AI-Based High-Resolution Forest Monitoring
This project develops advanced AI methods to monitor forests using satellite imagery, including radar and optical data. It creates scalable techniques for detailed, high-resolution global maps to monitor canopy height, biomass, and track forest disturbances.
IOL Project Members
External Project Members
Funding
This project is being funded by the Federal Ministry of Research, Technology and Space and the VDI/VDE Innovation + Technik GmbH (project ID 16IS23025B) from June 2023 to May 2027.
Project Description
Accurate and up-to-date information on global forest health, disturbances, evolution of above ground biomass and carbon storage is essential for policy-makers to design effective strategies for climate adaptation and mitigation. Advances in Earth observation and AI have paved the way for automated forest monitoring using satellite data, along with radar and LiDAR data. However, existing forest maps often have low resolution, cover limited areas or fail to track temporal dynamics.
The AI4Forest project brings together experts in AI, Applied Mathematics, Computer Science, Remote Sensing, and Climate Science, being a collaboration between ZIB, LSCE, CNRS, WWU and TUM. The project aims to develop scalable AI methods for forest monitoring and to efficiently and cost-effectively process large datasets. These advancements will enable the creation of detailed forest maps with high spatial and temporal resolution, down to individual trees and species, and allow for regular updates across Europe and globally.
Estimating Canopy Height at Scale
Recently, we released a global canopy height map with a 10m resolution, based on satellite data from 2020. You can find the paper on arXiv. Below is an image of the height map, you can explore the interactive global map on the Google Earth Engine.
The computational pipeline to produce the height map can be visualized as follows:

Comparing our map with two other existing global height maps (Lang et al., Potapov et al.), as well as a regional map for France, shows a significant improvement in visual quality.

Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation
Going a step further, we also built a model that takes the temporal evolution of canopy height over multiple years into account. A crucial step in this process, which we describe in our paper, is to use 12 images per year (i.e., one image per month) as input, as opposed to a median-composite. While this is computationally more expensive and could include more clouds, the model is able to capture seasonal variations as well as geolocation shifts. While we provide the first map at 10m resolution for the entire european continent and for years 2019-2022, our model is able to estimate canopy height much better, as the following comparison to the ALS ground truth shows.

Restricting our model to predicting only the height map for the year 2020, we can compare the results to other maps for the same year, including our previous map. We are able to significantly improve both the visual quality as well as the accuracy of the map.

Talks and posters
Poster presentations
- Apr 2026
- Neural Sum-of-Squares: Certifying the Nonnegativity of Polynomials with Transformers by Nico Pelleriti
14th International Conference on Learning Representations (ICLR), Rio de Janeiro - Apr 2026
- RECON: Robust Symmetry Discovery via Explicit Canonical Orientation Normalization by Alonso Urbano
14th International Conference on Learning Representations (ICLR), Rio de Janeiro - Dec 2025
- Computational Algebra with Attention: Transformer Oracles for Border Basis Algorithms by Nico Pelleriti
39th conference on neural information processing systems (NeurIPS), San Diego - Jul 2025
- Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation by Berkant Turan and Max Zimmer
42nd International Conference on Machine Learning (ICML), Vancouver - Jul 2025
- DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for Earth Observation Applications by Max Zimmer
42nd International Conference on Machine Learning (ICML), Vancouver - Jul 2024
- Estimating Canopy Height at Scale by Max Zimmer
41st International Conference on Machine Learning (ICML), Vienna - May 2024
- Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging by Max Zimmer
12th International Conference on Learning Representations (ICLR), Vienna - May 2023
- How I Learned to Stop Worrying and Love Retraining by Max Zimmer
11th International Conference on Learning Representations (ICLR), Kigali - Mar 2023
- How I Learned to Stop Worrying and Love Retraining by Max Zimmer
Workshop on Optimization and Machine Learning, Waischenfeld
Publications and preprints
Preprints and reports
- Roux, C., Zimmer, M., d’Aspremont, A., and Pokutta, S. (2025). Don’t Be Greedy, Just Relax! Pruning LLMs via Frank-Wolfe.
[arXiv]
[BibTeX]
@misc{2025_RouxZimmerDaspremontPokutta_Pruningfrankwolfe_2510-13713, archiveprefix = {arXiv}, eprint = {2510.13713}, arxiv = {arXiv:2510.13713}, primaryclass = {cs.LG}, author = {Roux, Christophe and Zimmer, Max and d'Aspremont, Alexandre and Pokutta, Sebastian}, title = {Don't Be Greedy, Just Relax! Pruning LLMs via Frank-Wolfe}, date = {2025-10-15} } - Pauls, J., Schrödter, K., Ligensa, S., Schwartz, M., Turan, B., Zimmer, M., Saatchi, S., Pokutta, S., Ciais, P., and Gieseke, F. (2026). ECHOSAT: Estimating Canopy Height Over Space And Time.
[arXiv]
[BibTeX]
@misc{2026_JanEtAl_EchosatCanopyheight_2602-21421, archiveprefix = {arXiv}, eprint = {2602.21421}, arxiv = {arXiv:2602.21421}, primaryclass = {cs.CV}, author = {Pauls, Jan and Schrödter, Karsten and Ligensa, Sven and Schwartz, Martin and Turan, Berkant and Zimmer, Max and Saatchi, Sassan and Pokutta, Sebastian and Ciais, Philippe and Gieseke, Fabian}, title = {ECHOSAT: Estimating Canopy Height Over Space And Time}, date = {2026-02-24} }
Conference contributions
- Wagner, M., Roux, C., Zimmer, M., and Pokutta, S. A Free Lunch in LLM Compression: Revisiting Retraining after Pruning. Advances in Neural Information Processing Systems, 39.
To appear.
Associated with NeurIPS 2026.
[arXiv]
[BibTeX]
@inproceedings{2025_WagnerRouxZimmerPokutta_Llmpruningretraining_2510-14444, booktitle = {Advances in Neural Information Processing Systems}, series = {Advances in Neural Information Processing Systems}, volume = {39}, publisher = {Curran Associates}, note = {To appear}, archiveprefix = {arXiv}, eprint = {2510.14444}, arxiv = {arXiv:2510.14444}, primaryclass = {cs.LG}, author = {Wagner, Moritz and Roux, Christophe and Zimmer, Max and Pokutta, Sebastian}, title = {A Free Lunch in LLM Compression: Revisiting Retraining after Pruning} } - Urbano, A., Romero, D. W., Zimmer, M., and Pokutta, S. (2026, January). RECON: Robust symmetry discovery via Explicit Canonical Orientation Normalization. International Conference on Learning Representations 2026.
Associated with ICLR 2026.
[arXiv]
[BibTeX]
@inproceedings{2025_UrbanoWZimmerPokutta_Reconsymmetrydiscovery, year = {2026}, booktitle = {International Conference on Learning Representations 2026}, publisher = {ICLR (OpenReview)}, month = jan, archiveprefix = {arXiv}, eprint = {2505.13289}, arxiv = {arXiv:2505.13289}, primaryclass = {cs.LG}, author = {Urbano, Alonso and Romero, David W. and Zimmer, Max and Pokutta, Sebastian}, title = {RECON: Robust symmetry discovery via Explicit Canonical Orientation Normalization} } - Pelleriti, N., Spiegel, C., Liu, S., Martínez-Rubio, D., Zimmer, M., and Pokutta, S. (2026). Neural Sum-of-Squares: Certifying the Nonnegativity of Polynomials with Transformers. International Conference on Learning Representations 2026.
Associated with ICLR 2026.
[URL]
[arXiv]
[BibTeX]
@inproceedings{2025_PelleritiEtAl_Neuralsos_2510-13444, year = {2026}, booktitle = {International Conference on Learning Representations 2026}, publisher = {ICLR (OpenReview)}, url = {https://proceedings.iclr.cc/paper_files/paper/2026/hash/1cfb3b066461f1c542be5242a622ce4b-Abstract-Conference.html}, archiveprefix = {arXiv}, eprint = {2510.13444}, arxiv = {arXiv:2510.13444}, primaryclass = {cs.LG}, author = {Pelleriti, Nico and Spiegel, Christoph and Liu, Shiwei and Martínez-Rubio, David and Zimmer, Max and Pokutta, Sebastian}, title = {Neural Sum-of-Squares: Certifying the Nonnegativity of Polynomials with Transformers} } - Kera, H., Pelleriti, N., Ishihara, Y., Zimmer, M., and Pokutta, S. (2025). Computational Algebra with Attention: Transformer Oracles for Border Basis Algorithms. Advances in Neural Information Processing Systems, 38, 10434–10476.
Associated with NeurIPS 2025.
DOI: 10.52202/085713-0319
[arXiv]
[BibTeX]
@inproceedings{2025_KeraEtAl_Transformeroracles_2505-23696, year = {2025}, booktitle = {Advances in Neural Information Processing Systems}, series = {Advances in Neural Information Processing Systems}, volume = {38}, publisher = {Curran Associates}, month = sep, pages = {10434--10476}, doi = {10.52202/085713-0319}, archiveprefix = {arXiv}, eprint = {2505.23696}, arxiv = {arXiv:2505.23696}, primaryclass = {cs.LG}, author = {Kera, Hiroshi and Pelleriti, Nico and Ishihara, Yuki and Zimmer, Max and Pokutta, Sebastian}, title = {Computational Algebra with Attention: Transformer Oracles for Border Basis Algorithms} } - Pauls, J., Zimmer, M., Turan, B., Saatchi, S., Ciais, P., Pokutta, S., and Gieseke, F. (2025). Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation. Proceedings of the 42nd International Conference on Machine Learning, 267.
Associated with ICML 2025.
[arXiv]
[BibTeX]
@inproceedings{2025_JanEtAl_Temporalcanopyheight, year = {2025}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, series = {Proceedings of Machine Learning Research}, volume = {267}, month = may, archiveprefix = {arXiv}, eprint = {2501.19328}, arxiv = {arXiv:2501.19328}, primaryclass = {cs.LG}, author = {Pauls, Jan and Zimmer, Max and Turan, Berkant and Saatchi, Sassan and Ciais, Philippe and Pokutta, Sebastian and Gieseke, Fabian}, title = {Capturing Temporal Dynamics in Large-Scale Canopy Tree Height Estimation} } - Fayad, I., Zimmer, M., Schwartz, M., Gieseke, F., Ciais, P., Belouze, G., Brood, S., De Truchis, A., and d’Aspremont, A. (2025). DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for Earth Observation Applications. Proceedings of the 42nd International Conference on Machine Learning, 267, 16375–16406.
Associated with ICML 2025.
[arXiv]
[BibTeX]
@inproceedings{2025_IbrahimEtAl_DuniaCrossmodalalignment, year = {2025}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, series = {Proceedings of Machine Learning Research}, volume = {267}, pages = {16375--16406}, archiveprefix = {arXiv}, eprint = {2502.17066}, arxiv = {arXiv:2502.17066}, primaryclass = {cs.CV}, author = {Fayad, Ibrahim and Zimmer, Max and Schwartz, Martin and Gieseke, Fabian and Ciais, Philippe and Belouze, Gabriel and Brood, Sarah and De Truchis, Aurelien and d'Aspremont, Alexandre}, title = {DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for Earth Observation Applications} } - Pauls, J., Zimmer, M., Kelly, U. M., Schwartz, M., Saatchi, S., Ciais, P., Pokutta, S., Brandt, M., and Gieseke, F. (2024). Estimating Canopy Height at Scale. Proceedings of the 41st International Conference on Machine Learning, 235.
Associated with ICML 2024.
[arXiv]
[code]
[BibTeX]
@inproceedings{2024_PaulsEtAl_Canopyheightestimation, year = {2024}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, series = {Proceedings of Machine Learning Research}, volume = {235}, month = may, archiveprefix = {arXiv}, eprint = {2406.01076}, arxiv = {arXiv:2406.01076}, primaryclass = {cs.CV}, author = {Pauls, Jan and Zimmer, Max and Kelly, Una M and Schwartz, Martin and Saatchi, Sassan and Ciais, Philippe and Pokutta, Sebastian and Brandt, Martin and Gieseke, Fabian}, title = {Estimating Canopy Height at Scale}, code = {https://github.com/AI4Forest/Global-Canopy-Height-Map} } - Zimmer, M., Spiegel, C., and Pokutta, S. (2024). Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging. International Conference on Learning Representations 2024.
Associated with ICLR 2024.
[URL]
[arXiv]
[BibTeX]
@inproceedings{2023_ZimmerSpiegelPokutta_Sparsemodelsoups, year = {2024}, booktitle = {International Conference on Learning Representations 2024}, publisher = {ICLR (OpenReview)}, url = {https://iclr.cc/virtual/2024/poster/17433}, archiveprefix = {arXiv}, eprint = {2306.16788}, arxiv = {arXiv:2306.16788}, primaryclass = {cs.LG}, author = {Zimmer, Max and Spiegel, Christoph and Pokutta, Sebastian}, title = {Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging} } - Zimmer, M., Spiegel, C., and Pokutta, S. (2023). How I Learned to Stop Worrying and Love Retraining. International Conference on Learning Representations 2023.
Associated with ICLR 2023.
[URL]
[arXiv]
[code]
[BibTeX]
@inproceedings{2021_ZimmerSpiegelPokutta_Retrainingpruning, year = {2023}, booktitle = {International Conference on Learning Representations 2023}, publisher = {ICLR (OpenReview)}, url = {https://iclr.cc/virtual/2023/poster/10914}, archiveprefix = {arXiv}, eprint = {2111.00843}, arxiv = {arXiv:2111.00843}, primaryclass = {cs.LG}, author = {Zimmer, Max and Spiegel, Christoph and Pokutta, Sebastian}, title = {How I Learned to Stop Worrying and Love Retraining}, code = {https://github.com/ZIB-IOL/BIMP} }







