Quantifying snow depth fluctuations based on a data-driven approach: Case study in Japan

The purpose of this study is to quantify snow depth fluctuations over a multi-year period using a data-driven approach. We analyze snow depth data from November 1 to April 30 over the period 1965–2024 in Nagaoka, a populous city in Japan that experiences heavy snowfall. This paper investigates two phenomena predicted by snow scientists: a decrease in the total snow depth throughout the year and an increase in heavy snowfall for a short period in early winter. Using an interpretable basis decomposition method, we represent the snow depth for each year as a nonnegative weighted sum of basis elements corresponding to the middle of February, late March, and early January. This enables us to prove a decrease in the total snow depth throughout the year. We show that the snow depth in late March has decreased over time, indicating a relationship with rising temperatures. In contrast, the proportion of the total snow depth reached in early January has increased in recent years. This may be related to extreme weather, including an increase in heavy snowfall for a short period in early winter. In this way, our method provides quantitative evidence that recent snowfall patterns deviate from historical trends. Our data-driven approach has the versatility and applicability to a variety of studies. • The purpose is to quantify snow depth fluctuations over a multi-year period. • We exploit a data-driven approach, which can be used without expert knowledge. • Our approach provides features corresponding to promising indicators in snow science. • It also enables more detailed analysis than conventional annual aggregation methods. • Our approach is versatile, and the effectiveness is demonstrated for Nagaoka, Japan.