NASA 与 IBM 开源月球基础模型：用 17 年轨道器数据构建月球科学基座

NASA-IBM 月球基础模型将几十年的月球观测数据转化为可用于机器学习的资源。它在预测极地冰存量和探测陨石坑方面特别强大。

“NASA 花了几十年时间建立了月球的非凡科学记录，但数据收集只是工作的一部分，”NASA 首席科学数据官 Kevin Murphy 说。他补充道，这些数据还必须更容易被科学家使用。NASA 与 IBM 研究院以及多所学术机构合作，现在发布了 NASA-IBM 月球基础模型。这两个组织称其为月球科学领域首批开源基础模型之一。

不同于针对特定任务的算法，基础模型是在大量未标记的数据上进行预训练的，然后只需少量标记样本即可适应特定任务。团队认为，这是月球研究的主要优势，因为观测数据丰富，但标注有限。

团队使用 SomBench 从零开始训练了该模型，他们称它是迄今为止最大的共注册多模态月球语料库。它包含了近 200 万个瓦片数据包，涵盖 11 种模态和两个空间尺度。约 100 万张高分辨率图像来自狭角相机（NAC），分辨率约为每像素 1 米，而略低于 96.4 万张多光谱图像来自广角相机，分辨率为每像素 100 米。

大部分数据来自月球勘测轨道器（Lunar Reconnaissance Orbiter，LRO）17 年的观测：https://science.nasa.gov/mission/lunar-reconnaissance-orbiter/。根据 NASA 的说法，其数据量超过所有其他 NASA 行星任务数据的总和。来自 GRAIL 任务（重力场）、Lunar Prospector（氢）以及 JAXA 的月亮/SELENE 探测器（矿物学）的数据补充了该数据集。

总体而言，该数据集汇集了来自 9 个仪器和 4 个任务的 30 多个空间对齐的数据层。为了防止训练、验证和测试数据之间的泄漏，语料库按地图分区进行地理划分，而不是随机分配单个瓦片。

该模型基于 TerraMind：https://arxiv.org/abs/2504.11171，一种多模态地球观测模型，但研究人员是从零开始训练的，而不是进行微调。对于每个图块，模型接收成像几何作为明确的上下文，包括照明角度、太阳位置和图块范围。在月球上，照明几何比地表的实际属性更大程度地决定了表面外观。研究团队并没有强迫模型从原始像素中纠正这一点，而是直接将这些信息输入模型。

该模型还在单次训练过程中同时学习高分辨率和低分辨率图像，同时捕捉细节和大尺度上下文。一种称为FlexiViT的技术使同一个训练好的模型能够适应不同图像块大小的任务而无需重新训练。

团队在100米和1米尺度的撞击坑检测以及极地冰沉积预测上测试了该模型。它还需要分割不规则海洋斑块（IMP），这些年轻的火山特征挑战了既定的月球降温时间表。根据技术报告：https://huggingface.co/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model/blob/main/NI_LFM_Technical_Report.pdf，在所有任务中，预训练模型都与常用基线模型和具有随机初始化的结构相同控制模型表现相当甚至更好。

最大的提升出现在冰沉积预测上。永远处于阴影的极地区域足够寒冷，能够保存水冰数十亿年，并被认为是水、氧气和火箭燃料的潜在资源。根据IBM，该模型相比最佳基线SwinV2-B将预测误差降低了多达22%。

在粗尺度撞击坑检测中，该模型比SwinV2-B高出近19%，根据IBM的数据。这一数字是通过仅使用一半数据训练得到的，表明该模型需要更少的标注样本就能表现良好。

即使是随机初始化的控制模型在没有任何月球预训练的情况下，也在冰预测任务中超过了七个基线中的五个。根据研究人员的说法，部分优势来自模型处理不同数据类型的方式。它为每个数据层分配自己的处理路径，而基线模型则将所有输入视为堆叠的通道。

在米级陨石坑检测和IMP分割方面，该模型大致与最强基线持平，差异在训练运行间的方差范围内。IBM声称在IMP上比SwinV2-B领先3个百分点，但结果看起来更像是可比，而不是明显更好。

研究人员表示，LoRA，一种轻量级微调方法，只训练一部分参数，在各方面都能跟上全面微调的步伐，并在陨石坑检测上表现更好。全面微调仅在两个最小的任务上获胜。

IBM欧洲研究院院长Juan Bernabé-Moreno表示，该模型能够跨仪器连接观测数据，并揭示孤立情况下难以发现的模式。

研究显示，这种模型不适合用于绝对大地测量定位。在生成测试中，经度和纬度在某些情况下偏差达数十度，即使地形形状重建正确，海拔结构的绝对高度值仍可能出现偏移。

作者将其视为可重复使用的下游任务基础，而不是物理测量仪器的替代品。用于隔离每项创新贡献的受控实验仍在进行中，一些测试数据集规模较小。

该模型可在 Hugging Face 上公开获取：https://huggingface.co/collections/nasa-ibm-ai4science/nasa-ibm-lunar-fm-and-downstream-models，代码在 GitHub 上：https://github.com/NASA-IMPACT/NASA-IBM-Lunar-Foundation-Model，并且已集成到开源工具包 TerraTorch 中。团队还发布了适用于机器学习的预训练数据集和基准集合。

该模型是NASA-IBM“科学人工智能”合作的一部分。自2022年初开始，双方在《太空法案协议》下一直致力于基础模型的研发。2023年8月，他们在Hugging Face上发布了首个Prithvi模型：https://the-decoder.com/ibm-and-nasa-publish-ai-model-for-climate-research/，该模型基于全美本土连片地区的Landsat和Sentinel-2影像训练，并适用于洪水和野火地图绘制。

IBM 宣布了 Prithvi，作为其 Watsonx 平台的一部分：https://the-decoder.com/watsonx-ibm-unveils-new-enterprise-platform-for-ai/。IBM 于 2025 年与欧洲航天局（ESA）和尤里希研究中心（Forschungszentrum Jülich）合作开发了 TerraMind，即月球模型的基础，用于地球观测。

谷歌 DeepMind 正在通过 AlphaEarth Foundations 追求类似的方法：https://the-decoder.com/google-deepmind-unveils-an-ai-model-that-acts-as-a-virtual-satellite-for-mapping-the-entire-planet/，将光学卫星图像、雷达、激光雷达扫描和气候模拟压缩成地球表面的紧凑嵌入。

保持对人工智能的关注。清晰、有用，无冗余信息。

关注 The Decoder 获取人工智能新闻、背景故事和专家分析。

解码器：https://the-decoder.com/

The NASA-IBM Lunar Foundation Model makes decades of lunar observation data usable for machine learning. It's especially strong at predicting ice deposits at the poles and detecting craters.

"NASA has spent decades building an extraordinary scientific record of the Moon, but collecting data is only part of the job," says Kevin Murphy, NASA's chief science data officer. The data also has to be easier for scientists to use, he adds. NASA and IBM Research, working with several academic institutions, have now released the NASA-IBM Lunar Foundation Model. The two organizations describe it as one of the first open-source foundation models for lunar science.

Unlike task-specific algorithms, a foundation model is pretrained on large volumes of unlabeled data and can then be adapted to specific tasks with just a few labeled examples. The team sees this as the main benefit for lunar research, where observation data is plentiful but labels are scarce. Ad

The team trained the model from scratch using SomBench, which they say is the largest co-registered multimodal lunar corpus to date. It contains nearly 2 million tile bundles across 11 modalities and two spatial scales. About 1 million high-resolution images come from the Narrow Angle Camera (NAC) at roughly 1 meter per pixel, while just under 964,000 multispectral images come from the Wide Angle Camera at 100 meters per pixel. Ad

The bulk of the data comes from 17 years of observations by the Lunar Reconnaissance Orbiter：https://science.nasa.gov/mission/lunar-reconnaissance-orbiter/ (LRO). According to NASA, its data volume exceeds that of all other NASA planetary missions combined. Data from the GRAIL mission (gravity field), Lunar Prospector (hydrogen), and JAXA's Kaguya/SELENE probe (mineralogy) rounded out the collection.

In total, the dataset brings together more than 30 spatially aligned data layers from nine instruments and four missions. To prevent leakage between training, validation, and test data, the corpus is split geographically by map zones rather than randomly distributing individual tiles. Ad

The model is based on TerraMind：https://arxiv.org/abs/2504.11171, a multimodal Earth observation model, but the researchers trained it from scratch instead of fine-tuning it. For each tile, the model receives the imaging geometry as explicit context, including illumination angles, sun position, and tile extent. On the Moon, lighting geometry shapes how the surface looks far more than the surface's actual properties do. Rather than forcing the model to correct for this from raw pixels, the team simply feeds it that information.

The model also learns from high-resolution and coarse imagery together in a single training run, capturing both fine details and large-scale context. A technique called FlexiViT lets the same trained model adapt to tasks with different image patch sizes without retraining. Ad

The team tested the model on crater detection at 100-meter and 1-meter scales and on predicting polar ice deposits. It also had to segment Irregular Mare Patches (IMPs), young volcanic features that challenge established timelines of lunar cooling. Across all tasks, the pretrained model matched or beat both common baselines and an architecturally identical control model with random initialization, according to the technical report：https://huggingface.co/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model/blob/main/NI_LFM_Technical_Report.pdf. Ad

The biggest gains showed up in ice deposit prediction. Permanently shadowed polar regions stay cold enough to preserve water ice for billions of years and are considered a potential resource for water, oxygen, and rocket fuel. The model cut prediction error by up to 22 percent compared to the best baseline, SwinV2-B, according to IBM.

For coarse-scale crater detection, the model beat SwinV2-B by nearly 19 percent, according to IBM. That number comes from training with only half the data, suggesting the model needs fewer labeled examples to perform well.

Even the randomly initialized control model beat five of seven baselines on ice prediction without any lunar pretraining. According to the researchers, part of the advantage comes from how the model handles different data types. It assigns each data layer its own processing path, while baseline models treat all inputs as stacked channels.

On meter-scale crater detection and IMP segmentation, the model roughly ties the strongest baselines, with differences falling within the variance between training runs. IBM claims a 3 percent lead over SwinV2-B on IMPs, but the results look more comparable than clearly better.

According to the researchers, LoRA, a lighter fine-tuning method that trains only a fraction of the parameters, kept pace with full fine-tuning across the board and did better on crater detection. Full fine-tuning only wins on the two smallest tasks.

Juan Bernabé-Moreno, director of IBM Research Europe, said the model connects observations across instruments and reveals patterns that are difficult to spot in isolation.

The model isn't suited for absolute geodetic positioning, the research shows. In generation tests, latitude and longitude were off by dozens of degrees in some cases, and elevation structures could appear with shifted absolute height values even when their shapes were reconstructed correctly.

The authors see it as a reusable foundation for downstream tasks, not a replacement for physical measurement instruments. Controlled experiments isolating the contribution of each innovation are still pending, and some test datasets are small.

The model is publicly available on Hugging Face：https://huggingface.co/collections/nasa-ibm-ai4science/nasa-ibm-lunar-fm-and-downstream-models, the code is on GitHub：https://github.com/NASA-IMPACT/NASA-IBM-Lunar-Foundation-Model, and it's integrated into the open-source toolkit TerraTorch. The team also released the ML-ready pretraining datasets and benchmark collections.

The model is part of the NASA-IBM "AI for Science" collaboration. Both organizations have been working on foundation models under a Space Act Agreement since early 2022. In August 2023, they released the first Prithvi model on Hugging Face：https://the-decoder.com/ibm-and-nasa-publish-ai-model-for-climate-research/, trained on Landsat and Sentinel-2 imagery of the entire contiguous US and adapted for flood and wildfire mapping.

IBM had announced Prithvi as part of its Watsonx platform：https://the-decoder.com/watsonx-ibm-unveils-new-enterprise-platform-for-ai/. IBM developed TerraMind, the lunar model's base, in 2025 with ESA and Forschungszentrum Jülich for Earth observation.

Google Deepmind is pursuing a related approach with AlphaEarth Foundations：https://the-decoder.com/google-deepmind-unveils-an-ai-model-that-acts-as-a-virtual-satellite-for-mapping-the-entire-planet/, compressing optical satellite imagery, radar, lidar scans, and climate simulations into compact embeddings of Earth's surface.

Stay in the loop on AI. Clear, useful, no fluff.

Follow The Decoder for AI news, background stories and expert analyses.

The Decoder：https://the-decoder.com/