Maps of Active Layer Thickness in Northern Alaska via Upscaling of P-band SAR Retrievals

Abstract:

Detailed information on the spatial and temporal distribution of active layer thickness (ALT) in the permafrost regions of northern Alaska can be of great value, offering broad visibility into how climate change has been impacting northern landscapes. It also facilitates the estimation of greenhouse gas emissions resulting from permafrost degradation, a significant component in global carbon budgets. The vastness and inaccessibility of the region has, however, limited measurements of ALT to scattered small test sites. Efforts to map ALT for the region have, consequently, thus far relied on either upscaling of sparse point measurements or low resolution satellite measurements.

To bridge this gap in the extent and resolution of estimated ALT, we are constructing high-resolution (30 m) maps of ALT in northern Alaska. Each map is formed by extrapolating from narrow strips of estimated ALT derived from airborne polarimetric P-band synthetic aperture radar (SAR) imagery, a data source far more extensive and representative than in-situ test sites. These data were acquired as part of the NASA Arctic-Boreal Vulnerability Experiment (ABoVE) flights in Alaska in 2017, as well as pre-ABoVE flights in 2014 and 2015. P-band SAR signals are able to penetrate both vegetation and topsoil to sense the depth to the top of the permafrost. Radar scattering inversion processing applied to the SAR imagery yields strips of high resolution estimated ALT.

Spatial upscaling from the SAR-derived strips is achieved by applying the Random Forests ensemble decision tree algorithm to perform a nonlinear regression. Training data for the regression are obtained by means of a dense stratified sampling of the SAR-derived strips that captures the full range of ALT values. Independent variables for the regression are taken from readily-available spatial data sources. These have been chosen as exhibiting a significant correlation with ALT, and include land cover, topography, soil texture, bulk density, and organic carbon, thaw and freeze degree days for August and October, and proximity to water. The selection of input data layers will be refined using recursive feature elimination. Finally, thousands of SAR-derived pixels withheld from the regression training set are used to validate the accuracy of the resulting large-scale high resolution ALT map.