It may seem that there are predictions in the plan that accurately predict whether rain or snow is not very important, unless you are skiing, snowboarding or driving in a canyon.
The reality is that when scientists develop predictions and provide data to be used in making decisions about everything from agriculture to air traffic control, they predict that either rain or snow is essential.
Prediction and climate science relies on creating models from a moving amount of data. This is like a “big data” problem where artificial intelligence tools should be better at solving, but a team of researchers have discovered that there are important limitations that can be achieved when AI operates with weather data from sensors near the surface of the earth.
Published in Nature Communications, led by the University of Vermont and made a significant contribution from Utah State University climate scientist Wei Zhang, this study evaluates traditional methods of what stages of precipitation are at a particular location: liquid rain or solid snow.
The researchers ultimately found that the temperature of the freeze produces a limit on knowing whether precipitation has decreased as rain or snow. The findings have significant implications for weather forecasting, climate research, and the development of mathematical models of the hydrological cycle that informs decisions regarding water resource management and flood and drought forecasting. These and other data-driven decisions affect agriculture decisions, transportation safety, infrastructure management and building standards.
“Snow has a major impact on water resources, food security, recreation, local and local planning and transportation safety,” Zhang said. “For Utah and the rest of the western states, more than 95% of its water resources come from snow.”
Keith Jennings, director of research at the University of Vermont Water Research Institute and lead author of the study, explained behind the paper that one reason model cannot be consistently and reliably predicted. Most precipitation data in the US comes from sensors in lowland valley areas, and is an important difference between these speckles and mountainous regions.
This gap in knowing exactly where the rain and snow is falling is especially important in mountainous regions such as Utah’s Wasatch and Winta Mountains. There, the precipitation stage plays an important role in snow development. Snow rain promotes thawing and increases the risk of flooding, while snowfall contributes to long-term storage.
“Like many other important Earth system variables, there is little direct observation of precipitation stages,” Jennings writes. “This means you don’t know if it’s raining or snowing in certain places. This is a particularly serious challenge in mountain ranges with atmospheric temperatures near freezing.”
Teach your computer to recognize rain and snow
Historically, scientists and predictors have relied on mathematical methods to estimate precipitation stages using variables such as temperature, humidity, and pressure.
However, these models only work well in very warm or cold conditions. At temperatures near freezing, traditional methods struggle due to the meteorological similarity between rain and snow.
To address this challenge, researchers incorporated machine learning techniques to train computers to recognize patterns and determine variables traditionally used in accurately recognized rain or snow predictions.
“On the surface, machine learning seemed very suitable for this problem,” Jennings said. “We had two large datasets, one from crowdsourced visual observations, and the other from weather reports. We were hoping that machine learning models would improve accuracy by identifying complex patterns. However, the improvements were small. The best machine learning models only improved accuracy by 0.6% compared to the best traditional methods.”
Chang, an assistant professor at the USU’s Plant, Soil and Climate Bureau, who has been studying artificial intelligence applications in climate science since 2010, emphasized that the study highlights AI limitations when using only surface weather data.
“AI models still struggle between 0°C and 4°C, as rain and snow share almost identical weather conditions at these temperatures,” Zhang said. “More diverse data and physical variables are needed to improve predictions. That’s why my student, Cody Ratterman, is working with Logan’s Apogee Instruments Inc. to develop sensors that can measure rain and snow more accurately.”
This study highlights the value of citizen science and weather observation. To collect real-world data, the researchers have created the Mountain Rain or Snow Project, a NASA-funded initiative that recruits volunteers to report whether precipitation is raining or snow, through a smartphone app.
“Regardless of the data we collected (nearly 100,000 observations and counts), we continued to run into the same problem. Surface weather measurements alone cannot guarantee that rain and snow are distinguished with near freezing temperatures,” Jennings said.
The Mountain Rain or Snow Project also includes locations and ranges of Utah’s Great Basin, as well as opportunities for reporting from other regions. More information and participation details are available at rainornow.org.
“Observation is the most important source of information to advance science,” added Zhang. “USU is working to understand the precipitation stages, but data from multiple sources is needed to continue improving our models. Civic science plays an important role in that effort.”
