Multiscale Observational Platforms and Bayesian Data Integration to Estimate Snow Depth and Snow-water-equivalent over the Ice-wedge Polygonal Tundra

Abstract:

Snow has a profound impact on permafrost and ecosystem functioning in the Arctic tundra. Characterizing snow depths and snow-water-equivalent (SWE) is difficult over a large area, since they are quite heterogeneous, particularly in ice-wedge polygonal ground. In this study, we explore various strategies to characterize snow depths and SWE, using multiscale observational platforms, including a snow probe, ground penetrating radar (GPR), unmanned aerial system (UAS) and ground/airborne LiDAR. In addition, our statistical analysis and data locations are designed such that we can characterize the snow heterogeneity in multiple spatial scales. We demonstrate our approach using the datasets collected in the ice-wedge polygonal tundra near Barrow, AK.

We first document the characteristics of each platform. GPR can cover ~100s meters, and also provide depth-averaged snow density. The UAS-based snow-surface elevation is used to estimate the snow depth over a larger area with much less labor than GPR. Snow density analysis, using both cores and GPR, suggests that the depth-averaged snow density is quite uniform over the site.

The spatial variability of snow depths is then quantified, and correlated with various topographic measures (e.g., curvature) from the LiDAR digital elevation map as well as the wind direction. We explore different scales of moving average to separate micro and macro-topography. Results show that the wind factor is not significant in our case, except for at the edge of a drained thaw-lake basin. The wind distribution fills microtopographic lows (i.e., troughs and centers of low-centered polygons) and creates a smooth snow surface following the macro-topography, which depends on polygon sizes.

Finally, we develop a Bayesian geostatistical method to integrate these multiscale datasets for estimating snow depths and SWE over the study site (~1km by 1km). We compare two strategies: interpolating (1) the residual of the topographic correlations or (2) the snow surface. Results show that the former approach leads to more accurate estimates, suggesting the importance of considering macro and micro-topography to estimate snow depths and SWE. The estimated SWE is used in hydrological modeling, the results of which confirm the importance of considering the SWE heterogeneity for computing spring runoff.