The paper "On the Foundations of Earth and Climate Foundation Models" by Xiao Xiang Zhu et al. explores the potential of foundation models (FMs) in advancing Earth and climate sciences. The authors define eleven essential features for an ideal Earth and climate FM, emphasizing the need for geolocation embedding, balanced geographic representations, scale awareness, wavelength embeddings, time variables, multisensory capabilities, task-agnosticality, and carbon minimization. They also highlight three highly desirable features: uncertainty quantification, physical consistency, and AI assistants. The paper discusses the current state of Earth observation (EO) and climate FMs, noting that while significant progress has been made, there are still gaps in spatial resolution, data diversity, and the integration of EO with weather and climate data. The authors propose a comprehensive workflow for achieving the ideal FM, including data curation, model design, and training. They emphasize the importance of dynamic encoding, spatiotemporal analysis, multi-modal learning, and physical consistency. The evaluation of Earth FMs is addressed, with the authors calling for standardized benchmarking suites that cover diverse data modalities, representative tasks, and interactions between components. They identify gaps in existing benchmarks, such as the lack of evaluation tasks that integrate EO with weather and climate data, and suggest future research directions, including energy-efficient adaptation and the integration of heterogeneous modalities. Overall, the paper provides a comprehensive overview of the current landscape of Earth and climate FMs, identifies key challenges, and outlines a path forward for their development and application in Earth and climate sciences.The paper "On the Foundations of Earth and Climate Foundation Models" by Xiao Xiang Zhu et al. explores the potential of foundation models (FMs) in advancing Earth and climate sciences. The authors define eleven essential features for an ideal Earth and climate FM, emphasizing the need for geolocation embedding, balanced geographic representations, scale awareness, wavelength embeddings, time variables, multisensory capabilities, task-agnosticality, and carbon minimization. They also highlight three highly desirable features: uncertainty quantification, physical consistency, and AI assistants. The paper discusses the current state of Earth observation (EO) and climate FMs, noting that while significant progress has been made, there are still gaps in spatial resolution, data diversity, and the integration of EO with weather and climate data. The authors propose a comprehensive workflow for achieving the ideal FM, including data curation, model design, and training. They emphasize the importance of dynamic encoding, spatiotemporal analysis, multi-modal learning, and physical consistency. The evaluation of Earth FMs is addressed, with the authors calling for standardized benchmarking suites that cover diverse data modalities, representative tasks, and interactions between components. They identify gaps in existing benchmarks, such as the lack of evaluation tasks that integrate EO with weather and climate data, and suggest future research directions, including energy-efficient adaptation and the integration of heterogeneous modalities. Overall, the paper provides a comprehensive overview of the current landscape of Earth and climate FMs, identifies key challenges, and outlines a path forward for their development and application in Earth and climate sciences.