Exploring snow model parameter sensitivity using Sobol' variance decomposition

Abstract

This study advances model diagnostics for snowmelt-based hydrological systems using Sobol' sensitivity analysis, illuminating parameter sensitivities and contrasting model structural differences. We consider several distinct snow-dominated locations in the western United States, running both SNOW-17, a conceptual degree-day model, and the Variable Infiltration Capacity (VIC) snow model, a physically-based model. Model performance is rigorously evaluated through global sensitivity analysis and a temperature warming analysis is conducted to explore how model parameterizations affect portrayals of climate change. Both VIC and SNOW-17 produce comparable results with SNOW-17 performing slightly better for shallower snowpacks and VIC performing better for deeper snowpacks. However, the lack of sensitivity of SNOW-17 to climate warming suggests that it may not be as reliable as a more sensitive model like VIC. Inter-model differences presented here offer insights into physical features with greatest uncertainty and may inform future model development and planning activities.

Regional Snow Parameters Estimation for Large‐Domain Hydrological Applications in the Western United States

“…The energy exchange at the snow‐ground interface is usually small and thus neglected in modeling. Our analyses focused on five parameters identified in previous studies as important to modeling snowpack dynamics (Cristea et al, ; Houle et al, ; Koivusalo & Heikinheimo, ; Lynch et al, ; Malik et al, ; Pavelsky et al, ; Rajagopal & Harpold, ; Xue et al, ). The five parameters are fresh snow albedo ( α max ); the albedo decay coefficient during snow accumulation ( λ A ); the albedo decay coefficient during melt ( λ M ); and the precipitation partitioning threshold temperatures, including the minimum temperature for precipitation to be in the form of rain ( T R , in °C) and the maximum temperature for precipitation to be in the form of snow ( T S , in °C).…”

“…Recent studies (Chen et al, ; Essery et al, ; Etchevers et al, ; Feng et al, ; He et al, ; Houle et al, ; Koivusalo & Heikinheimo, ; Mendoza et al, ; Wayand et al, ) have explored the impact of model complexity on model performance and found that the model performance can be very dependent on model application. Feng et al () compared the performance of five physics‐based Land Surface Models (LSMs) with different physical complexities of snow processes on snow simulations at three experimental sites in north‐central Colorado.…”

The ecological stoichiometry and interrelationship between litter and soil under seasonal snowfall in Tianshan Mountain

“…, Houle et al. ). Due to variation in topography, vegetation cover, and wind strength, winter snow cover in alpine forests is often patchy, with areas of distinctly different snow depth.…”

“…However, an increasing number of studies have shown that litter decomposition occurs primarily in winter in forest ecosystems (Gavazov 2010, Atkins et al 2017. Authors have attributed this contrasting pattern to snow's insulating properties, which produces a relatively stable environment for decomposers, as well as strong nutrient leaching during snowpack formation and melting (Lemma et al 2007, Zeidler et al 2014, Houle et al 2017). Due to variation in topography, vegetation cover, and wind strength, winter snow cover in alpine forests is often patchy, with areas of distinctly different snow depth.…”

Making the most out of a hydrological model data set: Sensitivity analyses to open the model black‐box

“…The process has inherent uncertainty, and measures of uncertainty commonly accompany any modeling analysis [e.g., Pappenberger and Beven , ; Montanari , ; Beven , ; Nearing et al ., ]. Sensitivity analysis is conducted to understand the relation between inputs and outputs and to obtain insights in what often is a complicated model input‐output mapping [ Hill and Tiedeman , ; Saltelli et al ., ; Rosero et al ., ; Mendoza et al ., ; Norton , ; Hill et al ., ; Pianosi et al ., ; Razavi and Gupta , ; Markstrom et al ., ; Houle et al ., ]. These discoveries help the analyst to use simulated values appropriately in planning, risk assessment, and decision support.…”