The increasing frequency and intensity of wildfires pose severe threats to ecosystems, infrastructure, and human lives. Rapid and reliable wildfire monitoring systems are essential for effective disaster response and mitigation. This PhD project explores the development of foundation models for Earth observation, leveraging large-scale satellite imagery and recent advances in generative machine learning. Focusing on diffusion-based models, the research aims to build flexible and scalable architectures capable of generalizing across regions and sensor modalities for critical tasks such as wildfire detection, change localization, and damage assessment. By pretraining on diverse satellite datasets and fine-tuning on downstream tasks, the project investigates how generative modeling can bridge domain gaps and enhance performance in dynamic and data-scarce environments. The work contributes to the growing field of foundation models in remote sensing and sets the groundwork for more robust, generalizable, and adaptive geospatial AI systems.