Modeling and mapping permafrost active layer thickness using field measurements and remote sensing techniques

Abstract

Quantifying end of summer season Active Layer Thickness (ALT) of permafrost is critical for understanding the effects of climate warming, disturbance, and hydrologic changes on permafrost. Current research mainly focuses on ALT estimation and mapping at large scales using process-based or statistical-empirical models with biophysical variables as predictors. Here we modeled multi-year ALT field measurements between 2014 and 2019 at a site in Interior Alaska using 1-m hyperspectral imaging data and an object-based ensemble approach at a local scale (1 km2), examined the efficacy of the multispectral sensor WorldView (WV)-2 for ALT estimation, and explored the potential of integrating single-date imaging data with multi-year in-situ measurements for mapping the spatial and temporal variation of ALT. Modeling results showed hyperspectral imaging was accurate for estimating ALT with a correlation coefficient (r) larger than 0.7, while application of WV-2 data produced an r around 0.4. Reasonable ALT patterns were generated, and the spatial and temporal variation of ALT was delineated between the shallowest (2015) and deepest (2019) years using hyperspectral data. This study suggests hyperspectral imaging is a promising tool for predicting field ALT measurements and monitoring ALT change at local scales. We expect this study will stimulate hyperspectral optical sensors for permafrost studies in general, and particularly for ALT upscaling.

Keywords

• Permafrost active layer
• Hyperspectral imaging
• Machine learning modeling and mapping