Yang Mu

I am a PhD student in Data Science and Earth Observation at Technical University of Munich, advised by Prof. Xiao Xiang Zhu and Muhammad Shahzad. Since April 2026, I am also a Student Researcher at Google DeepMind, working with Maxim Neumann. I obtained M.Sc. in Geomatics from KTH Royal Institute of Technology and exchanged in ETH Zurich in 2022. Prior to that, I earned a B.Eng. in Remote Sensing from Wuhan University in 2020.

My research focuses on the intersection of multimodal learning, spatial-temporal analysis, and foundation models for Earth observation, advancing geospatial reasoning to extract valuable insights, especially for global forest and biodiversity monitoring.

Feel free to reach out for research collaborations, academic discussions, or if you’re interested in the intersection of AI and Earth observation!

news

Apr 13, 2026	Joined Google DeepMind as a Student Researcher, working with Maxim Neumann!
Jan 25, 2026	Our paper GlobalGeoTree has been accepted by Earth System Science Data (ESSD) and is now online!
May 18, 2025	GlobalGeoTree is now available on arXiv! Check out our latest work on global tree species multimodal benchmark and GeoTreeCLIP vision-language model! 🌍🌲
Apr 04, 2025	Our National Tree Species Mapping paper has been accepted by International Journal of Applied Earth Observation and Geoinformation.
Dec 18, 2024	Received the AAAI-25 Student Scholarship as travel grant.

Show older news (1)

Dec 10, 2024	Our paper MPTSNet has been accepted by AAAI 2025.

selected publications

ESSD
GlobalGeoTree: A Multi-Granular Vision-Language Dataset for Global Tree Species Classification

Yang Mu^*, Zhitong Xiong, Yi Wang, and 5 more authors

Earth System Science Data, 2026

Abs DOI arXiv Bib HTML Code

We built a global dataset of 6.3M remote sensing samples covering 21,001 tree species with rich taxonomic labels. We developed GeoTreeCLIP, a CLIP-based vision-language foundation model for joint understanding of remote sensing imagery and species descriptions. Our model achieved strong zero-shot and few-shot performance, setting new benchmarks for multimodal tree species classification. All data and code are open-sourced to support the geospatial AI community.
@article{mu2026globalgeo, title = {GlobalGeoTree: A Multi-Granular Vision-Language Dataset for Global Tree Species Classification}, author = {Mu, Yang and Xiong, Zhitong and Wang, Yi and Shahzad, Muhammad and Essl, Franz and Kreft, Holger and van Kleunen, Mark and Zhu, Xiao Xiang}, journal = {Earth System Science Data}, volume = {18}, pages = {1379--1403}, year = {2026}, publisher = {Copernicus Publications}, doi = {10.5194/essd-18-1379-2026}, url = {https://essd.copernicus.org/articles/18/1379/2026/}, dataset = {https://huggingface.co/datasets/yann111/GlobalGeoTree} }
Int. J. Appl. Earth Obs.
National-scale tree species mapping with deep learning reveals forest management insights in Germany

Yang Mu^*, Jianhua Guo, Muhammad Shahzad, and 1 more author

International Journal of Applied Earth Observation and Geoinformation, 2025

Abs DOI Bib HTML Code

We designed ForestFormer, a dual-branch neural network with spectral-spatial attention. Our model achieved 84% accuracy across 8 dominant tree species using Sentinel-2 time series data. We generated a comprehensive, high-resolution tree species distribution map for Germany and provided forest management insights on species resistance and biodiversity levels. Code and pre-trained models are released to support open research and applications.
@article{mu2025national, title = {National-scale tree species mapping with deep learning reveals forest management insights in Germany}, author = {Mu, Yang and Guo, Jianhua and Shahzad, Muhammad and Zhu, Xiao Xiang}, journal = {International Journal of Applied Earth Observation and Geoinformation}, volume = {139}, pages = {104522}, year = {2025}, publisher = {Elsevier}, doi = {10.1016/j.jag.2025.104522}, url = {https://www.sciencedirect.com/science/article/pii/S1569843225001694}, }
AAAI
MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification

Yang Mu^*, Muhammad Shahzad, and Xiao Xiang Zhu

In Proceedings of the AAAI Conference on Artificial Intelligence, 2025

Abs Bib HTML Code

We designed a Multiscale Periodic Time Series Network for multi-periodicity analysis. Our approach combined CNN-based local pattern extraction with attention mechanisms for global dependency modeling. MPTSNet outperformed 21 existing baseline methods in extensive benchmark datasets. We released codes to support reproducible research and open-source applications.
@inproceedings{mu2025mpts, title = {MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification}, author = {Mu, Yang and Shahzad, Muhammad and Zhu, Xiao Xiang}, booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence}, volume = {39}, year = {2025}, organization = {AAAI}, url = {https://ojs.aaai.org/index.php/AAAI/article/view/34155}, }

Since June 2025