Deep Learning-Based Precipitation Classification Using Global Precipitation Measurement Dual-Frequency Precipitation Radar
Year: 2026
Type: article
Abstract: The Global Precipitation Measurement (GPM) Dual-Frequency Precipitation Radar (DPR) is a cornerstone of meteorological and climate research. Current GPM data products classify precipitation types using threshold-based methods applied to dual-frequency measurements from vertical profiles of reflectivity (VPRs) and auxiliary parameters. While effective for operational classification, these methods rely on predefined decision rules that can be sensitive to calibration and environmental variability and may not capture the complex, nonlinear relationships present in precipitation structure, prompting interest in leveraging machine learning for improved classification. This study introduces a deep neural network (DNN) that utilizes statistically derived features as well as convolutional features extracted from VPRs as inputs. Using a novel labeling strategy based on data from the Multi-Radar Multi-Sensor network, the proposed model captures both the statistical and gradient patterns of precipitation for stratiform precipitation, snow, convective precipitation, and hail. For these 4 types, the DNN improves precision from 0.70 to 0.87 and recall from 0.67 to 0.88 relative to the GPM/DPR Version 07 product. These results demonstrate the DNN approach’s potential to provide an accurate solution aligned with the GPM mission to advance meteorological and climate research.
Source: Journal of Remote Sensing
Authors: Daniel Durbin, Yadong Wang, Lin Tang, Chao Lu
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
SDG: Climate action
Open Access status: diamond
OpenAlex ID: W7208738859