Assessing Seasonal Vegetation Change In Coastal Wetlands With Airborne Remote Sensing PDF Download

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. Despite their tremendous economic and ecological value, substantial uncertainty still exists about wetland ecosystem function, habitats and response to natural and anthropogenic stressors worldwide. This uncertainty is further aggravated by constrained field access and surface heterogeneity which limit the accuracy of wetland analyses from remote sensing images. 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. Following the assessments of Poyang Lake dynamics in the low water season, I further examined which landscape characteristics of the permanent sub-lakes and their 500-m neighborhoods extracted from 30-m Landsat satellite imagery could explain non-uniform spatial distribution of waterbird diversity and abundance in the ground bird survey of December 2006. I hypothesized that the indicators of habitat size, spectral greenness, spectral and geometric patch heterogeneity would be positively associated with bird diversity and abundance, while the proportions of cover types approximating human disturbance would be negatively related to response variables. In the best-fit regression models selected using the Akaike Information Criterion, on average higher bird diversity and abundance were associated with larger sub-lake size, higher spectral greenness of emergent grassland and lower spectral greenness of mudflat as well as lower proportion of flooded/aquatic vegetation. At the same time, predictive performance of the best-fit models was penalized by large amounts of unexplained variation and inconsistencies among bird survey and remote sensing data from another year. Significant spatial autocorrelation in linear regression models raised concerns about missing predictor variables and the utility of sub-lakes as spatial units for diversity analysis, but it also suggested new hypotheses on spatial ecological interactions in bird community variables and habitat characteristics among sub-lakes. Research challenges identified in this study suggest that future monitoring programs should take more rigorous steps to standardize the protocols of bird surveys and improve spatial and temporal frequency of both bird and habitat observations. Rapid short-term surface variation and problematic field access will likely continue to limit remote sensing-based analyses of Poyang Lake wetlands and their habitats by traditional, static-class approaches. Using "dynamic" classes representing characteristic wetland transitions and disturbance regimes may provide more ecologically informative targets for management, conservation and modeling of ecosystem change. Object-based image analysis is a potentially powerful and promising approach to enhance classification accuracy of remote sensing data and ecologically informative interpretations of complex, heterogeneous wetland surfaces such as the study area. However, this methodology should be developed further to allow for more automated optimization of landscape object properties to capture vegetation patch structure and quantitatively assess propagation of the uncertainty among different spatial scales of the analysis. Finally, future studies should explore new ways of overcoming the limitations of problematic field access and frequent cloudiness obstructing the view of remote sensors by more rigorous utilization of in situ wireless sensors to record environmental conditions and surface composition and by introducing airborne lake-wide imaging programs for periods of prolonged cloudiness.