Multi-Temporal Remote-Sensing-based Mapping and Characterization of Landscape Evolution of a Meandering River Floodplain
MetadataShow full item record
Large meandering river floodplains are critical components of the Earth ecosystems for their high biodiversity and productivity. However, it is challenging to study these regions because of their complex land-covers and dynamic surface processes. This study applies soft classification and change-detection analysis to five Landsat 5 Thematic Mapper (TM) satellite images to examine long-term surface-cover composition and configuration change of the Rio Beni floodplain in Bolivia from 1987 to 2006. One hard/crisp classification algorithm (i.e., ISODATA) and two soft classification algorithms (i.e., Bayes classification and fuzzy classification) were applied to the study-area satellite images to examine the performances of classifying and mapping meandering river-floodplain environments between hard and soft classification approaches. In all five scenes, three algorithms achieved ~90% classification accuracy via hard classification outputs. However, the two soft algorithms were of more utility in this study because their results were less affected by “salt-and-pepper” noise and provided extra land-cover probability/membership layers. A novel change-detection algorithm was proposed in this study, namely Modified Change Vector Analysis (MCVA). The MCVA operated in fuzzy-membership space, considered change uncertainty during the thresholding stage, and utilized change-vector directions to modify the determination of change/no-change status for each pixel. A fuzzy Markov Random Field (FMRF) model was applied to further refine the change maps by incorporating spatial change uncertainty. A second thresholding stage was also applied to separate a type of change referred to as “transitional change,” which preserved fuzzy membership information and provided a concise map output. Compared with three traditional change-detection algorithms, the MCVA achieved higher change-detection accuracy and provided more detailed change dynamics regarding the land-surface change. Dynamics of major floodplain cover types (i.e., oxbow lakes, river, sand, forest, non-forest vegetation, and dry and wet soil) were investigated via multi-temporal analysis. Over the observing period of 1987 to 2006, 74.4% of pixels remained the same land-cover, 20% experienced clear land-cover change and 5.6% experienced transitional land-cover change. The riparian area experienced more dramatic change than other parts of the Rio Beni floodplain during this period. Additional analysis of landscape metrics provided information regarding the spatial patterns of the land-cover, but future work would be needed to further examine its utility in understanding floodplain dynamics. This study provides information on remote-sensing-based mapping and quantitative characterization methods for meandering river floodplains. The spatiotemporal patterns of landscape on Rio Beni floodplain can be used in sustainable management and protection of floodplain ecosystems.
modified change vector analysis (MCVA)
meandering river floodplain
You, Mingde (2016). Multi-Temporal Remote-Sensing-based Mapping and Characterization of Landscape Evolution of a Meandering River Floodplain. Master's thesis, Texas A&M University. Available electronically from