The Snow Ratio and Gary's Snowfall Formula (GSF):

Through various measurements, forecasters predict liquid precipitation which falls as snow during cold weather. The amount of snow per amount of water is the snow ratio. Traditionally, a 10:1 ratio (10 inches of snow per inch of liquid precipitation) has been used as an approximation, but the actual ratio depends on several factors and can be much lower or higher. Ratios higher than 20:1 are unusual, but not unprecedented in western NY. Dense (wet, heavy) snow produces less snow per amount of liquid precipitation than low density (light, fluffy) snow. In 2002, Roebber et al described a neural network method that results in a table of probabilities, e.g., an 80% chance of an inch of snow and a 20% chance of two inches, and the snow ratio is now calculated with this and similar complex formulas.

The US National Weather Service sometimes provides a liquid precipitation prediction with no snowfall prediction, even when the predicted temperature is below freezing. In these cases, I still want my table to show a snowfall prediction. The Roebber method uses match techniques that are more sophisticated than I have been willing to code up to now, so I developed a custom formula (Gary's Snow Gormula, GSF) based on Roebber's principles. In practice, GSF shows lower predictions than the NWS. Actual snowfall is difficult to measure in the first place due to settling, melting and sublimation (the process by which ice and snow change directly to water vapor without first melting, i.e., icicles can disappear when the temperature is below freezing). The snow ratio can vary widely across a small geographic area, and large bodies of water affect the ratio, also known as "lake effect snow".

The big northeast storms of February, 2007 buried the Tug Hill plateau north of Syracuse, and later Chicago and Cleveland. The earlier lake effect storms largely spared my webcam location, and the NWS point forecasts over-predicted snowfall. The point forecasts for the later Nor'easter (in which warm, moist, southerly air pushed into an arctic cold front) predicted a 30:1 snow ratio. My lat/lon observation was that the NWS over-predicted the early-storm snowfall, under-predicted the late-storm snowfall, and slightly over-predicted the total. My GSF badly under-predicted the snowfall from the unusual combination of single-digit temperatures and near-100% humidity.

GSF first calculates a predicted snow density factor.
Begin with 1 for Jan/Feb, 2 for Nov/Dec/Mar/Apr, 3 for Oct, 4 for May/Jun/Jul/Aug/Sep.
Add [((temperature + 15) * 2) / 25]
Add [relHum / 25]
Add [(windSpd * 2) / 25]
To account for the effect of the Great Lakes, subtract 1 from the density if the wind direction is nearly NW, and subtract 0.5 if the direction is between WNW and N, but this is admittedly an approximation and is optimized for Rochester.
Based on expectations of maximum and minimum temperatures and wind speeds, the density should be between 0 and 16, and the resulting snow ratio is based on a typical ratio of 9 inches of snow per inch of water.
ratio = 81 / density
snowfall = predicted liquid * ratio