Variation analysis of lake extension in space and time from MODIS images using random sets

Abstract

Understanding inundation in wetlands may benefit from a joint variation analysis in changes of size, shape, position and extent of water bodies. In this study, we modeled wetland inundation as a random spread process and used random sets to characterize stochastic properties of water body extents. Periodicity, trend and random components were captured by monthly and yearly random sets that were derived from multitemporal images. The Covering-Distance matrix and related operators summarized and visualized the spatial pattern and quantified the similarity of different inundation stages. The study was carried out on the Poyang Lake wetland area in China, and MODIS images for a period of eleven years were used. Results revealed the substantial seasonal dynamic pattern of the inundation and a subtle interannual change in its extension from 2000 to 2010. Various spatial properties including the size, shape, position and extent are visible: areas of high flooding risk are very elongated and locate along the water channel; few of the inundation areas tend to be more circular and spread extensively; the majority of the inundation areas have various extent and size in different month and year. Large differences in the spatial distribution of inundation extents were shown to exist between months from different seasons. A unique spatial pattern occurred during those months that a dramatic flooding or recession happened. Yearly random sets gave detailed information on the spatial distributions of inundation frequency and showed a shrinking trend from 2000 to 2009. 2003 is the partition year in the declining trend and 2010 breaking the trend as an abnormal year. Besides, probability bounds were derived from the model for a region that was attacked by flooding. This ability of supporting decision making is shown in a simple management scenario. We conclude that a random sets analysis is a valuable addition to a frequency analysis that quantifies inundation variation in space and time.