An Ecosystem Mapping and Modeling Unit Approach for Snow Scaling
Authors
Eve Gasarch1* (egasarch@lanl.gov), Emma Lathrop1, Lauren Thomas1, Claire Bachand1, Shannon Dillard1, Katrina E. Bennett1, Colleen Iversen2
Institutions
1Earth and Environmental Sciences Division, Los Alamos National Laboratory, Los Alamos, NM; 2Oak Ridge National Laboratory, Oak Ridge, TN
URLs
Abstract
Scaling is a ubiquitous challenge in Earth system models (ESMs). Clustering methods applied in permafrost studies are applied here as a tool to upscale snow characteristics from watershed (~km scale) to regional scales (~10s of kms), in a construct the research team is referring to as ecosystem mapping and modeling units (EMMUs). The resulting products can be used to evaluate coarse-scale ESM results, and to improve the ESM to reflect subgrid cell variability in snow characteristics.
In this work, researchers focus on two basins on the Seward Peninsula in Alaska, drawing on a combination of field-based and remotely sensed observations collected over winter and spring in 2021 and 2022. Estimates of snow cover and duration as well as snow insulation are derived for approximately 275 points across the basins using measurements obtained from iButton temperature sensors placed at ground surface as well as air temperatures collected at meteorological stations at each basin. Peak snow depth within the basins was measured manually at approximately 7000 points in late March 2022.
EMMUs are derived using a k-prototype clustering method, selected for its ability to handle mixed data types. Features included in the EMMU construct are derived from Light Detection and Ranging (LiDAR) data collected in summer 2022, satellite imagery, and from vegetation maps developed for the area. Several cluster variations were developed using different selections of input features and evaluated for their relationship to snow characteristics. Features considered include topographic characteristics of slope, elevation, aspect, and curvature; as well as vegetative characteristics of height and density, shrub presence or absence, and plant community type.
While traditionally ecosystem types focus on plant communities, the research team’s initial results indicate a clustering approach that includes additional features is a stronger predictor of snow characteristics than vegetation type alone. Additionally, the predictive capabilities of EMMUs regarding snow characteristics (e.g., depth, duration, insulation) are compared with those of a machine learning (random forest) model.
Leveraging the relationship between EMMUs and snow characteristics may represent a simpler alternative to more computationally intensive methods of scaling snow from watershed-to-regional scales and potentially assist with informing ELM subgrid variability.