Article Preview
TopIntroduction
Analysis of vegetation life cycles is fundamental in monitoring and planning agricultural endeavors and optimizing land use. In particular, gaining knowledge of current vegetation trends and using them to make accurate predictions is essential to minimize times of food scarcity and manage the consumption of natural resources in underdeveloped countries. Needing to understand the Earth’s ecology and land cover is increasingly important as the impacts of climate change start to affect animal, plant, and human life. Vegetation trends are also closely related to sustainability issues, such as management of conservation areas and wildlife habitats, precipitation and drought monitoring, improving land usage for livestock, and finding optimum agriculture seeding and harvest dates for crops.
For this reason, there are many agencies and organizations that focus on the study of land use and land cover trends, linking them to climate change and the socioeconomic consequences of these changes. The United States Global Change Research Program (Land Use and Land Cover Change Interagency Working Group), the United Nations Framework Convention on Climate Change (Land Use, Land Use Change, and Forestry), and NASA’s Land Cover Land Use Change Program are just three examples of well-known interdisciplinary/ interagency programs that conduct and sponsor research related to the question of global land change as noted in OCHA (2011).
Assessment of changes in a region’s vegetation structure is challenging, especially in topographically diverse areas, like East Africa. Regions with vegetation changes are often the areas of most interest in land use management. Ideally, an automated screening process can identify areas with significant vegetation changes and facilitate objective decision making about land-use management such as in Cressie and Wikle (2011).
As a first step in creating an automatic screening processes, data collection on vegetation and land cover is needed. This is typically done through satellite remote sensing. The remote sensing imagery is used to convert the observed elements (i.e., the image color, texture, tone, and pattern) into numeric quantities at each pixel in the image. The image pixels correspond to a square grid of land, the size of which depends on the satellite’s resolution. One such numeric indicator is the normalized difference vegetation index (NDVI). In this article, the NDVI series came from satellite remote sensing data collected between 1982 and 2006 over 8,000-meter grid points. It has been shown to be highly correlated with vegetation parameters such as green-leaf biomass and green-leaf area, and hence is of considerable value for vegetation monitoring as in Curran (1980) and Jackson et al. (1983).
The NDVI standard scale ranges from −1 to 1, indicating how much live green vegetation is contained in the targeted pixel. An NDVI value close to 1 indicates more abundant vegetation. For example, low values of NDVI (say, 0.1 and below) correspond to scarce vegetation consisting mostly of rock, sand and dirt. A range of moderate values (0.2 to 0.3) indicates small vegetation such as shrub or grassland; larger NDVI values can be found in rainforests (0.6 to 0.8). Often, negative NDVI values are consolidated to be zero since negative values indicate non-vegetation and are of little use for vegetation monitoring. Vegetation activity is a continuous space-time process and NDVI data provide a space-time lattice system, in the sense that observations are available over equally spaced regular grids. Often, the spatial resolution ranges from 1000 to 8000 meters, while the temporal one ranges from 7 days to 1 month.
Statistical and computational methods are needed to analyze remotely sensed data, like NDVI values, to determine trends in land condition and to predict areas at risk from degradation. Methodologies that detect land cover changes need to be sensitive as well as accurate, since it can be costly and risky to relocate human populations, agriculture or livestock to new regions of detected change. In such spatio-temporal data, time series models are tempting for representing such processes. Other existing change detection methodologies include the geographically weighted regression of Foody (2003), the principal component analysis of Hayes and Sader (2001), and the smoothing polynomial regression of Chen & Tamura (2004). However, these methods are unable to provide an upper bound on false detections. Since there is large risk associated with falsely declaring an area to have significant vegetation changes, land use managers seek new methods that have a meaningful control over such errors.