GEOSCAN Search Results: Fastlink


TitleMapping seasonal inundation frequency (1985-2016) along the St-John River, New Brunswick, Canada using the Landsat archive
AuthorOlthof, I
SourceRemote Sensing vol. 9, issue 2, 143, 2017 p. 1-16, Open Access logo Open Access
Alt SeriesNatural Resources Canada, Contribution Series 20190169
PublisherMDPI AG
Mediaon-line; digital
File formatpdf (Adobe® Reader®); html
ProgramRemote Sensing Science
Released2017 02 10
AbstractExtreme flood events in recent years in Canada have highlighted the need for historical information to better manage future flood risk. In this paper, a methodology to generate flood maps from Landsat to determine historical inundation frequency is presented for a region along the St-John River, New Brunswick, Canada that experiences annual springtime flooding from snowmelt and river ice. 1985-2016 Landsat data from the USGS archive were classified by combining See5 decision trees to map spectrally variable water due to spring ice and sediment, and image thresholding to map inundated floodplains. Multiple scenes representing each year were overlaid to produce seasonal time-series of spring (March-May) and summer (June-August) maximum annual water extents. Comparisons of annual surface water maps were conducted separately for each season against historical hydrometric water depth as a measure of relative springtime flood severity, and 1 m water masks from digital orthophotos were used to perform a formal accuracy assessment of summer water. Due to Landsat's 16-day revisit time, peak flood depth was poorly related to flood extent; however, spring depth measured during Landsat acquisitions was significantly related to extent (tau = 0.6, p-value < 0.001). Further, summer maps validated against 30 m water fractions scaled from 1 m water masks were over 97% accurate. Limitations with respect to the assessment of flood extent from depth, timing differences between peak flood depth and extent due to Landsat revisit time and cloud cover, and suggestions to overcome limitations through multi-sensor integration including radar are discussed.
Summary(Plain Language Summary, not published)
This paper presents a methodology and validation of historical flood (inundation) frequency from Landsat over the St-John River, NB. A new water extraction methodology is developed that is unique to springtime conditions in Canada when water colour is variable due to ice and sediment, and applied to a time-series of historical Landsat images from 1985-2016. Water extents from individual years are compared to water depth from hydrometric stations, revealing strengths and limitations of water extents and depths as unique measures of relative flood severity. Summer water maps are compared to high resolution digital orthophotos, confirming a high water mapping accuracy from Landsat. Historical Landsat water maps are then used to calculate the frequency of flooding for each location in the study area.

Date modified: