arXiv:2511.21194v1 Announce Type: new 
Abstract: Foundation models have demonstrated a remarkable ability to learn rich, transferable representations across diverse modalities such as images, text, and audio. In modern machine learning pipelines, these representations often replace raw data as the primary input for downstream tasks. In this paper, we address the challenge of adapting a pre-trained foundation model to inject domain-specific knowledge, without retraining from scratch or incurring significant computational costs. To this end, we introduce BotaCLIP, a lightweight multimodal contrastive framework that adapts a pre-trained Earth Observation foundation model (DOFA) by aligning high-resolution aerial imagery with botanical relev\'es. Unlike generic embeddings, BotaCLIP internalizes ecological structure through contrastive learning with a regularization strategy that mitigates catastrophic forgetting. Once trained, the resulting embeddings serve as transferable representations for downstream predictors. Motivated by real-world applications in biodiversity modeling, we evaluated BotaCLIP representations in three ecological tasks: plant presence prediction, butterfly occurrence modeling, and soil trophic group abundance estimation. The results showed consistent improvements over those derived from DOFA and supervised baselines. More broadly, this work illustrates how domain-aware adaptation of foundation models can inject expert knowledge into data-scarce settings, enabling frugal representation learning.

تم تقديم BotaCLIP كإطار خفيف الوزن للتعلم المتباين متعدد الوسائط يقوم بتكييف نموذج مؤسس مسبق التدريب لمراقبة الأرض (DOFA) من خلال محاذاة الصور الجوية عالية الدقة مع الأهمية النباتية، مما يعالج تحدي دمج المعرفة الخاصة بالمجال دون إعادة تدريب شاملة.

Se ha presentado BotaCLIP como un marco contrastivo multimodal ligero que adapta un modelo de fundación de Observación de la Tierra preentrenado (DOFA) alineando imágenes aéreas de alta resolución con relevancia botánica, abordando el desafío de integrar conocimientos específicos del dominio sin un extenso reentrenamiento.

BotaCLIP a été introduit comme un cadre contrastif multimodal léger qui adapte un modèle de fondation d'observation de la Terre pré-entraîné (DOFA) en alignant des images aériennes haute résolution avec des éléments botaniques pertinents, répondant au défi d'intégrer des connaissances spécifiques au domaine sans retraining extensif.

BotaCLIP has been introduced as a lightweight multimodal contrastive framework that adapts a pre-trained Earth Observation foundation model (DOFA) by aligning high-resolution aerial imagery with botanical relevance, addressing the challenge of integrating domain-specific knowledge without extensive retraining.

BotaCLIP: Contrastive Learning for Botany-Aware Representation of Earth Observation Data

arXiv:2511.21523v2 Announce Type: replace 
Abstract: Recent advances in foundation models have shown great promise in domains such as natural language processing and computer vision, and similar efforts are now emerging in the Earth Observation community. These models aim to generalize across tasks with limited supervision, reducing the need for training separate models for each task. However, current strategies, which largely focus on scaling model size and dataset volume, require prohibitive computational and data resources, limiting accessibility to only a few large institutions. Moreover, this paradigm of ever-larger models stands in stark contrast with the principles of sustainable and environmentally responsible AI, as it leads to immense carbon footprints and resource inefficiency. In this work, we present a novel and efficient alternative: an Ensemble-of-Specialists framework for building Remote Sensing Foundation Models (RSFMs). Our method decomposes the training process into lightweight, task-specific ConvNeXtV2 specialists that can be frozen and reused. This modular approach offers strong advantages in efficiency, interpretability, and extensibility. Moreover, it naturally supports federated training, pruning, and continuous specialist integration, making it particularly well-suited for collaborative and resource-constrained settings. Our framework sets a new direction for building scalable and efficient RSFMs. All codes and pretrained models are available at https://github.com/pierreadorni/EoS-FM.

أدت التطورات الأخيرة في مراقبة الأرض إلى تطوير إطار عمل Ensemble-of-Specialists، الذي يهدف إلى إنشاء نماذج أساسية في الاستشعار عن بعد (RSFMs) يمكن أن تعمم عبر المهام مع إشراف محدود. يتناقض هذا النهج مع الاتجاه الحالي لزيادة حجم النموذج، والذي يتطلب موارد كبيرة وغير مستدام بيئيًا.

Los avances recientes en la Observación de la Tierra han llevado al desarrollo del marco Ensemble-of-Specialists, que busca crear Modelos de Fundación en Teledetección (RSFMs) que se generalicen a través de tareas con supervisión limitada. Este enfoque contrasta con la tendencia actual de aumentar el tamaño del modelo, que consume muchos recursos y no es sostenible ambientalmente.

Des avancées récentes dans l'observation de la Terre ont conduit au développement du cadre Ensemble-of-Specialists, qui vise à créer des modèles de fondation en télédétection (RSFMs) capables de se généraliser à travers des tâches avec une supervision limitée. Cette approche contraste avec la tendance actuelle à l'augmentation de la taille des modèles, qui est gourmande en ressources et non durable sur le plan environnemental.

Recent advancements in Earth Observation have led to the development of the Ensemble-of-Specialists framework, which aims to create Remote Sensing Foundation Models (RSFMs) that generalize across tasks with limited supervision. This approach contrasts with the current trend of scaling model size, which is resource-intensive and environmentally unsustainable.

BotaCLIP: Contrastive Learning for Botany-Aware Representation of Earth Observation Data

Was this article worth reading? Share it

LucidQuery AI

Airparser

Attentive AI