An Artificial Neural Network Based Precipitation Classification Algorithm Using GPM DPR Data

The Global Precipitation Measurement (GPM) Dual-frequency Precipitation Radar (DPR) has become an invaluable tool for meteorological research and applications, particularly in hydrometeor classification. The current GPM data products utilize a combination of bright band detection, the slope profile of Z above the 0°C isotherm, and dual-frequency ratio (DFRm) characteristics to identify various precipitation types, including stratiform, convective, hail, snow, and graupel. However, there is ongoing interest in refining and enhancing these classifications using advanced machine learning techniques.

In this work, we propose an artificial neural network (ANN)-based approach for precipitation classification using GPM DPR data, validated with a large dataset spanning the 2022 calendar year. Our methodology involves a statistical analysis of key features within the vertical profiles from both Ka- and Ku-bands to identify the most effective input features. The ANN model is then trained and validated using precipitation classification results from the Multi-Radar/Multi-Sensor (MRMS) network data products synchronized with the GPM DPR data collections.

The proposed ANN approach focuses on classifying precipitation into four primary types: stratiform, convective, hail, and snow. Initial results are highly promising, with accuracy comparisons showing improvements over the current GPM algorithms. Ongoing work aims to extend the classification to capture the full set of MRMS precipitation classes, identify additional useful features, and further enhance classification accuracy across all precipitation types.