Rain-on-snow (ROS) can lead to high flood risks due to potential runoff from both rainfall and snowmelt. While the historical devastating ROS flood events over the U.S. have been widely examined, their potential changes in a warmer climate are poorly understood. Here we first identify three historical ROS flood events with catastrophic impacts on humans and environment in the U.S.: 1) February 1996 Pacific Northwest flood; 2) January 1996 Mid-Atlantic flood; and 3) February 2017 Oroville flood. We then calibrate the runoff- and snow-related parameters of the state-of-the-art Energy Exascale Earth System Model (E3SM) Land Model (ELM) against the field measurements. The calibrated ELM model is used to hindcast the three ROS flood events at a high spatial resolution of 1 km and investigate their changes with an increase of air temperature from +1 ℃ to +5 ℃ but under similar meteorological conditions to the historical period. Specifically, we aim to uncover the sensitivity of runoff to increasing temperature for the three different events, in terms of the changes in magnitude, peak timing, spatial variations, and elevation gradients. We further quantify the snowmelt contribution to total runoff and identify dominant energy sources for snowmelt under different warming levels. This study will improve our understanding of future risks associated with the ROS flood extremes under global warming.