Graduate Project
 

Applying Object-Based Supervised Image Classification in Large-Scale Floodplain Monitoring

Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_projects/kp78gq287

Descriptions

Attribute NameValues
Creator
Abstract
  • Large, alluvial rivers are naturally diverse, both in structural complexity and as drivers of landscape dynamics. Floodplains provide a mosaic of habitat types for aquatic, semi-aquatic, and terrestrial organisms and act as the framework for vital chemical processes to occur. In large part, this variety is due to the ability of rivers to shift and transport sediment within their floodplains, directly affecting species abundance, habitat availability, and physical landscape structure. Capturing this complexity can make floodplain monitoring difficult due to the variation in scale at which these processes occur. Remote sensing tools, effective in analyzing change at a variety of scales, are therefore often used in floodplain monitoring; however, such projects utilizing supervised image classification are still relatively few. To explore the validity and accuracy of applying this methodology in large-scale floodplain monitoring, the Upper Quinault River was assessed. High resolution, multi-spectral band imagery from 2006 – 2017 was classified using numerous parameters and classifier algorithms. Results were compared for accuracy using reference data created from randomly sampled points throughout the study area. With 95% confidence, area estimates for both vegetated surfaces and active channels had changed significantly in each of the later years (2013 – 2017) relative to the early years (2006 – 2011), with total area of vegetated surfaces increasing and total area of active channels decreasing. Overall map accuracies were relatively high, with all but two maps achieving an accuracy of 90% or greater. In 10 out of 12 reaches, the total area of change was higher for vegetating areas than areas undergoing erosion. An average of 8.86-ha of vegetated area was gained per reach while only an average of 1.74-ha was lost. The flow record was evaluated and no peak annual flows exceeded the 2-year flood level from 2008 to 2016, suggesting this trend may be influenced by a lack of channel-forming flows during the time series analyzed. Therefore, analyzing this data over a longer time period would aid the effort of determining the cause of these changes; however, the availability of high spatial, multi-spectral band data is limited in the historical record. Supervised image classification will become increasingly important for floodplain monitoring over time for its accuracy, efficiency, and consistency, particularly as the amount of available high resolution spatial data increases.
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Rights Statement
Publisher
Peer Reviewed
Language

Relationships

Parents:

This work has no parents.

Items

Thumbnail Title Date Uploaded Visibility Actions
file details Vito_CapstoneFinalReport.pdf 2021-12-22 Public Download