Case Study 3: River Ice Delineation with RADARSAT SAR

rsmnl
Page Navigation
140

the advantage of seeing through clouds and atmospheric haze. RADARSAT SAR (synthetic

aperture radar) collects spectral data in the microwave region and is capable of imaging

ground targets during adverse weather conditions, such as storms. Additionally,

RADARSAT SAR collects 10-m pixel sized data, a high spatial resolution well suited for

studies examining ice in narrow river channels. The study reviewed here explored

RADARSAT SAR's potential in delineating and monitoring ice and ice floes in rivers

ranging in stream widths of 160 to 1500 m. A better estimate of ice conditions along large

streams will allow for better navigation planning and will provide river dam regulators the

information needed to plan and prepare for ice breakup and floes.

b. Description of Methods. Three rivers of varying widths were evaluated for ice cover

over the course of two winters (2002 and 2003). The first winter was relatively mild with

partial river ice development at the three sites. Winter 2003 possessed a number of below

freezing days and was an ideal time for examining river ice in the northern mid-west. The

rivers chosen for this study were the Mississippi River near St. Louis, Missouri, the Mis-

souri River at Bismarck, North Dakota, and the Red Lake River in Grand Forks, North Da-

kota. Each site offered unique contributions to the study. The Mississippi River represented

a stream with heavy navigation use, the Missouri River site included a hydropower dam,

while the Red Lake River had extensive ice jam and flood records. Coordinated efforts

among CRREL researchers, the local Corps Districts, and the RADARSAT International

(RSI) aided in the acquisition and timing of satellite data collection.

(1) Stream channels were subset and isolated for river ice classification. To accom-

plish this, a band ratio was applied to Landsat TM data. They were then classified by an un-

supervised process and extracted for mask overlay onto the radar data. This sufficiently out-

lined the land/water boundaries and isolated the stream in images with wide river channels.

This process omitted vegetation and islands from the resultant image. The subsequent SAR

subset did not include mixed pixels (land/water/ice).

(2) Images with narrow channels required hand-digitization and a textural analysis,

followed by a supervised classification (to further eliminate land pixels). The hand-digitiza-

tion proved less successful than the Landsat TM overlay and extraction method. Hand-dig-

itization did not thoroughly omit pixels with mixed water, vegetation, and land (i.e., river

islands).

(3) In the SAR images, only the channel reaches were analyzed for ice conditions us-

ing an unsupervised classification. The classification mapped brash ice (accumulated float-

ing ice fragments), river channel sheet ice, shore ice, and open water.

6-8

Index