Scaling Remotely Sensed Surface Temperatures of Forests and Melting Snow

In heterogeneous mountain and forest terrain, where in situ measurements of seasonal snow are sparse, thermal infrared (TIR) remote sensing from aerial or spaceborne platforms can give us spatially distributed measurements of surface temperatures. However, even very high resolution remote sensing data provides us with the surface temperatures of mixed pixels, where features of interest such as forest canopy and the snow surface are blurred together. Understanding the role that TIR image scale and resolution play in how surface temperatures are represented in remote sensing observations can help improve how these data are collected and applied to snow hydrology and mountain ecology research.

In this study, surface temperature observations were made over two forested mountain snowpack sites: Laret, Davos, Switzerland (March 27, 2017) in the Alps, and Sagehen Creek, California, USA (April 21, 2017) in the Sierra Nevada. In situ point temperature data from fixed and handheld radiometers provided surface temperature ground-truth for comparison with airborne remote sensing. TIR imagery was collected by small unoccupied aerial systems (UAS) at each site, from altitudes between 20-110 m above ground level (AGL) (image resolutions 0.03-0.1 m), and at Sagehen by an aircraft from 1000 m AGL (image resolutions 1.0 m).

Key findings of this study show that as the pixel size of TIR observations increased (i.e., image resolutions decreased) (1) the standard deviation of measured temperatures decreased, (2) the median forest canopy temperature decreased, (3) the median snow surface temperature increased, and (4) forest edges and solitary trees appeared to have the greatest decrease in median temperature relative to denser forest stands. These findings can be attributed to coarser resolution pixels sensing a mixture of both forest canopy and snow surface radiance, blurring together fine-scale surface temperature variations. The results of this work can help inform how TIR remote sensing should be interpreted over heterogeneous environments, and how image resolution may bias observations.