Snow to Liquid Ratio: Climatology and Forecast Methodologies
LSX WFO Winter Weather Workshop
7 November 2005
Martin A. Baxter
Cooperative Institute for Precipitation Systems
Saint Louis University
Dept. of Earth and Atmospheric Sciences
Forecasting Winter Precipitation is a Two-Step Process
First, the current dynamic and thermodynamic forcings of the storm must be assessed.
Numerical model forecasts must be studied, especially the model quantitative precipitation forecast (QPF).
Second, the evolution of the hydrometeors from their origin to the surface must be predicted.
This evolution will be determined by the vertical profile of temperature and moisture.
This profile will elucidate the type of precipitation- rain, snow, freezing rain, ice pellets, or any combination.
If the precipitation is expected to fall as snow, a snow to liquid equivalent ratio must be determined to forecast the actual snow amount.
Why is liquid ratio important to forecasters?
After forecasting liquid equivalent (QPF), the snow-liquid equivalent ratio must be estimated.
Significant variations in snow to liquid equivalent ratio can occur even within a single storm system
A more clear understanding of the processes that act to vary snow density will enable the forecaster to employ a more scientific process oriented method toward forecasting snowfall, versus commonly used empirical techniques.
A challenge exists to determine the extent of interaction between the dynamical forcing and the microphysical processes that determine snow density (i.e., how efficient is the forcing in producing snowfall from a given amount of liquid equivalent?).
SLR determines the wintertime precipitation efficiency
NWS “New Snowfall to Estimated Meltwater Conversion Table” (Kyle and Wesley 1996)
Utilizes surface temperatures to estimate snowfall from liquid equivalent
Is only marginally effective, as it does not account for geographic location or in-cloud microphysical processes.
Description of Dataset and Methods
A 30 year (1971-2000) climatology of snow to liquid ratios was compiled using NWS Cooperative Observer Summary of the Day data.
Only snowfalls greater than 2” and liquid equivalents greater than 0.11” were included, as this was the standard for Roebber et al. (2003).
Estimated events were discarded.
A station must have recorded at least 15 observations over the 30 year period to be included.
Station distribution of the 30 year climatology
12-13 to 1
11-12 to 1
13-14 to 1
11-12 to 1
Average SLR for each NWS County Warning Area
http://www.eas.slu.edu/CIPS/Research/snowliquidrat.html
Histogram for the Entire Dataset of SLR
Mean:
– 13.53
(Short dashed)
25th Percentile:
– 9.26
Median:
– 12.14
(Long dashed)
75th Percentile:
– 16.67
Sample SLR Climatological Distributions
3674 Observations
A very “average” distribution when compared to the histogram for the entire US
Skewed to high values (0.33)
Variability of Mean SLR within LSX
Ratio typically varies with storm track
Clipper type storms feature higher snow to liquid ratios, as they are colder and contain less moisture.
This leads to growth by deposition.
Storm tracks that are warmer or contain more Gulf moisture feature lower snow to liquid ratios.
This leads to growth by riming, possibly mixed with sleet.
Average SLR for southeastern Wisconsin with various storm tracks (Adapted from Harms, 1970 )
combination
of needles
combination
of bullets
sheath
crystal with
broad
branches
dendrite
simple plate
bullet
solid column
hollow column
(Pruppacher and Klett 1980)
Ice Crystal Habits
(Magono and Lee, 1966)
Crystal habit depends on temperature and degree of saturation
Ice Crystal Riming
Supercooled water droplets impact the crystal as it falls
If riming occurs late, crystals retain original form.
If riming occurs early, the droplet can provide a nucleus for a new crystal.
If riming is significant, graupel (soft hail can form).
Light Riming
Heavy Riming
(Photos from Libbrecht 2004)
Different crystal types will have different amounts of air in between them at the surface (More air = higher SLR) (Less air = lower SLR)
Stellar / Dendritic crystals
Graupel
Low Temperature Scanning Electron Microscope Photography
A Graupel Particle – notice the lack of air space in the particle itself as well as the lack of air space that will result upon stacking
A Dendritic Crystal – notice the air space in the crystal itself and the air space that will result upon stacking
Low Temperature Scanning Electron Microscope Photography
Air space is reduced as a snowpack settles
Air space is reduced even more as snow melts
Billings, MT ACARS Sounding
30:1 SLR Observed (crystals like these?)
(Libbrecht 2004)
A Physically-Based Method Using Climatology
SLR is determined largely by the vertical temperature profile
Thus, the 30-year average SLR is likely associated with an average vertical temperature profile
An SLR value that is higher or lower than the 30-year average is presumably associated with an anomalous vertical temperature profile that is colder or warmer, respectively
For this study, 850 mb anomalies were used, with modifications made based upon the temperature profile below this level and the surface conditions.
Warmer temps – more riming – lower SLR values
Why must we use Climatology?
No correlations between atmospheric variables (temperature and humidity aloft and at the surface) and SLR have been established
Climatology provides an “initial guess” that can be refined by examining the details of the situation
Where is the maximum vertical motion?
What crystal types will form and how will they evolve?
Will significant riming occur?
How will the surface conditions impact the fallen snow?
This method shows how a forecaster can understand the processes that affect SLR, as well as providing a crude quantitative guess
Technique is being used at some NWS offices already (Billings, MT; Glasgow, MT)
Average SLR for 1200 UTC 22 November 1996 – 1200 UTC 23 November 1996
Aberdeen, SD & Lacrosse, WI marked (SLR on bottom, # of reports on top)
Meteograms for 1200 UTC 22 November 1996 – 1200 UTC 23 November 1996
Figure reads left to right
LSE
ABR
Temp at ABR for 1200 UTC 22 November 1996 – 1200 UTC 23 November 1996
-20
-20
-20
-16.5
-16.5
-16.5
-13.5
-13.5
-13.5
-10
-8
-8
-8
-4
-4
0
-10
-10
0
-4
-8
-10
-13.5
-10
-13.5
-13.5
-13.5
-10
Temp at LSE for 1200 UTC 22 November 1996 – 1200 UTC 23 November 1996
-20
-16.5
-13.5
-20
-16.5
-13.5
-10
-10
-10
-8
-4
-8
-4
0
0
0
-4
-4
850 mb Temperature Anomalies 00 UTC 23 November 1996
ABR
9 °C cooler than average
18 SLR vs. 13-14 Avg SLR for Fall
850 mb Temperature Anomalies 12 UTC 23 November 1996
LSE
2 °C colder than average
11 SLR vs. 12 Avg SLR for Fall
Difference due to warm ground temps and warm air in low levels
How do I find the 850 mb Temp Anomalies?
Rich Grumm’s (SOO, State College) ensemble & anomaly page at Penn State University
http://eyewall.met.psu.edu/ensembles/java/ModelDisplay.html
Other Methods for Forecasting SLR – Neural Network
Neural Network – Paul Roebber (UW-MW), Dave Schultz (NSSL)
Neural network is “trained” with conditions (temp, RH, etc.) associated with SLR values for many cases
Network is then able to predict SLR for new cases based upon the nonlinear relationships derived from the training data
“nonlinear” – an exponential relationship for example – changes in a given variable are not associated with changes of equal proportion in another variable
Crude treatment of vertical motion
Gives likelihood of SLR in one of 3 classes, no exact number given
Light – 27:1 (36:1 if > 67% probability)
Moderate – 13:1 (wins tie)
Heavy – 8:1
Most Important Factors in Forecasting SLR
Solar radiation (month)
Low- to midlevel temperature (> 850 mb)
Mid- to upper-level temperature (875-400 mb)
Low- to midlevel relative humidity (> 850 mb)
Midlevel relative humidity (850-700 mb)
Upper-level relative humidity (700-500 mb)
External compaction
RH information useful, but less essential
Function of neural network is to sort out the non-linearity
Behaves like a human brain – takes in new data and matches it to patterns it has “experience” with – even though the new data does not exactly match any of the old data
Verification of the Roebber Algorithm
Neural network vs. surface chart
% Snowfall forecast error (cm)
15.2 cm = 6”
4.0 cm = 1.5”
37 cases
2004-05
Other Methods for Forecasting SLR – Cobb Method
Cobb Method – Dan Cobb (SOO, WFO CAR)
Similar to a top-down approach to forecasting
Uses vertical motion information
Keys in on the Snow Production Zone (SPZ), where temperatures are -12 to -18 °C and the Bergeron process is maximized
Accounts for temporal evolution of SLR
Code is not complex and a gridded product has been created
Beneficial for forecasting responsibilities at both national and local/regional levels
Somewhat empirical, but also often accurate
The Cobb Method
Find max UVV in a cloudy layer (RH with respect to water > 75%)
Calculate a weighting factor to be applied to all layers that meet criteria
Although the concept is physically sound, the determination of the formulation of the weighting factor is highly empirical (and very important).
The layer with the highest vertical motion will contribute the most to the observed snow ratio
It will determine the dominant crystal type
The Cobb Method
Calculate a snow ratio for each model layer based on temperature
Curve is generated via a cubic spline through 6 data points that are based on observations of SLR vs. crystal type by Ivan Dube (MSC) and from my climatology (this curve is also “bumped up” to account for extreme events)
Is this curve accurate? Does “one size fit all”?
The Cobb Method
4. Use this formula to calculate the weighted contribution to the snow ratio for each layer:
5. Sum the results of this formula over all the layers to receive the predicted snow ratio
The Cobb Method – Sample Calculation
Weight = 4%
250m thick layer
Weight = 36%
500m thick layer
Weight = 60%
250m thick layer
10 x 4% = 0.4
45 x 36% = 16.2
6 x 60% = 3.6
20:1
The Cobb Method
Vertical motion max collocated with SPZ:
High rate of high ratio snowfall (dendritic)
Vertical motion max below the SPZ:
Warmer temps, more supercooled water leading to riming, lower ratio snowfall
Vertical motion max above the SPZ:
More difficult to discern resulting ratio, crystals fall through many layers resulting in different types of growth
Vertical motion is required to supply supercooled water in lower levels – thus the effects of riming are implicitly included
Crystals falling from aloft will deplete the liquid
Verification of the Cobb Method
For the 2001-2002 winter season (October-April)
24 hour observations of SLR were paired with model vertical profiles where snowfall was observed during the 24 hour period
Using the 3 hourly, 32 km, Regional Reanalysis data
25 mb vertical resolution below 700 mb, 50 mb above
3 hourly Cobb SLR’s were summed to produce a daily total
Sources of error are considerable when doing point based verification
Bad SLR measurement
Reanalysis still not totally representative of real atmosphere
Compare 2 weighting functions –
Original (#1) – ωlayer/ωtotal
New (#2) – Previously discussed
Mean = 5.3 for MAE < 15
Mean = 6.7 for MAE < 15
Verification of the Cobb Method – Jan 31 2002
Cobb Method
True Values
One More Method – Dube Method (MSC)
Makes the “top-down method” quantitative via a decision-tree styled algorithm
http://meted.ucar.edu/norlat/snowdensity/from_mm_to_cm.pdf
Verification of algorithm by Jessica Cox (McGill University) indicates the method performs equal to or better than the 10:1 approximation 83% of the time
Not currently available in the US
Sample Case – Roebber Method
Using GFS
40 km
Assumes all QPF snow
24 hour snow total from 3 hourly SLR
Ending 12Z 10-26-05
Sample Case – Cobb Method
Using GFS
40 km
Assumes all QPF snow
24 hour snow total from 3 hourly SLR
Ending 12Z 10-26-05
Verification – 24 hr Snowfall Totals
http://www.nohrsc.noaa.gov/nsa/ (NOHRSC)
Snow to Liquid Ratio Overview
A 30 year (1971-2000) climatology was completed for snow to liquid ratio
Average snow to liquid ratios are typically higher than 10:1; more like 13:1 for much of the country
Certain storm tracks will exhibit varying snow to liquid ratios based upon in-cloud temperatures and relative humidities. Time of year and external compaction also play dominate roles.
We can attempt to extract out the effects of microphysical interactions by determining the type of crystals that formed and how these crystals grew and changed.
In most cases it is possible to determine relative areas of higher and lower SLR values & obtain a crude estimate through use of low-level temp anomalies
Two new forecasting methods are being tested at HPC during the 2005-2006 winter weather experiment
These may become operational next season