On 19 June 2026, Yang Mu successfully defended his doctoral dissertation at the Technical University of Munich (TUM). His research focused on scalable tree species mapping from satellite image time series using deep learning methods.

Forests play a fundamental role in biodiversity conservation, carbon cycling, and climate regulation. However, accurately mapping tree species over large geographic areas remains a major challenge due to high spectral similarity among species, complex seasonal dynamics, imbalanced species distributions, and the limited availability of large-scale reference datasets. Addressing these challenges is essential for biodiversity monitoring, sustainable forest management, and understanding ecosystem change.

Deep learning for Scalable Tree Species Mapping from Satellite Time Series

Dr. Mu's dissertation develops novel deep learning methods for tree species classification using satellite time series, ranging from national-scale mapping to globally transferable foundation models for biodiversity monitoring.

The dissertation makes several key contributions:

National-scale tree species mapping.
The proposed ForestFormer framework enables large-scale tree species mapping from Sentinel-2 time series and was used to produce one of the first nationwide tree species maps of Germany, providing valuable information for forest monitoring and management.

Multiscale temporal representation learning.
The MPTSNet architecture advances multivariate time series classification by combining local temporal pattern extraction with global dependency modeling, improving the representation of seasonal vegetation dynamics.

Phenology-aware tree species classification.
The Hi-PhenoNet framework incorporates phenological information and biological taxonomy into deep learning models, substantially improving fine-grained tree species identification, particularly for rare species.

Global foundation models for biodiversity monitoring.
The dissertation introduces GlobalGeoTree, a global dataset containing 6.3 million tree occurrence records covering more than 21,000 species, together with GeoTreeCLIP, a vision-language learning framework that enables scalable zero-shot and few-shot tree species classification.

Together, these contributions advance the development of scalable AI methods for biodiversity monitoring and demonstrate the potential of satellite image time series and foundation models for global forest observation.

Examination Committee

The examination committee included:

• Prof. Xiaoxiang Zhu (Technical University of Munich, PhD supervisor)
• Prof. Katharina Anders (Technical University of Munich), chair of the committee
• Prof. Muhammad Shahzad (University of Reading), examiner
• Prof. Richard Bamler (Technical University of Munich), examiner

We warmly congratulates Dr. Yang Mu on this important academic milestone and wish him every success in his future research career.