Characterizing The Remotely Sensed Time Series Data For Ecological Applications

Land cover changes have always been the focus of ecological researches. In the scenes of global changes, better understanding the processes of land cover changes are undoubtly of great importance. Though traditional ecological experiments can acquire detailed land cover characteristics, however, as a result of limited human resources and funds support, these researches are mainly restricted at local scales.Satellite remote sensing sensors can acquire land cover reflectance at the global scale. Among them, the wide-view-of-field sensors have caught much attention, for they are of wide swaths and high revisit frequencies, and the data they acquired possess multiply kinds of information (spatial, spectral, and temporal information). This gives us an opportunity to thoroughly explore the ecosystem characters (including land cover characteristics) at larger scale. And the way to achieve this goal is by exploring time series data. However, as time series data is not designed for ecological studies, it is inevitable that time series data cannot fully matched with ecological requirements, and a series of problems arises in ecological applications. Therefore, in this paper, the time series data were adapted, and analyses were performed from various aspects to handle those problems. In this process, revised methods or solutions were proposed to increase the efficiencies of time series data analyses, and to improve its applications in ecological reasearches. The paper consists of following parts:(1) A time series data analysis method was first proposed in this paper. Though other methods had been proposed previously, they are often based on mathematical standards and it is difficult to choose an ideal method that only based on mathematics principles. Therefore, a new method for analyzing time series data was introduced from the ecological perspective. To simulate ecosystems with different growth patterns, five land cover types that differ in vegetation coverage and disturbances were chosen for study. Parameters aimed to reflect land cover characters were extracted based on sustained vegetation growth trend, and improvement that inspired by a method used for determining the hyperspectral red edge position was also made in this process. Results showed that this revised analysis method can identify land cover characteristics of different vegetation growth patterns, and the extracted parameters can be used for land cover classification, and the growth differences between herbaceous and woody plants, evergreen and deciduous plants can also be conveyed.(2) Then, an analysis was taken to evaluate whether the extent of study area can affect researches aimed to identify land cover characteristics by using remote sensing data, and time series data was used as a type of remote sensing data. A region covers a MODIS tile (h26v04) was used for analysis, and pixel size of was used to represent the coverage area. This region contains 8 natural land cover types in the IGBP system. And the total pixel of each land cover type varies from 634 to 1,734,460. Random samplings (100,000 runs) were used to test whether changes in sample pixel size can affect the land cover characteristics. Results showed that, large research area is not necessary for ecological research that use remote sensing data, and a relative small area (<100 pixels) can represent the characteristics of a large region. This result does not vary as the analyzing year, studied land cover type, or studied parameter changes. This indicates when doing land cover related researches, targeted experiments can be designed in a small area, and the relations between ground measured results and remote sensing analysis results can be carefully surveyed for further exploration of land cover changes.(3) After that, an analysis was conducted to test whether the temporal resolution would affect identifying land cover characteristics. This question was analyzed from a perspective of noise, and both original data quality and practical use were considered in the study. The noise exists because remote sensing sensors cannot detect vegetation signals as a result of cloud or snow. As climate varies around the world, the noise frequencies differ in regions. Five land cover types that in two regions, where noise contamination are apparent in time series data, were selected for study. Time series data that of different temporal resolutions were created based on the quality assessments of daily data, and this composition strategy follows the principles adopted by MOD09 products. Results showed that the quality of time series data varied with changing temporal resolution, and the best data quality exist when temporal resolution is between 10 and 11 day. Temporal resolutions suitable for application were further detected based on practical aspects of various ecological applications, including study aim, noise pattern of studied area, and procedure to deal with time series data. Results showed that the 8-day and 10-day composition schemes adopted by current remote sensing time series products could satisfy most ecological studies, and can be used for analyze land cover changes.(4) Finally, in the last chapter, an analysis was performed to test whether the application of time series data could cause errors when analyzing land cover characteristics. Results showed that, the phenological information extracted from time series data cannot be used to reflect land cover changes if there is no other auxiliary data. Also, if these data cannot be used properly, misunderstanding could arise.