GROUND-TRUTHING A REMOTE SENSING LANDSLIDE MODEL USING REACTION WOOD PRESENCE IN TREE RINGS NEAR GLENNALLEN, ALASKA

Alaska’s interior transportation corridors are susceptible to landslide events and are often burdened by excessive repair costs and prolonged repair time. The state requires a new model that can project long-term landscape stability given the constraints of a limited landslide inventory. Miandad et al. (2020) developed a remote sensing model using LiDAR and Normalized Difference Vegetation Index (NDVI) to identify stable slopes, landslide susceptible slopes, and landslide slopes across Alaska with tests at four study sites. We ground-truthed an identified landslide area between the Trans-Alaskan pipeline and the Richardson Highway south of Glennallen, AK; the area is characterized topographically by a steep slope and several areas of exposed ground cover. We sampled 29 black spruce (Picea mariana) that were preferentially tilted by obtaining core samples (n=58), which were taken from each tree—one from the tree’s direction of tilt and the other perpendicular to the first. Samples were dot-counted and statistically verified using a digitized measuring system that generated a master chronology of individual growth years spanning from 1821-2021 (correlation, 0.63). With a focus on the tree-tilt cores (n=29) and data prior to 1921 excluded from analysis due to a small sample size, the mean value of recorded reaction wood is 16.2% for the 100 years (1921-2021) with a maximum of 47.6% in 1938 and a minimum of 3.4% in 2014. Exceptionally, 1930-1946 saw the largest increase in reaction wood with an average of 31.2%, which likely indicates slope instability. A 5.9% average increase, above background, in reaction wood from 2003 to 2007 potentially implies a recent slope instability. By extension, a site event must have occurred in 2021-2022, for all trees were tilted (mean=11.8°) across a disturbed slope revealing tensional cracks, mud slurries, and extensionally split trees. Given the LiDAR data was collected in 2011, it is unclear whether the model picked up on background instability or peak instability from 1930-1946 or 2003-2007 and from what factors of slope, curvature, roughness or NDVI determined this area as a landslide. Incorporating yearly rates of NDVI and LiDAR model parameters, when available, into the model could give greater insight into future stability based on the fluctuation of data values.