Machine Learning-Based Downscaling of Glacier Surface Temperature to 10-m Resolution Using NDSI-Driven Emissivity Correction

Accurate high-resolution glacier surface temperature (GST) mapping is critical for monitoring Himalayan glacier dynamics. This study introduces a method to predict GST at 10-m resolution by integrating Sentinel-2, Landsat thermal infrared sensor (TIRS), and shuttle radar topography mission (SRTM) digital elevation model (DEM) data in Google Earth engine (GEE). A dynamic emissivity correction, using normalized difference snow index (NDSI) thresholds (clean ice: NDSI <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\geq 0.4$ </tex-math></inline-formula>, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\varepsilon =0.99$ </tex-math></inline-formula>; debris-covered ice: <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\varepsilon =0.96$ </tex-math></inline-formula>), addresses a critical limitation in existing GST retrievals. A random forest (RF) regression model trained on stratified elevation classes achieved an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${R} ^{2}$ </tex-math></inline-formula> of 0.91 and mean absolute error (MAE) of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.53~^{\circ }$ </tex-math></inline-formula>C, downscaling Landsat’s 30-m thermal data. Results reveal fine-scale GST variability driven by debris cover and topography, demonstrating the method’s utility for glacier studies.