Detecting avalanche path ground cover and vegetation change across multiple scales through time using remote sensing tools

Abstract

Large-magnitude avalanches often alter vegetation composition, avalanche path dimensions, and subsequent avalanche return periods. Understanding temporal changes in individual avalanche path trimlines, runout zones, and geomorphic characteristics helps forecasters, planners, and engineers estimate potential avalanche destructive size and impact on infrastructure or settlements in the runout zone. Understanding these changes on a large scale also provides information on post-cycle avalanche distribution. Here, we use remote sensing platforms and change detection techniques to examine vegetation change in avalanche paths in Montana and Colorado. In northwest Montana, we implemented a novel approach using lidar, aerial imagery, and a random forest model to classify imagery-observed vegetation within avalanche paths. We calculated spatially explicit avalanche return periods using a physically based spatial interpolation method and characterized the vegetation within those return period zones. In Colorado, we investigated changes in avalanche path vegetation characteristics prior to and after a widespread large-magnitude avalanche cycle. The highest frequency of avalanche return periods was broadly characterized by grassland and shrubland, but topography greatly influences vegetation classes and return periods. Furthermore, statistically significant differences in lidar-derived vegetation canopy height exist between categorical return periods. We used optical sensors from satellite imagery to analyze changes in Normalized Difference Vegetation Index (NDVI) to calculate ground cover change over time. NDVI, a measure of near-infrared and red bands within the imagery, allowed us to distinguish between green vegetation (e.g., trees and shrubs) and non-vegetated ground cover (e.g., dead and downed trees, rocks, and dirt) within avalanche paths. For this study, we calculated changes in NDVI values by comparing imagery from 2018 to imagery from 2019 after a widespread large magnitude avalanche cycle occurred in March 2019 in Colorado, United States. We applied a filtering process to reduce error, classified NDVI change based on the value distribution, and then calculated area change of all areas within each avalanche path. We completed this process for 1633 avalanche paths throughout Colorado. We found that using NDVI difference values pre- and post-avalanche cycle allowed us to identify ground cover change in avalanche paths throughout Colorado. These changes span from a slight expansion of existing avalanche paths to substantial landscape disturbance. For example, a size D5 avalanche caused severe ground cover change in 18% of one single path near Aspen, Colorado. This suggests that large magnitude avalanches can redefine avalanche path dimensions and could impact subsequent avalanche size and frequency. Using NDVI from satellite imagery is a simple way to detect ground cover changes in avalanche paths on a large scale or in remote areas. In general, remote sensing products to detect and examine vegetation and ground cover change in avalanche paths can help inform avalanche distribution and benefit planning efforts.