Assessing the accuracy of Landsat-derived stream temperature for use in juvenile salmonid habitat assessments on the Anchor River, Alaska
This thesis investigates the environmental variables which influence the accuracy of Landsat 8 thermal infrared imagery-derived stream temperatures in a ground-truthing study on the Anchor River basin in Southcentral Alaska. From May 1st to September 30th 2015, in-situ temperature data were collected concurrently with the remotely sensed Landsat 8 TIR images at ten field sites throughout the Anchor River watershed. At each field site, stream delineations were preformed to assess the influence of riparian vegetation, stream morphology, discharge and air temperature on TIR-derived stream temperatures. The accuracy of TIR data was assessed by calculating the Landsat Thermal Offset (LTO), the difference between TIR-derived temperature and the in-situ temperature (TIR – In-situ =LTO). A Generalized Additive Model was employed to examine the relationship between environmental variables and LTO, and model selection with Akaike information criterion (AIC) determined that Rosgen Stream Classification (Rosclass), 7 day trends in stream discharge (Gagediff), and daily average air temperature (Airdaily) were strongly associated with Landsat 8 thermal offset (LTO). Stability in stream discharge (Gagediff = ±0.2 ft.) in the week preceding a Landsat TIR data collection event, coupled with stream morphology characteristics consistent with Rosclass B and C reaches, were associated with the highest accuracy of raw Landsat TIR-derived stream temperatures. Across all sites and dates, GAM regressions accurately translated raw Landsat TIR data into temperatures reflecting in-situ water temperatures in small order streams. While regression analyses supported Rosclass, Gagediff, and Airdaily environmental covariates as being influential in explaining LTO, the underlying mechanisms which these variables reflect remains an open question. These variables may provide indication of conditions that lead to stream stratification, or alternatively, may be reflecting other environmental processes leading to stronger temperature differentials between temperature in the stream and that on the surrounding riparian land. Continued validation efforts in other high latitude watersheds over longer time periods will be important for testing the robustness of methods to utilize Landsat TIR-based stream temperature predictions, progressing towards a goal of providing a broad-scale tool for fisheries and watershed managers to assess the degree of and potential impacts from climate change on Alaskan freshwater aquatic and fishery resources.
Funding for this work was provided by the Pollock Conservation Cooperative via the Alaska Education Tax Credit Program.
Current Status: Biologist – Tyonek Tribal Conservation District (Anchorage, AK)