Trends and causal structures of rain-on-snow flooding

Rain-on-Snow (ROS) events have been under increased scrutiny in recent years due to their devastating impacts. An ROS event is marked by rain falling on pre-existing snowpacks, which poses a considerable risk of flooding. In this study, we proposed a new approach to defining ROS events with potential flooding (ROS-PF) by establishing thresholds on rainfall, snow water equivalent, air temperature, and dew point temperature simultaneously, thereby overcoming the limitations of existing definitions. We also included a threshold at the 90th percentile over discharge to identify the ROS events that lead to actual floods (ROS-AF). Using this framework, we analyzed the frequency and trends of ROS-PF and ROS-AF events across thousands of basins in North America, Europe, Chile, Brazil, and Australia. Our findings indicate that the western US, central Chile, and central Europe are the most vulnerable regions with the highest frequency of ROS events, all of which showed a significant increasing trend. Additionally, we employed two causal discovery algorithms to uncover the causal structures leading to ROS flooding: Fast Causal Inference (FCI) and Fast Greedy Equivalence Search (FGES). Each algorithm offers a distinct path to infer causality from observational data. We combined the outputs of FCI and FGES to establish the final causal structure illustrating the causal mechanisms of ROS-based floods. This study also identified rainfall, soil moisture, snow water equivalent, maximum temperature, and DPT as critical drivers of ROS flooding, although the causal mechanisms resulting in ROS flooding differ across the four continents