A statistical analysis of the relationship between surface observations and Snow-to-Liquid Ratio
Notebook Link:
https://colab.research.google.com/github/hunter3789/SLR-study/blob/main/SLR-study.ipynb
Intro:
- Accurate winter snowfall forecasting requires not only precise precipitation prediction but also consideration of the Snow-to-Liquid Ratio (SLR), which determines how much snowfall accumulates from a given amount of liquid precipitation.
- SLR varies depending on snow particle characteristics and compaction effects and is highly sensitive to subtle atmospheric conditions such as temperature changes, making it difficult to characterize accurately.
- Using snowfall observations from laser snow depth sensors (introduced by KMA in 2014; currently 643 sites), this study analyzes (1) the relationship between surface meteorological variables and SLR, and (2) regional differences in SLR characteristics.
Exploratory Data Analysis:
- Temperature, humidity, wet-bulb temperature, Snow Water Equivalent (SWE), and snow depth all show negative correlations with SLR.
- Higher temperature and humidity reduce SLR due to changes in snow crystal formation.
- Larger SWE and existing snow depth reduce SLR through compaction effects.
- Wind speed showed no significant correlation with SLR, while some previous studies (Ware et al., 2006) suggest wind may reduce SLR via snow crystal fracturing.
- Wet-bulb temperature had a slightly higher correlation with SLR than air temperature.
Statistical Significance – Linear Regression Model
- Linear regression was used to test statistical significance of predictors.
- Wet-bulb temperature, SWE (Snow Water Equivalent), and snow depth were all statistically significant (p < 0.001).
- Wet-bulb temperature was used as a representative variable to avoid multicollinearity.
- (discussion) Heteroskedasticity in the SLR regression is observed when SWE or existing snow depth are near zero.
Regional (Mechanism-Based) Differences in SLR
- West Coast Snowfall
- Caused by cold continental air interacting with relatively warm Yellow Sea waters
- Generally associated with low temperatures
- Warm Advection / Trough Type (Regions: Seoul metropolitan area)
- Occurs when warm air overrides cold air
- Often near the rain–snow temperature threshold
- Easterly Flow Snowfall (East Coast Type)
- Driven by easterly winds associated with high pressure over northeastern China
- Higher temperatures due to warm sea surface temperatures
Statistical test for regional effects - Linear Mixed Model
- Assuming common effects of wet-bulb temperature and SWE across regions with regional random effects
- A chi-square test using parametric bootstrapping (n = 1,000) showed that adding regional random effects improves model fitting statistically.
- However, (1) the SLR difference is small (approximately 1–2), and (2) random-effect variance is much smaller than mean squared error.
- Thus, regional effects are statistically significant but physically weak.