I work at the intersection of physics-based models, satellite imagery, and deep learning, aiming at tackling critical Earth science challenges, such as terrestrial carbon cycle modeling, ecological forecasting, and climate change combatting. George Box famously stated, "All models are wrong, but some models are useful". In this spirit, I study physics-guided/informed deep learning, with a focus on how real observations from satellite imagery can effectively bridge the gap between theoretical physics models and dynamic realities of our planet, revealing unknown sides of Earth. I ground my work in experiments with diverse ecosystem and energy models (e.g. Ecosystem Demography, Mortran) and in mining valuable information from both historical and real-time remote sensing images across the globe.