Graduate Exam Abstract

Abstract: Accurate precipitation retrieval from geostationary satellites remains challenging due to sensor limitations and the complex relationship between satellite observations and surface precipitation. To address this problem, this research develops a deep learning (DL) framework for precipitation estimation using GOES-R satellite measurements. In addition, we propose null-space-based continual learning (CL) algorithms that improve model generalization across multiple domains without retraining from scratch. Three incremental methods are introduced to efficiently update the null space as new domains are incorporated. Experimental results show that the proposed DL framework improves existing GOES precipitation products, while the CL-based approach enhances cross-domain generalization and maintains robust performance across diverse precipitation regimes. These results demonstrate the potential of scalable DL-based precipitation estimation for large-area geostationary satellite applications.

Adviser: Mahmood Azimi-Sadjadi
Co-Adviser: Haonan Chen
Non-ECE Member: Michael Kirby, Mathematics
Member 3: J. Rockey Luo, ECE
Addional Members: N/A

Publications:
Yang, Yifan, et al. "Deep learning for precipitation retrievals using ABI and GLM measurements on the goes-r series." IEEE Transactions on Geoscience and Remote Sensing 61 (2023): 1-14.

Program of Study:
ECE 513
ECE 514
ECE 516
ECE 653
ECE 654
MATH560
MATH580
STAT620