Rain and snow are difficult to forecast at near-freezing temperatures

Predicting whether it will rain or snow might seem like a basic job for weather forecasters. But, as it turns out, classifying precipitation as either rain or snow – especially when temperatures hover around freezing – is far trickier than we might think.

A recent study, led by researchers from the University of Vermont, puts this challenge into sharp focus.

Using massive datasets and both traditional and machine learning models, the experts found that even the most advanced tools can struggle with one key problem: nature doesn’t always make a clear distinction between rain and snow.

The importance of getting it right

Knowing whether precipitation is falling as rain or snow matters more than most people realize. It affects road and air travel, flood forecasting, reservoir planning, and infrastructure maintenance.

In mountainous regions, it’s even more critical. A snow-heavy storm may boost a ski season or replenish water reserves. But if that same storm delivers rain instead, it could trigger floods or damage roads and bridges.

Despite the stakes, weather stations rarely have direct measurements that can tell us whether it’s raining or snowing. Most surface data – such as temperature, humidity, and pressure – come from airports and don’t reflect the complex conditions in mountainous terrain.

Rain and snow at the freezing point

To fill the gap, meteorologists use something called “precipitation phase partitioning.” These are mathematical methods that estimate rain or snow based on surface data.

But there’s a catch: they work best when it’s clearly cold or warm. Around the freezing point, where conditions for rain and snow are nearly identical, the predictions often miss the mark.

“The challenge is at those temperatures near freezing, the air and wet bulb temperature distributions of rain and snow overlap heavily,” explained Dr. Keith Jennings, lead researcher on the project. “This means the traditional partitioning methods cannot consistently separate rain from snow.”

“What surprised us is that the machine learning models did not perform much better. Even by using more data and complex mathematics, they are still trying to tease apart the same information, and they’re seeing rain and snow with almost the exact same meteorological properties.”

New tech, same problem

The team analyzed two major sources of data. One was a collection of nearly 40,000 crowd-sourced weather observations from NASA’s Mountain Rain or Snow project. The other was a dataset of over 17 million weather reports from across the Northern Hemisphere.

They tested both older partitioning techniques – like basic temperature thresholds – and more sophisticated machine learning tools such as random forest, XGBoost, and artificial neural networks. These models had access to the same types of surface weather data.

Despite their complexity, the machine learning models only performed slightly better than traditional methods. The boost in accuracy was tiny – just 0.6% at most.

All methods struggled to classify precipitation correctly near freezing, especially in the 1.0°C to 2.5°C range. The models also failed to reliably identify mixed precipitation (like sleet) and rain that falls when it is below freezing.

The study pinpointed the core issue: rain and snow simply occur under very similar conditions at these borderline temperatures. When the input data doesn’t clearly separate the two, no amount of model sophistication can compensate.

Increasing rain-on-snow events

This research suggests that it’s time to rethink how we approach the rain-versus-snow problem.

Instead of trying to squeeze more performance out of limited surface weather data, scientists may need to explore new types of input. That could mean tapping into different data streams – like radar, satellites, and more crowd-sourced reports.

These tools might provide the missing context that surface measurements alone can’t offer. As the climate continues to shift, getting precipitation forecasts right is only becoming more important.

Rain-on-snow events are expected to become more common, increasing the risk of floods and infrastructure stress. Knowing the difference between a snowstorm and a rainstorm could be the key to managing those risks.

If we want to be prepared for the storms of tomorrow, we may need to stop asking surface data to do what it simply cannot. It is time to start building smarter, more integrated systems that reflect the full picture of what’s happening in the sky.

The study was a collaboration between experts from UVM, Lynker, the Desert Research Institute, the Cooperative Institute for Research in the Atmosphere, the University of Nevada Reno, and Utah State University.

The full study was published in the journal Nature Communications.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–