Snow-fed mountain watersheds supply freshwater to ∼2 billion people worldwide (Viviroli et al., 2007; Sturm et al., 2017; Immerzeel et al., 2020). However, climate change is endangering this natural service. Declines in winter snowfall and snow albedo are reducing snow’s natural reservoir function by producing both smaller and earlier snowmelt (Dudley et al., 2017; Huss et al., 2017; Mote et al., 2018; Skiles et al., 2018; Lynn et al., 2020; Kraaijenbrink et al., 2021; Siirila-Woodburn et al., 2021). Winter rain-on-snow (ROS), particularly on high elevation alpine snow (Freudiger et al., 2014; Beniston and Stoffel, 2016; Musselman et al., 2018; Hock et al., 2019), further diminishes the snowpack’s ability to accumulate, and simultaneously escalates the risk of extreme hazards including ROS floods, ROS initiated avalanches, and debris flows (Harr, 1981; DeGraff, 1994; Stimberis and Rubin, 2011; Surfleet and Tullos, 2013; Hatchett et al., 2020).

What is ROS? The simplest definition is: measurable rainfall on an existing snow cover (Pomeroy et al., 2016). However, observing and/or measuring the process is easier said than done (Harpold et al., 2017), which gives rise to numerous “definitions”. For example, McCabe et al. (2007) defined ROS as any day in which precipitation occurs and snow depth declines—a good definition given available instrumentation, but the metric fails to quantify flood risk. Alternatively, Musselman et al. (2018) and Huang et al. (2022) defined ROS (with flood potential) as heavy rainfall ≥ 10   m m   d − 1 ,     or ≥ 25   m m   d − 1 , respectively, that falls on a snow water equivalent (SWE) ≥ 10   m m and where the snowmelt contribution to Terrestrial Water Input (TWI; integrated rainfall and snowmelt reaching the land surface over the event duration) exceeds 20%, or 25%, respectively. However, as Wayand et al. (2015) noted, ROS flooding can still occur with minimal snowmelt contributions, and therefore depending on the ROS definition, could lead to an undercount of rainfall-driven, but nonetheless still technically ROS-type floods.

Definitions aside, rain falling on a snowpack tends to occur between a few to ∼10 times a season (Moore and Prowse, 1988; Cohen et al., 2015; Würzer et al., 2016; Li et al., 2019; Juras et al., 2021; López-Moreno et al., 2021), and is more frequent in maritime mid-latitude mountains vs. continental, or arctic locations (López-Moreno et al., 2021). Most ROS events do not cause flooding (i.e., Merz and Blöschl, 2003; Yang et al., 2020). But, when floods do arise—they can be historic (e.g., Kattelmann, 1997; Marks et al., 1998; Rössler et al., 2014; Garvelmann et al., 2015; Pomeroy et al., 2016).

This review covers the physical processes governing ROS for midlatitude mountain snowpacks and touches on flood-generating mechanisms. Conceptually, ROS floods exist on a continuum—on one end, rainfall on a snow-dominated landscape producing TWI, and on the other, TWI resulting from both rainfall and snowmelt. Improvements to ROS flood forecasting skill is predicated on knowing the contributions of both fluxes as they impact the timing, duration, and volume of runoff (Tarasova et al., 2019). To help process-based discussion we introduce two new terms— “passive” vs. “active” snowpacks. Passive snowpacks convey rainfall but provide minimal melt, whereas active snowpacks readily melt, enhancing TWI.

Many past ROS floods have been largely rainfall driven (Kattelmann, 1997; Merz and Blöschl, 2003; Wayand et al., 2015; Li et al., 2019). A good example of a more “extreme” snowmelt contribution would be Henn et al. (2020) who found that snowmelt enhanced TWI by 37% in California’s Feather River. But, broadly speaking, what drives a large ROS runoff response?

Simply put, a large ROS runoff response requires an anomalously low snowline, a large snow-covered area (SCA) and prolonged, high-intensity rainfall over most of the SCA. A seasonally low snowline can result from either a preceding cold storm, or within a single storm (Hatchett et al., 2016)—but both scenarios lead to a large SCA. Prolonged, high intensity rainfall can occur under a range of atmospheric conditions. These include: atmospheric rivers (ARs; e.g., Guan et al., 2016; Ralph et al., 2020), cloud microphysical processes such as seeder-feeder mechanisms and localized convection (e.g., Rössler et al. (2014)), enhanced mid-tropospheric moisture fluxes (e.g., Kaplan et al. (2012)), and orographic effects (Houze, 2012). We note that these atmospheric conditions operate in unison or independently, but their identification is fundamental to understanding the spatiotemporal variability in the basin-wide runoff response.

While atmospheric conditions favor prolonged, high-intensity precipitation for a large runoff response, surface processes can either amplify or dampen the runoff. Primary factors include: the snow energy balance, preferential flow, and the antecedent snowpack and soil conditions (Figure 1; Dunne and Black, 1971; Berris and Harr, 1987; Kattelmann, 1997; Singh et al., 1997; Marks et al., 1998; Kattelmann and Dozier, 1999; Rössler et al., 2014; Guan et al., 2016). An active snowpack driven by a large positive energy balance can bolster rainfall driven TWI—particularly across ephemeral (i.e., transient), low-elevation snow with little cold content (Jennings et al., 2018). However, without a saturated landscape and a snowpack with established preferential flow routing (see Section 4), runoff responses could still be dampened and/or delayed (Kattelmann, 1989; McGurk and Marsh, 1995; Singh et al., 1997; Garvelmann et al., 2015; Würzer et al., 2016).

Many midlatitude snow-dominated basins have extensive forest coverage, and vegetation structure/density are known to influence snow accumulation via interception, reduced wind effects, and an altered energy balance (Storck et al., 2002; Lundquist et al., 2013; Broxton et al., 2015; Stevens, 2017). Given that ROS involves some of the same processes, it would follow that vegetation should also exert control on resulting ROS runoff. However, results from field studies remain inconclusive. For example, some studies suggest that ROS produces more outflow from glades (forest open areas) compared to forests (Beaudry and Golding, 1983; Berris and Harr, 1987; Storck et al., 2002), while others observed little difference (Kattelmann, 1987; Berg et al., 1991; Garvelmann et al., 2015). A good discussion can be found in Garvelmann et al. (2015)—but, more research is needed.

While our binary classification (passive vs. active) is by design simple, we note that snowpacks exist on a spectrum that can vary both temporally and spatially (Figure 2). For example, at a point location, antecedent snowpack conditions (e.g., the pre-event cold content) coupled with the meteorological conditions will initially drive a singular response (passive or active), however a shift in the meteorological drivers (i.e., a change in weather fronts) could elicit a switch. From a basin perspective, snowpacks are likely composed of a transition from active to passive snowpacks because of elevation-driven temperature differences—particularly when basins encompass large topographic relief. Forest coverage and/or structure will only increase the variability in the snowpack response. Regardless, both snowpack modes (passive vs. active) can dampen and/or delay TWI through an unsatisfied irreducible water content, but only active snowpacks will amplify TWI.

The rate of liquid water transmission through snow governs outflow timing and magnitude. Despite being studied—from theoretical (e.g., Colbeck, 1972; Colbeck, 1973; Colbeck, 1975; 1976; 1978), to experimental (e.g., Colbeck, 1974; Singh et al., 1997; Juras et al., 2017), and observational standpoints (e.g., Marsh and Woo, 1984; Conway and Benedict, 1994; McGurk and Marsh, 1995; Kattelmann and Dozier, 1999; Eiriksson et al., 2013; Rücker et al., 2019)—we have yet to fully grasp the complexities of water routing through snow. Much of the “unknown” lies in the variability of the snow’s stratigraphy and the complex exchanges of water and energy between layers.

Because mountain ROS environments are arguably one of the most complex encountered, parsing (and forecasting) active vs. passive snowpacks requires a full, physically-based characterization of the energy balance, as opposed to a degree day model—temperature alone does not drive ROS melt (Qi et al., 2017). With reasonable meteorological forcing data, energy balance models are capable of realistic estimates of snow state, i.e., the “bulk” cold content (Jennings et al., 2018). However, the quality of model output is largely predicated on the quality of the meteorological forcing—which can be poor even in observation-rich regions like California’s Sierra Nevada (Lundquist et al., 2019; Terzago et al., 2020). Denser, and improved station networks to capture complex meteorological drivers would greatly enhance our modeling capabilities.

While the meteorological environment during ROS is difficult to observe, arguably harder still are the exchanges of water and energy within the snowpack itself. To represent these processes, models require multi-layer snowpacks, with both Darcy and Richard’s equations for flow, and ways to incorporate preferential flow (i.e., Wever et al. (2016)). But our physical understanding of these processes and their variability is still in its infancy. Advancing modeling efforts requires more rigorous measurements, e.g., TWI from networks of lysimeters (Marsh and Woo, 1985; Kattelmann, 2000; Eiriksson et al., 2013; Webb et al., 2018b; Rücker et al., 2019), observations of stratigraphy and measurements of LWC (e.g., Koch et al. (2019), Mavrovic et al. (2020); Eole and Michel (2021), Capelli et al. (2022)). This “measurement space” needs clever, new engineering solutions to achieve a predictive understanding of ROS.

Impactful ROS floods are rare, unique, and occur during varied meteorological conditions with swings in flood driving mechanisms. The causative components for ROS floods (rainfall vs. snowmelt) can vary storm-by-storm depending on whether a snowpack is passive or active, and the routing mechanisms at play. Moving forward, we first encourage a more vigorous debate on the definition(s) of ROS—having agreed upon definitions will foster better intercomparison studies and understanding of future ROS risk. Second, we encourage funding for established and novel measurements and improved networking across systems (Hatchett et al., 2020). Specifically, we suggest: 1) improved and denser observation networks for the phase, temperature, and intensity of precipitation; and 2) the development of quickly deployable and movable instruments to measure the snow energy balance, snow cold content, and snowpack flow regimes. These new observations would help to constrain model physics with regards to passive vs. active snowpacks, and ensure extreme events are skillfully forecast for the correct physically-based reasons (Kirchner, 2006).