NWP Snow Accumulation Products - Forecast Guide

What you see in AWIPS, WSUP, and various websites is usually not straight from the model! Only the HRRR and RAP produce explicit snowfall output (and most websites don't display it). The other NCEP models produce a snow water equivalent (SWE) product, but AWIPS and most graphics on the web use the QPF output instead of SWE.

The ECMWF and Environment Canada models keep track of precipitation type during each model step but then output an overall SWE covering multiple hours, and end users (AWIPS and websites) have to convert to snow accumulation on their own.

Below are how various model snow accumulation product displays are derived:

AWIPS Snow Accumulation Product

The AWIPS baseline software produces snow accumulation plots/graphics using an outdated thickness-based method. Bottom line -- DO NOT USE in and around marginally-warm thermodynamic environments and in mixed precipitation areas! There are often significant differences seen in AWIPS plots compared to what various websites (rapidrefresh.noaa.gov, pivotalweather, etc.) plot.

The issue revolves around how AWIPS processes the data. For most models (except NBM), the AWIPS snow accumulation products use "rules of thumb" techniques that are 25+ years old. For all models except the NBM, AWIPS snow accumulation maps are:

• Using model QPF and NOT snow water equivalent

• Calculating on-the-fly based on model 1000-500 hPa thickness and surface topography/elevation (NOT temperature profiles!)

• Employing one of four possible Snow:Liquid Ratio (SLR) calculations at each grid point based on thickness:

  • SLR 0:1 (thickness implies too warm for snow)

  • SLR 5:1 (thickness implies marginal liquid/snow and/or presence of sleet)

  • SLR 10:1 (thickness implies cold enough for all snow)

  • SLR of 15:1 (thickness implies very cold air mass)

  • SLR category adjusted upward at higher elevations

  • Multiplying the QPF by the calculated SLR in CAVE/D2D to produce contour/image plot

NOTE: This does NOT impact the raw model data. As a workaround, you can load individual hours into GFE and view the output there to see what the models may be forecasting for snowfall, but when producing final grids, use of standard regional or national tools designed for spatial consistency is recommended (e.g., ForecastBuilder).
 

1-hr HRRR Snow Accum Using Thickness from CAVE/D2D

The 1 hour forecast graphic of Snow Accumulation Using Thickness from AWIPS shows a swath of 2-4" snowfall surrounded by a large area of 1-2" across parts of the west Texas. Note the artificial discontinuities in snowfall where the AWIPS algorithm tries to take into account changes in elevation.
 

HRRR Snow Depth via rapidrefresh.noaa.gov for the same time period

The snow depth reported from the HRRR barely breaks 0.1" anywhere in the area of the 1-4" of snow accumulation depicted in the AWIPS product.

HRRR Snow Depth Change via PivotalWeather for the same time period

The output from this external website is similar to the rapidrefresh.noaa.gov site, so confidence is high that the AWIPS method is producing unreasonable output.

Post-produced by NCEP for most of their models (GFS, GEFS, NAM, and HiResW, but NOT the HRRR/RAP), the snow water equivalent (SWE) product attempts to determine what proportion of the precipitation produced by the model will be snow upon reaching the surface, which can then be converted to snow accumulation using one of the techniques below. This is more advanced than simple thickness/elevation techniques applied to QPF, but is still limited by the inability to separate ice and/or freezing rain.

• Does attempt to take into account surface processes (melting, sublimation, and compaction) in each model time step

• Output every hour

• Available in AWIPS through the Volume Browser or Product Browser but is displayed in liquid equivalent amount

• Users must apply their own SLR to SWE to get a snowfall accumulation amount

• Snow and sleet are tallied together

• Does not separate out freezing rain

• Some external websites use SWE to create their snowfall graphics, but most use QPF (check with website)

Constant 10:1 SLR Product

• Not currently available in AWIPS but ubiquitous on websites

• Most websites use QPF, some attempt to use SWE if available

• Multiplies model QPF or SWE by 10 to get snowfall in areas deemed cold enough for snow

• Methods for determining snow vs. rain temperature profile vary by website

• Multiple of 10 or nothing, so will overestimate snow in marginally-warm temperature environments and underestimate snow in very cold air masses containing dendrites

• Websites will often categorize sleet as 10:1 snow, causing potentially-significant overestimation where sleet mixes in or is the primary precipitation type

Kuchera Product

• Not available in AWIPS but available on many external websites

• Uses a simple variable density formula to determine SLR based on maximum temperature in the column (Tmax):

  • If Tmax > 271.16 K: SLR = 12 + 2(271.16-Tmax)

  • If Tmax <= 271.16 K: SLR = 12 + (271.16-Tmax)

• Does a good job lowering SLR in marginally-warm environments and raising SLR in very cold environments where dendrites are expected

• However, it does not take into account the breaking off of dendrite branches during descent in windy conditions, so it can overestimate snow accumulation in very cold and windy environments (blizzards)

• Can still capture some freezing rain or sleet as snow, but generally at lower amounts than constant 10:1 or thickness-based methods

• Most external websites use QPF to multiply with the SLR, some attempt to use SWE if available (check with website)

• Generally captures gradients in accumulation well in environments with temperature gradients during dendrite-predominate snow events

Snow Depth Product

• Directly produced by the HRRR and RAP, this uses a simplistic SLR that is a function tied only to the 2 meter temperature

• SLR and melting are assessed at each model time step which usually provides better accumulations compared to other post-processed products like Kuchera, thickness, constant 10:1, etc.

  • GFS uses a rough parameterization starting with data from the Interactive Multisensor Snow and Ice Mapping System to see where there is current snow cover.

• Depth is updated at 00 UTC with a blend of the 23km USAF snow depth field and 6-hour forecasted snow depth from the previous cycle (so, for example, the 6 hour forecast from the 18 UTC run for the 00 UTC run initialization)

• Once the model has started running, snow cover and water equivalent snow depth are allowed to evolve, affecting albedo for radiative transfer and impacting the surface/land interaction. This will change the water equivalent snow depth but does not explicitly produce snow depth (must use conversions!)

  • NAM starts the same as the GFS with the USAF analysis, but also uses the 4 km daily NESDIS northern hemisphere snow cover analysis and is updated daily with data valid as of 18 UTC, and is available in time for the 06 UTC and later runs of the model

  • Once the model has started running, NAM microphysics takes into account fractional snow amounts (to separate snow from rain), albedo, snowmelt, and sublimation

  • Model tries to take into account new snow density from air temperature in the first layer and surface temperature -- a rough proxy to account for compaction

  • Boundary layer microphysics scheme also tries to account for heat flux and sublimation over time, with surface land model used for ground temperature

Change-in-Snow Depth Product

• Produced by many websites, this is simply the net positive increase in snow depth reported from the model over the indicated time duration

• Does not take into account any negative contributions to depth over time such as melting or rain falling on top of snow, nor compaction or drifting

• Does not take into account freezing rain or sleet mixing with falling snow

• Often found to be the best representation of verifying storm total snow accumulation in events where precipitation type is not a source of uncertainty

A Note on NOHRSC Products

NOHRSC snow depth products feed into most models to initialize surface snow cover.

• This can have a big effect on 2 meter/surface temperature forecasts, so check satellite imagery to estimate actual snow surface cover and compare against model snow depth displays at the same time step

• In many areas, the NOHRSC analysis relies on automated remote liquid precipitation sensors where an estimated SLR is applied

• Keep in mind measurement uncertainties that might creep into the NOHRSC analysis