Graduate Exam Abstract

Abstract: Accurate precipitation estimation from satellite sensors remains a significant challenge due to uncertainties inherent in traditional parametric retrieval algorithms. To overcome these limitations, we develop a deep learning (DL)-based precipitation estimation framework utilizing observations from geostationary satellites. However, conventional DL models are typically limited to learning from data within predefined study domains and fixed sets of atmospheric variables, restricting their adaptability to new geographic regions and evolving environmental conditions. This study introduces a novel DL-based precipitation estimation system that enhances model generalization across diverse domains and integrates additional atmospheric variables to improve accuracy. The framework primarily leverages observations from the Geostationary Operational Environmental Satellites (GOES) as input features, with ground-based Multi-Radar Multi-Sensor (MRMS) rainfall rate estimates serving as reference data. The proposed model is designed to extend its applicability beyond the initial training regions, supporting precipitation estimation across previously unseen domains. Furthermore, by incorporating supplementary atmospheric variables such as brightness temperature, cloud properties, and moisture profiles, the system demonstrates improved capability in capturing complex precipitation dynamics. This work highlights the significance of adaptive DL frameworks in advancing toward hemispheric and global-scale precipitation retrieval using geostationary satellite observations.

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

Publications:
Deep Learning for Precipitation Retrievals Using ABI and GLM Measurements on the GOES-R Series

Program of Study:
ECE-513
ECE-514
ECE-516
ECE-520
ECE-656
MATH-560
MATH-580
STAT-620