDVI with SG filter, based on the temporal continuity, which is the first try of using SG in long time series NDVI rebuilt. Secondly, the lost value caused by long-term cloud is filled based on interannual anomaly. Then boundary abnormal values are rectified using spatial filter and high quality NDVI time series dataset is got.2) The spatial temporal variation of NDVI and its response regulation of land cover type are analyzed. Various kinds of indictors are used to intensively study the spatial-temporal variation discipline of NDVI, including the spatial-temporal distribution, seasonal and interannual spatial-temporal variation. The reponse of NDVI evolution to land cover type is explored, with the assistant of DEM and land cover map. Spatial temporal heterogeneity is found in the absolute value of NDVI variation when the land cover types change.3) A noval land cover change detection methos is proposed based on time series abnormal point detection theory. According to the abnormal essence of land cover change in long time series, land cover change is linked to NDVI anormality based on the analysis of NDVI interannual variation in various scenes of land cover change. The the theory of land cover change detection based on anormaly detection is proposed, which gives on solution to the contradiction faced in traditional methods that thresholds differ spatially because of sptial heterogeneity. Based on the review of methods of spatial temporal outlier detection, a scheme for the land cover change detection based on the current NDVI dataset according the small sample character of NDVI dataset. Then the feasibility of this method is proved by experiment with real data.4) An Intergrated Change Detection Indictor (ICDI) for multi-dimention vector based on similarity of time series curve is proposed, which can restrain the effection of phonological drift to land cover change detection. The accuracy of land cover change detection using NDVI time series depends on the ability of describing the similarity of NDVI curve, while the traditional indictiors of land cover change are designed from the aspect of similarity coefficient or distance separately. Therefore, the indictor ICDI is proposed combing curve shape and distance, which can greatly reduce the accuracy loss caused by the interannual phonological drift and resist noise well. Then a scheme of land cover change detection based on similarity and anomaly detection methods for NDVI time series dataset according. Finally, the method is proved to be operational with real data experiment and applied to Jiangxi province.5) A feasible research system is built using NDVI, an easily obtained remote sensing parameter, to study the complicated change of land cover type. Based on the study of spatial-temporal variation of NDVI and its response to land cover change, the NDVI variation is related to land cover change. Then, some land cover change detection methods are proposed using NDVI time series images from many aspects based on the their characteristion of land cover type. Solutions are given to some specific application problem, which gives some new methods to land cover change detection theory.