Variability in Rain-on-Snow Event Detection: Discrepancies Across Climate Datasets
Schematic showing how the RoS event detection algorithm works. Basin pixels are isolated (top), followed by dataset-specific threshold definition (middle), then concurrent precipitation, snowmelt, and surface runoff are flagged (bottom).
Colin Zarzycki, Penn State
We developed an automated algorithm to detect rain-on-snow (RoS) events in gridded climate data, applied it to the Susquehanna River Basin (SRB), and compared four different historical datasets over two decades. The results show large discrepancies in RoS event frequency and magnitude across the datasets, primarily driven by differences in snowmelt and surface runoff estimates.
This research is important because RoS events contribute to slow-rise flooding, particularly in regions with ephemeral snowpacks, like the northeastern U.S. Understanding how these events are captured in climate datasets is critical for flood risk management, improving operational decision-making, and developing more reliable flood forecasting tools.
Rain-on-snow (RoS) events occur when rain falls on existing snowpacks, triggering rapid snowmelt that can lead to flooding. In this study, we developed an automated algorithm to identify RoS events in gridded climate data by detecting concurrent precipitation, runoff, and snowmelt thresholds over a defined catchment area. We applied this method to historical data over the Susquehanna River Basin (SRB) and compared the event frequency and magnitude across four widely used hydrometeorological datasets. Although the datasets cover the same 21-year period, the algorithm revealed significant differences in RoS event detection. For instance, when analyzing the January 1996 RoS event, which caused severe flooding in the SRB, datasets varied in how they simulated snowmelt and rain/snow partitioning, highlighting variability in data performance. Across the datasets, event frequencies differ by up to a factor of ten. Even when dataset-specific thresholds were applied, differences remained. This work shows that the underlying meteorological forcing and land surface processes are crucial to understanding and predicting these events. It also emphasizes the need for careful selection of climate datasets in operational contexts, particularly for flood risk mitigation in regions with compound hydrological extremes like RoS events.