In many areas of the circumpolar north, it is often
prohibitively expensive to use aircraft or ground
reconnaissance techniques to map the landscape due to the remoteness
of the terrain and large extent of areas to be mapped.
Satellite remote sensing offers a more cost-effective method for
map- ping and monitoring large areas in remote regions. A
single satellite scene can cover an area that would require
hundreds of medium-scale aerial photographs. Furthermore,
the regular orbit cycle of satellites makes long-term
There are two basic types of satellite sensors: active microwave sensors and passive optical sensors. These two types gather complementary data for mapping wetlands. Active microwave sensors, like Earth Resources Satellite-1 synthetic aperture radar (ERS-1 SAR), emit energy toward the earth which scatters off the earth's surface. Some of the scattered energy returns to be detected by the sensor and is called backscatter. Since geometric characteristics and the presence or absence of water on the landscape influence backscatter, wetland properties such as spacing and height of plants, plant shape, and flooding conditions may be detected by SAR remote sensing Passive optical sensors, like Landsat Thematic Mapper (TM), primarily sense sun- light reflected from the earth, and may be influenced by shadow between plants, pigment absorption, and plant cellular structure. Thus, Landsat TM and ERS-1 SAR gather different information and are sensitive to different parameters on the landscape.
The overall goal of this study was to evaluate the capa- bilities of combined imagery from the ERS-1 SAR and Landsat TM sensors for classifying boreal wetlands and to compare the results with classifications using the two sen- sors individually. The general hypothesis was that strengths of ERS-1 SAR (sensitivity to water, reliable multitemporal coverage regardless of cloud cover, and sensitivity to land- scape texture) will complement strengths of Landsat TM (discrete spectral bands and sensitivity to several vegetational factors) to improve detail and accuracy.
I used Landsat TM and multitemporal ERS-1 SAR to classify wetlands in the Tanana Flats area, interior Alaska. Thematic Mapper bands 1 through 5 from a single image and 23 separate radar images were co-registered and clipped to create a combined/multitemporal stack of grids. A maximum likelihood classifier was then used to classify (1) single-data, single-band radar images, (2) multitemporal radar images, (3) multispectral TM image, and (4) combined multitemporal radar and multispectral TM imagery. Overall classification accuracy for the multitemporal radar images was significantly higher than any single-data radar classification. The TM classification separated classes well within the wetlands and non-wetlands categories. However, the TM classification often confused classes between wetland and non-wetland categories. Overall accuracy for the combined multitemporal radar and multispectral TM imagery was sig- nificantly higher than either the multitemporal radar or the multispectral TM classification. The combination retained the strengths of each sensor: radar's wetland versus non-wetland delineation, and TM's relatively accurate classification within the wetland and non-wetland categories.
Return to global glimpses
The University of Alaska Fairbanks is an AA/EO employer and educational institution and prohibits illegal discrimination against any individual: Learn more about UA's notice of nondiscrimination.