Predicting/Extrapolating Active Layer Thickness Using Statistical Learning from Remotely-Sensed High-resolution Data in Arctic Permafrost Landscapes: Improved parameterization of Ice-wedge polygons from LiDAR/WorldView-2 derived metrics

Wednesday, 17 December 2014
Chandana Gangodagamage1, Joel C Rowland1, Susan S. Hubbard2, Steven P. Brumby1, Anna Liljedahl3, Haruko Murakami Wainwright2, Victoria L Sloan4, Garrett Altmann1, Alexei N Skurikhin1, Eitan Shelef5, Cathy Jean Wilson6, Baptiste Dafflon2, John Peterson2, Craig Ulrich2, Ann Gibbs7, Craig E Tweedie8, Scott L Painter1 and Stan D Wullschleger9, (1)Los Alamos National Laboratory, Los Alamos, NM, United States, (2)Lawrence Berkeley National Laboratory, Berkeley, CA, United States, (3)University of Alaska Fairbanks, Fairbanks, AK, United States, (4)ORNL, Bristol, United Kingdom, (5)Stanford University, Stanford, CA, United States, (6)Los Alamos National Lab, Los Alamos, NM, United States, (7)USGS California Water Science Center Menlo Park, Menlo Park, CA, United States, (8)University of Texas at El Paso, El Paso, TX, United States, (9)Oak Ridge National Laboratory, Oak Ridge, TN, United States
Landscape attributes that vary with micro-topography, such as active layer thickness (ALT) in ice-wedge polygon ground, are labor-intensive to document in the field at large spatial extents, necessitating remotely sensed methods. Robust techniques to estimate ALT over large areas would improve understanding of coupled dynamics between permafrost, hydrology and landsurface processes, and improve simulations of the rate and timing of release of soil carbon from permafrost settings. In particular, it would provide critically needed data to parameterize and initialize soil property information in permafrost models and evaluate model predictions for large, complex domains. In this work, we demonstrate a new data fusion approach using high-resolution remotely sensed data for estimating cm scale ALT in a 5 km2 area of ice-wedge polygon terrain in Barrow, Alaska. We used topographic (directed distance, slope, wavelet-curvature) and spectral (NDVI) metrics derived from multisensor data obtained from LiDAR and WorldView-2 platforms to develop a simple data fusion algorithm using statistical machine learning. This algorithm was used to estimate ALT (2 m spatial resolution) across the study area. A comparison of the estimates with ground-based measurements documented the accuracy (±4.4 cm, r2=0.76) of the approach. Our findings suggest that the broad climatic variability associated with warming air temperature will govern the regional averages of ALT, but the smaller-scale variability could be controlled by local eco-hydro-geomorphic variables. This work demonstrates a path forward for mapping subsurface properties over large areas from readily available remote sensing data.

Methodology of Mapping and Characterization Polygons:

We convolve LiDAR elevations with multiscale wavelets and objectively chose appropriate scales to map interconnected troughs of high- and low-centered polygons. For the ice wedges where LiDAR surface expressions (troughs) are not well developed, we used a Delaunay triangulation to connect the ice-wedge network and map the topologically connected polygons.

Polygon slopes and curvatures as a function of DD were used to develop a microtopographic classification scheme for rims/elevated ridges (Zone 1), centers (Zone 2), and troughs (Zone 3) in both high- and low-centered polygon.