The study of seasonal composition and dynamics of wetland ecosystems and wintering bird habitat at Poyang Lake, PR China using object-based image analysis and field observations

Wetlands are among the most productive ecosystems in the world which support critical ecological services and high biological diversity yet are vulnerable to climate change and human activities. In this thesis, I investigated the capabilities of satellite remote sensing with medium spatial resolution and object-based image analysis (OBIA) methods to elucidate seasonal composition and dynamics of wetland ecosystems and indicators of habitat for wintering waterbirds in a large conservation hotspot of Poyang Lake, PR China.;I first examined changes in major wetland cover types during the low water period when Poyang Lake provides habitat to large numbers of migratory birds from the East Asian pathway. I used OBIA to map and analyze the transitions among water, vegetation, mudflat and sand classes from four 32-m Beijing-1 microsatellite images between late fall 2007 and early spring 2008. This analysis revealed that, while transitions among wetland classes were strongly associated with precipitation and flood-driven hydrological variation, the overall dynamics were a more complex interplay of vegetation phenology, disturbance and post-flood exposure. Remote sensing signals of environmental processes were more effectively captured by changes in fuzzy memberships to each class per location than by changes in spatial extents of the best-matching classes alone. The highest uncertainty in the image analysis corresponded to transitional wetland states at the end of the major flood recession in November and to heterogeneous mudflat areas at the land-water interface during the whole study period. Results suggest seasonally exposed mudflat features as important targets for future research due to heterogeneity and uncertainty of their composition, variable spatial distribution and sensitivity to hydrological dynamics.;I further explored the potential of OBIA to overcome the limitations of the traditional pixel-based image classification methods in characterizing Poyang Lake plant functional types (PFTs) from the medium-resolution Landsat satellite data. I assessed the sensitivity in PFT classification accuracy to image object scale, machine-learning classification method and hierarchical level of vegetation classes determined from ecological functional traits of the locally dominant plant species. Both the overall and class-specific accuracy values were higher at coarser object scales compared to near-pixel levels, regardless of the machine-learning algorithm, with the overall accuracy exceeding 85-90%. However, more narrowly defined PFT classes differed in their highest-accuracy object scale values due to their unique patch structure, ecology of the dominant species and disturbance agents. To improve classification agreement between different levels of vegetation type hierarchy and reduce the uncertainty, future analyses should integrate spectral and geometric properties of vegetation patches with species' functional ecological traits.;In periodically flooded wetlands such as Poyang Lake, rapid short-term surface dynamics and frequent inundation may constrain detection of directional long-term effects of climate change, succession or alien species invasions. To address this challenge, I proposed to classify Poyang Lake wetlands into "dynamic cover types" (DCTs) representing short-term ecological regimes shaped by phenology, disturbance and inundation, instead of static classes. I defined and mapped Poyang Lake DCTs for one flood cycle (late summer 2007-late spring 2008) from combined time series of medium-resolution multi-spectral and radar imagery. I further assessed sensitivity of DCTs to hydrological and climatic variation by comparing results with a hypothetical change scenario of a warmer wetter spring simulated by substituting spring 2008 input images with 2007 ones. This analysis identified the major steps in seasonal wetland change driven by flooding and vegetation phenology and spatial differences in change schedules across the heterogeneous study area. Comparison of DCTs from the actual flood season with the hypothetical scenario revealed both directional class shifts away from expanding permanent water and more complex location-specific redistributions of vegetation types and mudflats. These outcomes imply that changes in flooding may have non-uniform effects on different ecosystems and habitats and call for a thorough investigation of the future change scenarios for this landscape. The possibility to disentangle short-term ecological "regimes" from longer-term landscape changes via DCT framework suggests a promising research strategy for landscape ecosystem modeling, conservation and ecosystem management. (Abstract shortened by UMI.).