Linking Atmospheric Transport Resistance to Extreme Snowfall Risk Using Non-Stationary Extreme Value Models

hiroshi matsuda , Shin'ichi HOMMA

This study develops a hybrid physical--statistical framework for evaluating extreme snowfall risk along the Sea of Japan coast by integrating causal-network analysis, Dijkstra-based atmospheric transport physics, and non-stationary extreme value theory. Using a 135-year record (1889--2024) of annual maximum snow-depth anomalies from 23 meteorological stations, we first quantify atmospheric moisture transport pathways through a meteorological transport-cost model based on Dijkstra's shortest-path algorithm, incorporating distance, topographic barriers, and wind-direction penalties. The transport-cost analysis supports a physically consistent four-block spatial classification (Hokkaido, Tohoku, Hokuriku/Niigata, and San-in/Kinki) and identifies Hokuriku/Niigata as a low-resistance transport hub. Non-stationary Generalized Extreme Value (GEV) models driven by a local winter monsoon index ($MOI_{\text{SN}}$) reveal strong regional contrasts in monsoon sensitivity. The Hokuriku/Niigata block exhibits the highest sensitivity ($\beta_2 = 12.067$), while station-specific Bayesian analyses identify localized hotspots at Shirakawa ($\beta_2 = 26.238$), Takada ($\beta_2 = 22.722$), and Nagaoka ($\beta_2 = 18.116$). A direct comparison between transport cost ($c_a$) and monsoon sensitivity ($\beta_2$) demonstrates a strong negative relationship at both the regional-block scale ($r = -0.95$) and the station scale ($r = -0.52$, $p = 0.011$), indicating that reduced atmospheric transport resistance is systematically associated with increased extreme-snowfall sensitivity. Although long-term climatological trends generally indicate declining snowfall baselines ($\beta_1 < 0$), positive shape parameters in Hokuriku/Niigata ($\xi = 0.166$) reveal heavy-tailed Fr{\'e}chet behavior and elevated susceptibility to rare high-impact events. These results demonstrate a decoupling between mean climatic trends and extreme-event risks: average snowfall may decline under warming conditions, while extreme snowfall hazards remain strongly amplified when intense winter monsoon surges propagate through low-resistance transport pathways. The proposed framework provides a physically interpretable approach for linking atmospheric transport processes with non-stationary extreme climate risks.