Ian Brown

Abstract:
Snowcloud is an embedded data collection system for snow hydrology field research campaigns conducted in harsh climates and remote areas. The system combines distributed wireless sensor network technology and computational techniques to provide data at lower cost and higher spatio-temporal resolution than ground-based systems using traditional methods. Snowcloud has seen multiple Winter deployments in settings ranging from high desert to arctic, resulting in over a dozen node-years of practical experience. The Snowcloud system architecture consists of multiple TinyOS mesh-networked sensor stations collecting environmental data above and, in some deployments, below the snowpack. Monitored data modalities include snow depth, ground and air temperature, PAR and leaf-area index (LAI), and soil moisture. To enable power cycling and control of multiple sensors a custom power and sensor conditioning board was developed. The electronics and structural systems for individual stations have been designed and tested (in the lab and in situ) for ease of assembly and robustness to harsh winter conditions. Battery systems and solar chargers enable seasonal operation even under low/no light arctic conditions. Station costs range between 500 and 1000 depending on the instrumentation suite. For remote field locations, a custom designed hand-held device and data retrieval protocol serves as the primary data collection method. We are also developing and testing a Gateway device that will report data in near-real-time (NRT) over a cellular connection. Data is made available to users via web interfaces that also provide basic data analysis and visualization tools. For applications to snow hydrology studies, the better spatiotemporal resolution of snowpack data provided by Snowcloud is beneficial in several aspects. It provides insight into snowpack evolution, and allows us to investigate differences across different spatial and temporal scales in deployment areas. It enables the relation of local variations in accumulation to environmental parameters such as slope, micro-topography and vegetation cover. The relative importance of such factors is likely to change over the snow season, for example as deposition fills depressions reducing surface roughness due to micro-topography and low lying vegetation. Snowcloud has also been deployed to support the acquisition of satellite imagery from Radarsat-2 and TanDEM-X missions. In capturing spatial variability Snowcloud provides better validation data than single node systems. An important use case of Snowcloud is its application to modeling snow-water equivalent SWE evolution, and more accurate areal average SWE estimation in locations with a typical paucity of snowpack data. We are especially interested in applying machine learning techniques to find relations between average SWE in remote areas and regional infrastructure data (e.g. regional snow pillows). These techniques leverage a combination of data provided by Snowcloud systems and periodic manual snow courses in short field campaigns. Furthermore, using machine learning we can explore the importance of snowpack variables for synthetic aperture radar backscatter at co- and cross-polarizations to better understand scattering processes and unify models of scattering with in situ observations at high resolution.