Identifying Precipitation Types From Surface Meteorological Variables With Machine Learning

Precipitation type prediction is crucial for various sectors, including aviation, agriculture, and public safety. For instance, freezing rain and sleet can severely disrupt transportation, while heavy rainfall may lead to flash floods. Using METAR weather reports, we sought to classify precipitation types based on surface variables such as temperature (°C), dew point deficit (°C), wind speed (knots). The dataset was divided into training and testing subsets. Known precipitation types in the training set allowed us to fit classifying algorithms. We evaluated several classification models. The k-nearest neighbors (k-NN) method was initially applied, with parameter optimization performed to enhance accuracy. Precipitation types were first categorized into two groups: liquid and non-liquid types. Liquid included rain, fog and drizzle, excluding convective precipitation. The non-liquid category included solid and supercooled types like snow, sleet, freezing rain, graupel, and supercooled fog. Further refinements classified precipitation into six categories: liquid precipitation, convective precipitation, snow, sleet, ice pellets, and supercooled water (including freezing rain and rime fog). Additionally, classifications into 5 and 4 categories were analyzed. Evaluation metrics, such as sensitivity, precision, specificity, and accuracy, were employed to assess model performance in classifying precipitation types. This work has been implemented by the National Multidisciplinary Laboratory for Climate Change (RRF-2.3.1-21-2022-00014) project within the framework of Hungary's National Recovery and Resilience Plan supported by the Recovery and Resilience Facility of the European Union.