数十年来的轨道探测任务已获取大量月球多模态遥感数据，涵盖光学影像、光谱学、热辐射、雷达、重力场及元素成分等类型。然而，这些数据仍分散存档于不同机构，且尚无用于评估机器学习模型在月球数据上性能的基准。本文提出 Moonstone，首个面向月球遥感的多模态基础模型基准。我们的贡献包括：（1）构建一个覆盖五项任务、七类仪器的全球月球预训练数据集，包含28个通道、空间分辨率为每度128像素（约237米）；（2）提出 MG-MAE（Modality-Grouped Masked Autoencoder），一种按模态分组的掩码自编码器，其特点包括：针对各模态组设计的卷积标记器（convolutional tokenizers）、共享的 Vision Transformer 编码器、面向缺失模态的注意力掩码（attention masking）、适配异构空间覆盖范围的覆盖率自适应掩码（coverage-adaptive masking），以及保障物理合理重建的光谱连续性正则化（spectral continuity regularization）；（3）构建涵盖分类、回归与分割任务的六项下游任务基准。MG-MAE 预训练特征在全部下游任务上均显著优于从零训练（scratch）基线，并大幅超越 ImageNet 预训练及标准 MAE 基线。数据与代码发布于 https://huggingface.co/datasets/ayushprd/Moonstone 和 https://github.com/ayushprd/Moonstone。

Decades of orbital missions have produced multi-modal remote sensing data for the Moon, spanning optical imagery, spectroscopy, thermal emission, radar, gravity, and elemental composition. Yet these datasets remain fragmented across archives, and no benchmark exists for evaluating machine learning on lunar data. We introduce Moonstone, the first multi-modal foundation model benchmark for lunar remote sensing. Our contributions are: (1) a 28-channel, 128 pixels-per-degree (~237 m) global lunar pretraining dataset from seven instrument families across five missions, (2) MG-MAE, a modality-grouped masked autoencoder with per-group convolutional tokenizers, a shared Vision Transformer encoder, attention masking for missing modalities, coverage-adaptive masking for heterogeneous spatial coverage, and spectral continuity regularization for physically plausible reconstructions, and (3) a benchmark of six downstream tasks covering classification, regression, and segmentation. MG-MAE pretrained features outperform scratch baselines on all tasks and surpass both ImageNet-pretrained and vanilla MAE baselines by large margins. Data and code are available at https://huggingface.co/datasets/ayushprd/Moonstone and https://github.com/ayushprd/Moonstone .