Thomas Brunschwiler

Hendrik F. Hamann, Thomas Brunschwiler, Blazhe Gjorgiev et al.

Foundation models (FMs) currently dominate news headlines. They employ advanced deep learning architectures to extract structural information autonomously from vast datasets through self-supervision. The resulting rich representations of complex systems and dynamics can be applied to many downstream applications. Therefore, FMs can find uses in electric power grids, challenged by the energy transition and climate change. In this paper, we call for the development of, and state why we believe in, the potential of FMs for electric grids. We highlight their strengths and weaknesses amidst the challenges of a changing grid. We argue that an FM learning from diverse grid data and topologies could unlock transformative capabilities, pioneering a new approach in leveraging AI to redefine how we manage complexity and uncertainty in the electric grid. Finally, we discuss a power grid FM concept, namely GridFM, based on graph neural networks and show how different downstream tasks benefit.

Johannes Jakubik, Michal Muszynski, Michael Vössing et al.

Climate change results in an increased probability of extreme weather events that put societies and businesses at risk on a global scale. Therefore, near real-time mapping of natural hazards is an emerging priority for the support of natural disaster relief, risk management, and informing governmental policy decisions. Recent methods to achieve near real-time mapping increasingly leverage deep learning (DL). However, DL-based approaches are designed for one specific task in a single geographic region based on specific frequency bands of satellite data. Therefore, DL models used to map specific natural hazards struggle with their generalization to other types of natural hazards in unseen regions. In this work, we propose a methodology to significantly improve the generalizability of DL natural hazards mappers based on pre-training on a suitable pre-task. Without access to any data from the target domain, we demonstrate this improved generalizability across four U-Net architectures for the segmentation of unseen natural hazards. Importantly, our method is invariant to geographic differences and differences in the type of frequency bands of satellite data. By leveraging characteristics of unlabeled images from the target domain that are publicly available, our approach is able to further improve the generalization behavior without fine-tuning. Thereby, our approach supports the development of foundation models for earth monitoring with the objective of directly segmenting unseen natural hazards across novel geographic regions given different sources of satellite imagery.

Alban Puech, Matteo Mazzonelli, Celia Cintas et al.

We introduce gridfm-datakit-v1, a Python library for generating realistic and diverse Power Flow (PF) and Optimal Power Flow (OPF) datasets for training Machine Learning (ML) solvers. Existing datasets and libraries face three main challenges: (1) lack of realistic stochastic load and topology perturbations, limiting scenario diversity; (2) PF datasets are restricted to OPF-feasible points, hindering generalization of ML solvers to cases that violate operating limits (e.g., branch overloads or voltage violations); and (3) OPF datasets use fixed generator cost functions, limiting generalization across varying costs. gridfm-datakit addresses these challenges by: (1) combining global load scaling from real-world profiles with localized noise and supporting arbitrary N-k topology perturbations to create diverse yet realistic datasets; (2) generating PF samples beyond operating limits; and (3) producing OPF data with varying generator costs. It also scales efficiently to large grids (up to 10,000 buses). Comparisons with OPFData, OPF-Learn, PGLearn, and PF$Δ$ are provided. Available on GitHub at https://github.com/gridfm/gridfm-datakit under Apache 2.0 and via `pip install gridfm-datakit`.

Johannes Jakubik, Felix Yang, Benedikt Blumenstiel et al.

We present TerraMind, the first any-to-any generative, multimodal foundation model for Earth observation (EO). Unlike other multimodal models, TerraMind is pretrained on dual-scale representations combining both token-level and pixel-level data across modalities. On a token level, TerraMind encodes high-level contextual information to learn cross-modal relationships, while on a pixel level, TerraMind leverages fine-grained representations to capture critical spatial nuances. We pretrained TerraMind on nine geospatial modalities of a global, large-scale dataset. In this paper, we demonstrate that (i) TerraMind's dual-scale early fusion approach unlocks a range of zero-shot and few-shot applications for Earth observation, (ii) TerraMind introduces "Thinking-in-Modalities" (TiM) -- the capability of generating additional artificial data during finetuning and inference to improve the model output -- and (iii) TerraMind achieves beyond state-of-the-art performance in community-standard benchmarks for EO like PANGAEA. The pretraining dataset, the model weights, and our code are open-sourced under a permissive license.

Benedikt Blumenstiel, Viktoria Moor, Romeo Kienzler et al.

Image retrieval enables an efficient search through vast amounts of satellite imagery and returns similar images to a query. Deep learning models can identify images across various semantic concepts without the need for annotations. This work proposes to use Geospatial Foundation Models, like Prithvi, for remote sensing image retrieval with multiple benefits: i) the models encode multi-spectral satellite data and ii) generalize without further fine-tuning. We introduce two datasets to the retrieval task and observe a strong performance: Prithvi processes six bands and achieves a mean Average Precision of 97.62% on BigEarthNet-43 and 44.51% on ForestNet-12, outperforming other RGB-based models. Further, we evaluate three compression methods with binarized embeddings balancing retrieval speed and accuracy. They match the retrieval speed of much shorter hash codes while maintaining the same accuracy as floating-point embeddings but with a 32-fold compression. The code is available at https://github.com/IBM/remote-sensing-image-retrieval.

Benedikt Blumenstiel, Paolo Fraccaro, Valerio Marsocci et al.

Large-scale foundation models in Earth Observation can learn versatile, label-efficient representations by leveraging massive amounts of unlabeled data. However, existing public datasets are often limited in scale, geographic coverage, or sensor variety. We introduce TerraMesh, a new globally diverse, multimodal dataset combining optical, synthetic aperture radar, elevation, and land-cover modalities in an Analysis-Ready Data format. TerraMesh includes over 9~million samples with eight spatiotemporal aligned modalities, enabling large-scale pre-training. We provide detailed data processing steps, comprehensive statistics, and empirical evidence demonstrating improved model performance when pre-trained on TerraMesh. The dataset is hosted at https://huggingface.co/datasets/ibm-esa-geospatial/TerraMesh.

Clive Tinashe Marimo, Benedikt Blumenstiel, Maximilian Nitsche et al.

Vision-language models for Earth observation (EO) typically rely on the visual spectrum of data as the only model input, thus failing to leverage the rich spectral information available in the multispectral channels recorded by satellites. Therefore, we introduce Llama3-MS-CLIP, the first vision-language model pre-trained with contrastive learning on a large-scale multispectral dataset and report on the performance gains due to the extended spectral range. Furthermore, we present the largest-to-date image-caption dataset for multispectral data, consisting of one million Sentinel-2 samples and corresponding textual descriptions generated using Llama3-LLaVA-Next and Overture Maps data. We develop a scalable captioning pipeline, which is validated by domain experts. We evaluate Llama3-MS-CLIP on multispectral zero-shot image classification and retrieval using three datasets of varying complexity. Our results demonstrate that Llama3-MS-CLIP significantly outperforms other RGB-based approaches, improving classification accuracy by +6.77% on average and retrieval performance by +4.63% mAP compared to the second-best model. Our results emphasize the relevance of multispectral vision-language learning. The image-caption dataset, code, and model weights are available at https://github.com/IBM/MS-CLIP.

Alban Puech, Jonas Weiss, Thomas Brunschwiler et al.

The energy transition, crucial for tackling the climate crisis, demands integrating numerous distributed, renewable energy sources into existing grids. Along with climate change and consumer behavioral changes, this leads to changes and variability in generation and load patterns, introducing significant complexity and uncertainty into grid planning and operations. While the industry has already started to exploit AI to overcome computational challenges of established grid simulation tools, we propose the use of AI Foundation Models (FMs) and advances in Graph Neural Networks to efficiently exploit poorly available grid data for different downstream tasks, enhancing grid operations. For capturing the grid's underlying physics, we believe that building a self-supervised model learning the power flow dynamics is a critical first step towards developing an FM for the power grid. We show how this approach may close the gap between the industry needs and current grid analysis capabilities, to bring the industry closer to optimal grid operation and planning.

Alban Puech, François Mirallès, Jonas Weiss et al.

Quasi-static time series (QSTS) simulations have great potential for evaluating the grid's ability to accommodate the large-scale integration of distributed energy resources. However, as grids expand and operate closer to their limits, iterative power flow solvers, central to QSTS simulations, become computationally prohibitive and face increasing convergence issues. Neural power flow solvers provide a promising alternative, speeding up power flow computations by 3 to 4 orders of magnitude, though they are costly to train. In this paper, we envision how recently introduced grid foundation models could improve the economic viability of neural power flow solvers. Conceptually, these models amortize training costs by serving as a foundation for a range of grid operation and planning tasks beyond power flow solving, with only minimal fine-tuning required. We call for collaboration between the AI and power grid communities to develop and open-source these models, enabling all operators, even those with limited resources, to benefit from AI without building solutions from scratch.

Luis Gilch, Isabelle Wittmann, Maximilian Nitsche et al.

Earth observation (EO) missions produce petabytes of multispectral imagery, increasingly analyzed using large Geospatial Foundation Models (GeoFMs). Alongside end-to-end adaptation, workflows make growing use of intermediate representations as task-agnostic embeddings, enabling models to compute representations once and reuse them across downstream tasks. Consequently, when GeoFMs act as feature extractors, decisions about how representations are obtained, aggregated, and combined affect downstream performance and pipeline scalability. Understanding these trade-offs is essential for scalable embedding-based EO workflows, where compact embeddings can replace raw data while remaining broadly useful. We present a systematic analysis of embedding design in GeoFM-based EO workflows. Leveraging NeuCo-Bench, we study how backbone architecture, pretraining strategy, representation depth, spatial aggregation, and representation combination influence EO task performance. We demonstrate the usability of GeoFM embeddings by aggregating them into fixed-size representations more than 500x smaller than the raw input data. Across models, we find consistent trends: transformer backbones with mean pooling provide strong default embeddings, intermediate ResNet layers can outperform final layers, self-supervised objectives exhibit task-specific strengths, and combining embeddings from different objectives often improves robustness.

Carlos Gomes, Isabelle Wittmann, Damien Robert et al.

Over the past decades, there has been an explosion in the amount of available Earth Observation (EO) data. The unprecedented coverage of the Earth's surface and atmosphere by satellite imagery has resulted in large volumes of data that must be transmitted to ground stations, stored in data centers, and distributed to end users. Modern Earth System Models (ESMs) face similar challenges, operating at high spatial and temporal resolutions, producing petabytes of data per simulated day. Data compression has gained relevance over the past decade, with neural compression (NC) emerging from deep learning and information theory, making EO data and ESM outputs ideal candidates due to their abundance of unlabeled data. In this review, we outline recent developments in NC applied to geospatial data. We introduce the fundamental concepts of NC including seminal works in its traditional applications to image and video compression domains with focus on lossy compression. We discuss the unique characteristics of EO and ESM data, contrasting them with "natural images", and explain the additional challenges and opportunities they present. Moreover, we review current applications of NC across various EO modalities and explore the limited efforts in ESM compression to date. The advent of self-supervised learning (SSL) and foundation models (FM) has advanced methods to efficiently distill representations from vast unlabeled data. We connect these developments to NC for EO, highlighting the similarities between the two fields and elaborate on the potential of transferring compressed feature representations for machine--to--machine communication. Based on insights drawn from this review, we devise future directions relevant to applications in EO and ESM.

Ruben Gonzalez Avilés, Linus Scheibenreif, Nassim Ait Ali Braham et al.

Hyperspectral imaging provides detailed spectral information and holds significant potential for monitoring of greenhouse gases (GHGs). However, its application is constrained by limited spatial coverage and infrequent revisit times. In contrast, multispectral imaging offers broader spatial and temporal coverage but often lacks the spectral detail that can enhance GHG detection. To address these challenges, this study proposes a spectral transformer model that synthesizes hyperspectral data from multispectral inputs. The model is pre-trained via a band-wise masked autoencoder and subsequently fine-tuned on spatio-temporally aligned multispectral-hyperspectral image pairs. The resulting synthetic hyperspectral data retain the spatial and temporal benefits of multispectral imagery and improve GHG prediction accuracy relative to using multispectral data alone. This approach effectively bridges the trade-off between spectral resolution and coverage, highlighting its potential to advance atmospheric monitoring by combining the strengths of hyperspectral and multispectral systems with self-supervised deep learning.

Carlos Gomes, Thomas Brunschwiler

As repositories of large scale data in earth observation (EO) have grown, so have transfer and storage costs for model training and inference, expending significant resources. We introduce Neural Embedding Compression (NEC), based on the transfer of compressed embeddings to data consumers instead of raw data. We adapt foundation models (FM) through learned neural compression to generate multi-task embeddings while navigating the tradeoff between compression rate and embedding utility. We update only a small fraction of the FM parameters (10%) for a short training period (1% of the iterations of pre-training). We evaluate NEC on two EO tasks: scene classification and semantic segmentation. Compared with applying traditional compression to the raw data, NEC achieves similar accuracy with a 75% to 90% reduction in data. Even at 99.7% compression, performance drops by only 5% on the scene classification task. Overall, NEC is a data-efficient yet performant approach for multi-task EO modelling.

Gianfranco Basile, Johannes Jakubik, Benedikt Blumenstiel et al.

We propose a task-agnostic framework for multimodal fusion of time series and single timestamp images, enabling cross-modal generation and robust downstream performance. Our approach explores deterministic and learned strategies for time series quantization and then leverages a masked correlation learning objective, aligning discrete image and time series tokens in a unified representation space. Instantiated in the Earth observation domain, the pretrained model generates consistent global temperature profiles from satellite imagery and is validated through counterfactual experiments. Across downstream tasks, our task-agnostic pretraining outperforms task-specific fusion by 6% in R^2 and 2% in RMSE on average, and exceeds baseline methods by 50% in R^2 and 12% in RMSE. Finally, we analyze gradient sensitivity across modalities, providing insights into model robustness. Code, data, and weights will be released under a permissive license.

Jannis Fleckenstein, David Kreismann, Tamara Rosemary Govindasamy et al. · ibm-research

As urbanization and climate change progress, urban heat island effects are becoming more frequent and severe. To formulate effective mitigation plans, cities require detailed air temperature data, yet conventional machine learning models with limited data often produce inaccurate predictions, particularly in underserved areas. Geospatial foundation models trained on global unstructured data offer a promising alternative by demonstrating strong generalization and requiring only minimal fine-tuning. In this study, an empirical ground truth of urban heat patterns is established by quantifying cooling effects from green spaces and benchmarking them against model predictions to evaluate the model's accuracy. The foundation model is subsequently fine-tuned to predict land surface temperatures under future climate scenarios, and its practical value is demonstrated through a simulated inpainting that highlights its role for mitigation support. The results indicate that foundation models offer a powerful way for evaluating urban heat island mitigation strategies in data-scarce regions to support more climate-resilient cities.

Rikard Vinge, Isabelle Wittmann, Jannik Schneider et al.

We introduce NeuCo-Bench, a novel benchmark framework for evaluating (lossy) neural compression and representation learning in the context of Earth Observation (EO). Our approach builds on fixed-size embeddings that act as compact, task-agnostic representations applicable to a broad range of downstream tasks. NeuCo-Bench comprises three core components: (i) an evaluation pipeline built around reusable embeddings, (ii) a new challenge mode with a hidden-task leaderboard designed to mitigate pretraining bias, and (iii) a scoring system that balances accuracy and stability. To support reproducibility, we release SSL4EO-S12-downstream, a curated multispectral, multitemporal EO dataset. We present initial results from a public challenge at the 2025 CVPR EARTHVISION workshop and conduct ablations with state-of-the-art foundation models. NeuCo-Bench provides a first step towards community-driven, standardized evaluation of neural embeddings for EO and beyond.

Ruben Gonzalez, Conrad M Albrecht, Nassim Ait Ali Braham et al.

Hyperspectral imaging provides detailed spectral information, offering significant potential for monitoring greenhouse gases like CH4 and NO2. However, its application is constrained by limited spatial coverage and infrequent revisit times. In contrast, multispectral imaging delivers broader spatial and temporal coverage but lacks the spectral granularity required for precise GHG detection. To address these challenges, this study proposes Spectral and Spatial-Spectral transformer models that reconstruct hyperspectral data from multispectral inputs. The models in this paper are pretrained on EnMAP and EMIT datasets and fine-tuned on spatio-temporally aligned (Sentinel-2, EnMAP) and (HLS-S30, EMIT) image pairs respectively. Our model has the potential to enhance atmospheric monitoring by combining the strengths of hyperspectral and multispectral imaging systems.

Maximilian Nitsche, S. Karthik Mukkavilli, Niklas Kühl et al.

We explore the implementation of deep learning techniques for precise building damage assessment in the context of natural hazards, utilizing remote sensing data. The xBD dataset, comprising diverse disaster events from across the globe, serves as the primary focus, facilitating the evaluation of deep learning models. We tackle the challenges of generalization to novel disasters and regions while accounting for the influence of low-quality and noisy labels inherent in natural hazard data. Furthermore, our investigation quantitatively establishes that the minimum satellite imagery resolution essential for effective building damage detection is 3 meters and below 1 meter for classification using symmetric and asymmetric resolution perturbation analyses. To achieve robust and accurate evaluations of building damage detection and classification, we evaluated different deep learning models with residual, squeeze and excitation, and dual path network backbones, as well as ensemble techniques. Overall, the U-Net Siamese network ensemble with F-1 score of 0.812 performed the best against the xView2 challenge benchmark. Additionally, we evaluate a Universal model trained on all hazards against a flood expert model and investigate generalization gaps across events, and out of distribution from field data in the Ahr Valley. Our research findings showcase the potential and limitations of advanced AI solutions in enhancing the impact assessment of climate change-induced extreme weather events, such as floods and hurricanes. These insights have implications for disaster impact assessment in the face of escalating climate challenges.

Anirban Das, Thomas Brunschwiler

Federated Learning enables training of a general model through edge devices without sending raw data to the cloud. Hence, this approach is attractive for digital health applications, where data is sourced through edge devices and users care about privacy. Here, we report on the feasibility to train deep neural networks on the Raspberry Pi4s as edge devices. A CNN, a LSTM and a MLP were successfully trained on the MNIST data-set. Further, federated learning is demonstrated experimentally on IID and non-IID samples in a parametric study, to benchmark the model convergence. The weight updates from the workers are shared with the cloud to train the general model through federated learning. With the CNN and the non-IID samples a test-accuracy of up to 85% could be achieved within a training time of 2 minutes, while exchanging less than $10$ MB data per device. In addition, we discuss federated learning from an use-case standpoint, elaborating on privacy risks and labeling requirements for the application of emotion detection from sound. Based on the experimental findings, we discuss possible research directions to improve model and system performance. Finally, we provide best practices for a practitioner, considering the implementation of federated learning.