Satellite data represents a distinct modality in machine learning (ML), requiring specialized approaches due to its unique characteristics and challenges. Unlike traditional data modalities such as natural images, satellite data spans logarithmic scales in spatial and temporal dimensions, with varying resolutions, spectral channels, and data volumes. It is collected from diverse sensors, often at high spatial and temporal resolutions, and includes rich spectral information beyond standard RGB. The data volume is massive, with petabytes of information available, and annotations are often sparse, biased, and challenging to obtain. Deployment of satellite-based ML models requires efficient processing of dense predictions, which can be computationally intensive. Evaluation of these models is also complex, as traditional ML practices are not well-suited for the spatio-temporal nature of satellite data. Ethical concerns also arise, as satellite data can enable large-scale monitoring of human activities, raising issues of privacy and fairness. The ML community must recognize satellite data as a distinct modality and develop specialized methods for its analysis. Current approaches that "lift and shift" solutions from other modalities are suboptimal for satellite data. Instead, new algorithms, architectures, and models tailored to satellite data are needed. These include self-supervised learning strategies, rotation-equivariant models, and domain adaptation techniques. Satellite data also offers opportunities for cross-cutting ML research, such as distribution shift, self-supervised learning, and multi-modal learning. The unique properties of satellite data can enrich these areas, providing new datasets, challenges, and contexts for research. To advance SatML, the community must prioritize specialized research, develop inclusive and collaborative practices, and ensure that research benefits both global and local communities. This includes creating benchmarks that reflect real-world conditions, emphasizing practical impact, and addressing ethical concerns. The ML community must also rethink open science practices to ensure that satellite data is used responsibly and ethically, with a focus on accessibility and fairness. By recognizing satellite data as a distinct modality, the ML community can drive innovation and ensure that SatML contributes meaningfully to real-world challenges.Satellite data represents a distinct modality in machine learning (ML), requiring specialized approaches due to its unique characteristics and challenges. Unlike traditional data modalities such as natural images, satellite data spans logarithmic scales in spatial and temporal dimensions, with varying resolutions, spectral channels, and data volumes. It is collected from diverse sensors, often at high spatial and temporal resolutions, and includes rich spectral information beyond standard RGB. The data volume is massive, with petabytes of information available, and annotations are often sparse, biased, and challenging to obtain. Deployment of satellite-based ML models requires efficient processing of dense predictions, which can be computationally intensive. Evaluation of these models is also complex, as traditional ML practices are not well-suited for the spatio-temporal nature of satellite data. Ethical concerns also arise, as satellite data can enable large-scale monitoring of human activities, raising issues of privacy and fairness. The ML community must recognize satellite data as a distinct modality and develop specialized methods for its analysis. Current approaches that "lift and shift" solutions from other modalities are suboptimal for satellite data. Instead, new algorithms, architectures, and models tailored to satellite data are needed. These include self-supervised learning strategies, rotation-equivariant models, and domain adaptation techniques. Satellite data also offers opportunities for cross-cutting ML research, such as distribution shift, self-supervised learning, and multi-modal learning. The unique properties of satellite data can enrich these areas, providing new datasets, challenges, and contexts for research. To advance SatML, the community must prioritize specialized research, develop inclusive and collaborative practices, and ensure that research benefits both global and local communities. This includes creating benchmarks that reflect real-world conditions, emphasizing practical impact, and addressing ethical concerns. The ML community must also rethink open science practices to ensure that satellite data is used responsibly and ethically, with a focus on accessibility and fairness. By recognizing satellite data as a distinct modality, the ML community can drive innovation and ensure that SatML contributes meaningfully to real-world challenges.