4.1 Conditional Generative Adversarial Neural Networks for Radar Noise Postprocessing

Monday, 8 January 2018: 3:30 PM
Room 7 (ACC) (Austin, Texas)
Sarvesh Garimella, ACME AtronOmatic, LLC, Portland, OR

Removal of radar noise is often accomplished using alogrithms that leverage the dual polarization capabilities of modern weather radar. Such techniques are very effective at noise removal, but operating on data on a gate-by-gate or radial-by-radial basis is computationally intensive. This study presents a deep learning technique for noise reduction in radar data that is applied to gridded radar imagery. Using a conditional generative adversarial network (CGAN) achitecture implemented with the TensorFlow deep learning library, this approach is portable across platforms and uses GPU acceleration for increased performance. This postprocessing technique can be used in conjunction with or as an emulator of traditional techniques and provides a scalable option for radar noise reduction.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner