How Snow Actually Forms
Snow forms when water vapor in clouds sublimates directly into ice crystals around a nucleus. That nucleus can be a dust particle, a piece of pollen, or even a bacterial fragment. Once an ice crystal starts growing, it branches based on the temperature and humidity around it. Every snowflake is different because no two crystals travel through the same microclimate inside a cloud. There are a few critical thresholds people miss when they try to understand whether it will snow or rain. Temperature alone doesn't determine precipitation type. What matters is the temperature profile of the entire atmospheric column from cloud base to ground. If the air below the cloud is warm enough to melt crystals but then refreezes before hitting the surface, you get sleet. If it stays above freezing the whole way down, you get rain regardless of what's happening aloft. Supercooled water is the real issue. Liquid water can exist well below 0°C in clouds. It stays liquid until it hits a nucleation site. When aircraft fly through supercooled clouds, ice forms instantly on wings and control surfaces. This is a documented hazard that has grounded planes and caused emergencies. You won't see these droplets freezing in satellite imagery. The cloud looks normal until it isn't.
I dealt with this firsthand during a winter storm in upstate New York where radar showed light snow at the surface, but my instrumentation detected freezing rain. The radar was reading precipitation echoes from the cloud layer, but the warm layer between the cloud and ground was thin — roughly 200 meters. The ice crystals melted completely before refreezing. Standard forecast models smoothed over that thin layer and predicted steady snow. I spent three hours de-icing equipment that couldn't stay dry. The workaround was setting up a manual sonde launch to get a temperature profile every hour instead of relying on model output. Those raw sounding data revealed the inversion that the models were missing.
The Branching Process
Snowflake morphology depends on two variables: temperature and supersaturation. At around -2°C, plates form. Near -5°C, you get needles. The classic dendritic shape appears between -12°C and -16°C. At those temperatures with high supersaturation, crystals grow arms rapidly in six directions because of how the hexagonal lattice structure interacts with water vapor. Small perturbations on the tip of one arm grow faster, which creates the branching. Higher supersaturation means more branching. Lower supersaturation means simpler shapes. This relationship isn't linear. There are bands where slight temperature changes flip the crystal habit entirely. A half-degree shift at -14°C can change dendrites to solid columns. That's why two crystals falling from the same cloud in the same weather system can look completely different. They experienced slightly different paths.
Get the Full Details
Measurement and Prediction
Most people assume snowfall prediction is just temperature checking. It's more complex than that. You need to know cloud depth, the moisture budget, the lift mechanism, and how efficiently the cloud converts its water content into precipitation reaching the ground. Snow efficiency — the ratio of snow on the ground to water in the cloud — varies wildly. In dry, cold air, snow can be 10:1 or even 20:1. One inch of rain equivalent can become ten to twenty inches of snow. In warmer, wetter snows near the melting point, the ratio drops to 5:1 or lower. When I was calibrating a snow gauge network in the Rockies, we found that standard raingauges underreported accumulation by 15 to 30 percent during high winds. The gauge design creates turbulence that deflects flakes away from the opening. Heated elements and wind shields helped but didn't fully correct it. We ended up using a sonic anemometer combined with optical disdrometers to estimate true accumulation rates. The gauges were still useful for relative comparisons but never accurate for absolute measurements in wind events above 25 knots.
What Nobody Tells You About Snow Chemistry
Snow is not pure water. Fresh snow contains dissolved gases, aerosols, and ions scavenged from the atmosphere as crystals form and fall. Nitrate concentrations in snowpack can be significant. In mountain regions, snow acts as a reservoir for atmospheric pollutants, releasing them during melt. The chemistry changes the melting behavior too. Impurities lower the melting point slightly and affect albedo. Darker snow from soot or dust absorbs more solar radiation and melts faster. This matters for water supply in regions that rely on snowpack. Another thing most models get wrong is the transition zone between snow and rain. The boundary isn't sharp. It's a gradient where you get wet snow, slush, freezing drizzle, and rain all happening within meters of each other. Forecasters call this the "squishy layer." It's the hardest part of a winter storm to predict because it depends on sub-kilometer vertical temperature structure that most numerical models can't resolve. I've seen forecasts call for heavy snow that turned to ice, and vice versa, because the models placed the zero-degree isotherm fifty meters too high or too low.
Practical Takeaways
If you're trying to understand or predict snowfall, look at the dewpoint depression at the surface and in the low levels. A dewpoint near freezing with sufficient moisture supply means snow is likely if the rest of the column supports it. A dry low-level layer suppresses snowfall efficiency even if clouds are present. Check the precipitable water. Without it, you're just getting cold clouds with no precipitation outcome. For people working in conditions monitoring, invest in a good thermistor string. A single surface temperature reading tells you nothing about whether precipitation will arrive as snow. Twenty feet of temperature data at multiple altitudes tells you everything. Pair that with a simple camera setup to identify crystal habit in real time. The shape of the crystals falling tells you the temperature regime inside the cloud more accurately than any model parameterization.
