Introducing OlmoEarth embeddings: Custom embedding exports from OlmoEarth Studio for downstream analysis

介绍 OlmoEarth 嵌入：从 OlmoEarth Studio 导出自定义嵌入以进行下游分析

OlmoEarth Studio, our platform for building Earth observation models, now lets you compute and export embedding vectors—compact numerical representations of Earth-observation data produced by our open source OlmoEarth foundation models. The source code and model weights are publicly available alongside the research paper, so the community can inspect exactly how these embeddings are generated. OlmoEarth Studio 是我们用于构建地球观测模型的平台，现在支持计算并导出嵌入向量（embedding vectors）——这是由我们的开源 OlmoEarth 基础模型生成的地球观测数据的紧凑数值表示。源代码和模型权重已与研究论文一同公开，因此社区可以准确地检查这些嵌入是如何生成的。

Embeddings are a fast, cost-effective entry point for leveraging OlmoEarth: they support a wide range of downstream tasks, from similarity search to segmentation to unsupervised exploration. Locations with similar surface characteristics end up with similar vectors; locations that differ land far apart. OlmoEarth embeddings have shown strong performance in our own benchmarking and in independent evaluations. The exported Cloud-Optimized GeoTIFFs (COGs) are lightweight and easy to share. 嵌入是利用 OlmoEarth 的一种快速且经济高效的切入点：它们支持广泛的下游任务，从相似性搜索到分割，再到无监督探索。具有相似地表特征的地点会产生相似的向量；而不同的地点则在向量空间中相距甚远。OlmoEarth 嵌入在我们自己的基准测试和独立评估中均表现出色。导出的云优化 GeoTIFF (COG) 文件轻量且易于共享。

Choose your area of interest, time range, encoder variant, resolution, and imagery sources via the Studio UI or API, and get back a COG you can use however you like. If your application requires higher performance, Studio also supports supervised fine-tuning (SFT). Custom-computed embeddings are now available for users of OlmoEarth Studio. Reach out if you’re interested in gaining access. 通过 Studio UI 或 API 选择您的感兴趣区域、时间范围、编码器变体、分辨率和影像源，即可获得可按需使用的 COG 文件。如果您的应用需要更高的性能，Studio 还支持监督微调 (SFT)。自定义计算的嵌入功能现已向 OlmoEarth Studio 用户开放。如果您有兴趣获取访问权限，请与我们联系。

Computing embeddings in Studio

在 Studio 中计算嵌入

Computing embeddings follows the same workflow as any other prediction in Studio. First configure a model and run it, and then download the results. Several parameters tailor the output: 计算嵌入遵循与 Studio 中任何其他预测相同的流程。首先配置模型并运行，然后下载结果。可以通过以下几个参数来定制输出：

• Area of interest: Draw or upload any polygon; Studio handles imagery acquisition and tiling.
• 感兴趣区域： 绘制或上传任何多边形；Studio 会自动处理影像获取和切片。
• Time span: 1-12 monthly periods.
• 时间跨度： 1-12 个月周期。
• Encoder variant: Nano (128-dim, 1.4M params), Tiny (192-dim, 6.2M params), or Base (768-dim, 89M params).
• 编码器变体： Nano（128 维，140 万参数）、Tiny（192 维，620 万参数）或 Base（768 维，8900 万参数）。
• Spatial resolution: 10 meter, 20 meter, 40 meter, or 80 meter per pixel.
• 空间分辨率： 每像素 10 米、20 米、40 米或 80 米。
• Imagery sources: Sentinel-2 L2A, Sentinel-1 RTC, or both.
• 影像源： Sentinel-2 L2A、Sentinel-1 RTC 或两者兼有。

Studio delivers a COG with one band per embedding dimension. Vectors are stored as signed 8-bit integers (int8). Values range from -127 to +127, with -128 reserved for nodata. To recover floating-point vectors, see dequantize_embeddings in olmoearth_pretrain. Studio 交付的 COG 文件中，每个嵌入维度对应一个波段。向量以有符号 8 位整数 (int8) 存储。数值范围从 -127 到 +127，其中 -128 保留用于无数据值。如需恢复浮点向量，请参阅 olmoearth_pretrain 中的 dequantize_embeddings 函数。

Because everything is computed on demand rather than pulled from a pre-computed global archive, your embeddings reflect exactly the conditions you care about. You can generate monthly embeddings to capture seasonal dynamics, not just annual snapshots. 由于所有内容都是按需计算，而非从预先计算的全局存档中提取，因此您的嵌入能够精确反映您所关心的条件。您可以生成月度嵌入来捕捉季节性动态，而不仅仅是年度快照。

What you can do with OlmoEarth embeddings

您可以用 OlmoEarth 嵌入做什么

Similarity search: Finding “more like this” 相似性搜索：寻找“更多类似内容”

Pick a query pixel, extract its embedding, and compute cosine similarity against every other pixel. The result is a heatmap showing where the landscape looks most and least like your query pixel. This query sits near the Merced urban center in California. Urban fabric and road corridors light up coherently while agricultural parcels stay dark. The model distinguishes built-up surfaces from cropland without any labels. 选择一个查询像素，提取其嵌入，并计算其与所有其他像素的余弦相似度。结果是一张热力图，显示了哪些区域的地貌与您的查询像素最相似或最不相似。此查询位于加利福尼亚州默塞德 (Merced) 城市中心附近。城市建筑和道路走廊呈现出一致的亮度，而农田则保持暗色。该模型无需任何标签即可区分建筑表面和农田。

Few-shot segmentation: Labeling the landscape 少样本分割：地貌标注

Similarity search tells you “where is it like this?” but sometimes you need discrete labels across a region. Because the representations are already rich, a simple linear classifier can produce a wall-to-wall land-cover map from very few labeled pixels. 相似性搜索告诉您“哪里像这样？”，但有时您需要整个区域的离散标签。由于这些表示已经非常丰富，一个简单的线性分类器仅需极少量的标注像素即可生成覆盖全区域的土地覆盖图。

The core of the analysis is a few lines of Python: 分析的核心代码仅需几行 Python：

import rasterio
import numpy as np
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

# Load the 192-band embedding COG exported from Studio
with rasterio.open("embeddings.tif") as ds:
    emb = ds.read().astype(np.float32) # (192, H, W)

C, H, W = emb.shape
X = emb.reshape(C, -1).T # (H*W, 192)

# Train on labeled pixels, predict everywhere
clf = make_pipeline(StandardScaler(), LogisticRegression())
# ... (后续代码省略)