A recent study led by the University of Vermont (UVM) has shed light on the difficulties of accurately classifying precipitation as rain or snow. This research focuses on the challenges posed by near-freezing temperatures in using surface weather data for this task. Published in Nature Communications, the study evaluates both traditional methods and advanced machine learning models, highlighting an inherent limitation in distinguishing rain from snow in such conditions.
The ability to accurately identify precipitation type is crucial for effective weather forecasting, hydrologic modeling, and climate research, impacting transportation, infrastructure, and water management. However, this is particularly challenging in mountain regions, where distinguishing between rain and snow is vital for natural resource management. A predominantly snowfall event can be beneficial to ski resorts and water resources, while a rain-dominated event can result in flooding and damage.
Traditional precipitation phase classification methods rely on weather data to estimate rain or snow, but these methods are typically effective only at extreme temperatures. Near the freezing point, these approaches struggle due to the meteorological similarities between rain and snow. "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, the lead researcher. He noted that despite using complex data and models, machine learning techniques also face difficulties under these conditions.
The study utilized data from nearly 40,000 crowdsourced observations and more than 17 million synoptic weather reports. Dr. Jennings and the research team evaluated various classification methods, including traditional models and machine learning techniques like random forest, XGBoost, and artificial neural networks. The machine learning models offered minimal improvements, highlighting the fundamental challenge of using only meteorological data for classification.
The research underscores the need for alternative methods to improve precipitation classification. Dr. Jennings recommends focusing on new techniques that incorporate diverse data sources, such as crowd-sourced observations, weather radars, and satellite data. "In this new study, we further demonstrate that even the most advanced machine learning methods do not perform well in distinguishing between rain and snow without the incorporation of novel data sources," remarked Guo Yu, Assistant Research Professor and coauthor of the study.