Open Collections

A technique for sub-pixel temperature retrieval from multispectral imagery and its application to Quesnel Lake, British Columbia

University of British Columbia

Water temperature is a significant environmental indicator in lakes, rivers and streams. It is important to many resource management and environmental fields: forestry practices in adjacent lands modify runoff patterns which in-turn affects water temperatures; accelerated glacial melting caused by climate change increases the flux of cold water; and spawning salmon stocks are lethally affected by thermal shock from small (<5°C) temperatures changes. Quesnel Lake in the interior of British Columbia, Canada, is an important lake in all of these respects, and is the subject of this study due to recent mass mortality of spawning salmon believed to be caused by extreme temperature fluctuations. The conventional means of tracking temperature are in-situ temperature loggers installed in a small number of critical locations. This approach is costly and prone to data loss however, and does not provide a spatially comprehensive measure of lake temperature patterns. Coupled with satellite Thermal Infrared (TIR) imagery, the two measures can provide a contiguous dataset in space, depth, and time. Achieving an adequate balance between temporal and spatial coverage in TIR imagery can be difficult. Sensors with a sufficiently short revisit time often do not provide adequate spatial resolution to resolve narrow reaches of lakes and rivers. The MODIS sensor provides very high temporal coverage, up to four visits per day, but at a resolution of 1km, which is insufficient for Quesnel Lake's narrow reaches. This research presented in this thesis attempts to reconcile this deficiency in available TIR products, to produce an adequately accurate and fine spatial and temporal resolution water surface temperature dataset. By using a priori vectorized water boundary data, we estimate the sub-pixel surface temperature to within 1°C using a gradient descent solution to the mixed pixel at-sensor radiance equation. Ground-leaving radiance is estimated from standard MODIS temperature and emissivity data products for pure pixels and a simple regression technique to estimate atmospheric effects. The resulting algorithm is simple, effective, and useful for scientists without extensive remote sensing expertise, essentially unlocking a largely untapped resource for limnological and hydrological studies.

Text

2009-12-16

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/16758

Civil Engineering

University of British Columbia

UBCV

DSpace