《地空学术讲座 第475期》张恩泽 创新及科技基金研究人才库博士后：从冷到热：利用遥感数据进行地球科学中的深度学习应用

• 题目: 从冷到热：利用遥感数据进行地球科学中的深度学习应用
• 主讲人：张恩泽 创新及科技基金研究人才库博士后 @ 香港科技大学创新及科技基金研究人才库
• 时间：2026年3月5日 15:30-16:30
• 地点：理学院E3152会议室

主讲人简介

张恩泽，香港科技大学創新及科技基金研究人才庫博士后。本科毕业于中国科大地球物理系；博士毕业于香港中文大学地球与环境科学系。长期致力于人工智能在地球科学领域应用，发表TC、RSE、JGR-ML等SCI论文多篇。担任13个包括npj Climate and Atmospheric Science, Remote Sensing of Environment, IEEE Transactions on Geoscience and Remote Sensing等SCI期刊以及美国国家自然基金评审。曾获得德州大学奥斯汀分校杰出博士后奖，香港中文大学全球卓越研究獎學金等。

讲座简介

The proliferation of Earth observation data, particularly from remote sensing, has catalyzed the widespread adoption of deep learning in Earth sciences. While this vast data volume poses significant processing challenges, it simultaneously offers unprecedented opportunities for data-intensive methodologies. The diversity and scale of these datasets enhance model training, fostering robust generalization capabilities that ensure reliability in practical applications. This presentation highlights two applications of deep learning in this domain: (1) the automated extraction of glacier termini to construct a comprehensive dataset, and (2) a deep learning-based downscaling algorithm for high-resolution tropical cyclone (TC) wind forecasting.

In the first study, we successfully identified 278,239 termini across 295 Greenlandic glaciers using multi-sensor remote sensing imagery. The resulting dataset, characterized by high temporal sampling and rigorous quality control, facilitates improved quantification of ice sheet dynamics and refines parameterizations of ice-ocean interactions. In the second study, we significantly enhance the accuracy of AI-driven weather models for tropical cyclones. Our approach delivers an operationally ready solution capable of generating timely, high-fidelity TC wind forecasts, marking a substantial advance in extreme weather prediction.

海报链接