Causal Network Analysis of Heavy Snowfall Systems: North American Great Lakes vs. Sea of Japan Coast

hiroshi matsuda , Shin'ichi HOMMA

This study conducts a hydro-climatic comparative analysis of two regions prone to heavy urban snowfall: the eastern shores of the North American Great Lakes (Lake-Effect Snow: LES) and the coast of the Sea of Japan (Sea-Effect Snow: SES). We introduce a quantitative metric—water vapor conversion efficiency (η= snowfall amount /PW)—and apply a data-driven causal discovery framework (DirectLiNGAM). Building upon challenges identified in previous studies (e.g., Fujisaki-Manome et al., 2022; West et al., 2019; Steenburgh and Nakai, 2020), we present the following five new findings:
(1)Mathematical modeling of a discontinuous "risk jump" (Jrisk) using the Heaviside step functions.
(2)Construction of causal Directed Acyclic Graphs (DAGs) that distinguish between physical topological differences.
(3)Characterization of properties through non-stationary Generalized Extreme Value (GEV) distribution analysis spanning over 80 years (confirming a Weibull-type distribution with an upper bound (ξ= -0.12 < 0) for the Great Lakes and a heavy-tailed Fréchet-type distribution (ξ= +0.18 > 0) for Japan).
(4)Identification of a memory effect regarding water vapor accumulation over 1–2 days (lag 1–2: r = 0.78).
(5)Quantification of differences in amplification effects attributable to topographic conditions, Japan's steep mountain ranges (elevations of 2,000–3,000 m) versus the US's gentle plateaus (elevations of 300–500 m).