Remote Sensing Surface Temperatures of Forests and Melting Snow at Different Spatial Scales

Abstract

Uncooled thermal infrared (TIR) imagers, commonly used on aircraft and small unoccupied aerial systems (UAS, “drones”), can provide high-resolution surface temperature maps, but their accuracy is dependent on reliable calibration sources. A novel method for correcting surface temperature observations made by uncooled TIR imagers uses observations over melting snow, which provides a constant 0 °C reference temperature. This bias correction method is applied to remotely sensed surface temperature observations of forests and snow over two mountain study sites: Laret, Davos, Switzerland (27 March 2017) in the Alps, and Sagehen Creek, California, USA (21 April 2017) in the Sierra Nevada. Surface temperature retrieval errors that arise from temperature-induced instrument bias, differences in image resolution, retrieval of mixed pixels, and variable view angles were evaluated for these forest-snow scenes. Applying the snow-based bias correction decreased the root-mean-squared error by about 1 °C for retrieving snow, water, and forest canopy temperatures from airborne TIR observations. The degree to which mixed pixels change the observed surface temperature distributions across a forest-snow scene depends not only on image resolution, but also on how the underlying forest is distributed. Airborne observations over forests showed that near the edges of the TIR images, at more than 20° from nadir, the snow surface within forest gaps smaller than 10 m was obscured by the surrounding trees. These off-nadir views could then provide unmixed pixels of canopy surface temperature.