![GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL](https://developers.google.com/earth-engine/datasets/images/GOOGLE/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL_sample.png) Dataset Availability : 2017-01-01T00:00:00Z--2025-01-01T00:00:00Z Dataset Producer : [Google Earth Engine](https://earthengine.google.com/) [Google DeepMind](https://deepmind.google/) Tags : [annual](https://developers.google.com/earth-engine/datasets/tags/annual) [global](https://developers.google.com/earth-engine/datasets/tags/global) [google](https://developers.google.com/earth-engine/datasets/tags/google) [landsat-derived](https://developers.google.com/earth-engine/datasets/tags/landsat-derived) [satellite-imagery](https://developers.google.com/earth-engine/datasets/tags/satellite-imagery) [sentinel1-derived](https://developers.google.com/earth-engine/datasets/tags/sentinel1-derived) [sentinel2-derived](https://developers.google.com/earth-engine/datasets/tags/sentinel2-derived) #### Description The Google Satellite Embedding dataset is a global, analysis-ready collection of learned geospatial [embeddings](https://developers.google.com/machine-learning/crash-course/embeddings/embedding-space). Each 10-meter pixel in this dataset is a 64-dimensional representation, or "[embedding vector](https://developers.google.com/machine-learning/glossary#embedding-vector)," that encodes temporal trajectories of surface conditions at and around that pixel as measured by various Earth observation instruments and datasets, over a single calendar year. Unlike conventional spectral inputs and indices, where bands correspond to physical measurements, embeddings are feature vectors that summarize relationships across multi-source, multi-modal observations in a less directly interpretable, but more powerful way. [See usage examples and more detailed explanations](https://medium.com/google-earth/ai-powered-pixels-introducing-googles-satellite-embedding-dataset-31744c1f4650). The dataset covers terrestrial land surfaces and shallow waters, including intertidal and reef zones, inland waterways, and coastal waterways. Coverage at the poles is limited by satellite orbits and instrument coverage. The collection is composed of images covering approximately 163,840 meters by 163,840 meters, and each image has 64 bands `{A00, A01, ..., A63}`, one for each axis of the 64D embedding space. All bands should be used for downstream analysis as they collectively refer to a 64D coordinate in the embedding space and are not independently interpretable. All images are generated in their local Universal Transverse Mercator projection as indicated by the UTM_ZONE property, and have `system:time_start` and `system:time_end` properties that reflect the calendar year summarized by the embeddings; for example, an embedding image for 2021 will have a `system:start_time` equal to `ee.Date('2021-01-01 00:00:00')` and a `system:end_time` equal to `ee.Date('2022-01-01 00:00:00')`. The embeddings are unit-length, meaning they have a magnitude of 1 and do not require any additional normalization, and are distributed across the [unit sphere](https://en.wikipedia.org/wiki/Unit_sphere), making them well-suited for use with clustering algorithms and tree-based classifiers. The embedding space is also consistent across years, and embeddings from different years can be used for condition change detection by considering the dot product or angle between two embedding vectors. Furthermore, the embeddings are designed to be linearly composable, i.e., they can be aggregated to produce embeddings at coarser spatial resolutions or transformed with vector arithmetic, and still retain their semantic meaning and distance relationships. The Satellite Embedding dataset was produced by [AlphaEarth Foundations](https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/), a geospatial embedding model that assimilates multiple datastreams including optical, radar, LiDAR, and other sources (Brown, Kazmierski, Pasquarella et al., 2025; preprint available [here](https://arxiv.org/abs/2507.22291)). Because representations are learned across many sensors and images, embedding representations generally overcome common issues such as clouds, scan lines, sensor artifacts, or missing data, providing seamless analysis-ready features that can be directly substituted for other Earth Observation image sources in classification, regression, and change detection analyses. The embeddings in this image collection were generated using v2.1 of the AlphaEarth Foundations model, which includes a number of improvements over the v2.0 model evaluated in the AlphaEarth Foundations paper. Namely, the training dataset was re-generated to include a large number of samples from Antarctica that had previously been dropped due to limited sensor coverage, effectively increasing the count of training video sequences from over 8.4 million to over 10.1 million sequences; and the USDA NASS Cropland Data Layer was included as an additional target during training; loss weights for NLCD and CDL were lowered from 0.50 to 0.25; and several other minor changes were implemented to better mitigate visual artifacts associated with input sensor swaths, tiling, and multi-resolution pixel targets. These changes did not notably affect the model's performance in terms of evaluation metrics, but generally improved the quality of the resulting embeddings. While some large scale swath and data availability artifacts remain, these typically represent minor vector offsets and generally do not significantly affect downstream processing or results. The dataset is also available on Google Cloud Storage (GCS) in the bucket `gs://alphaearth_foundations`. You can browse the bucket contents via the [Google Cloud Console](https://console.cloud.google.com/storage/browser/alphaearth_foundations). and refer to the bucket [README](https://developers.google.com/earth-engine/guides/aef_on_gcs_readme) for full details on the data structure. As of July 2026, this bucket is set up as "provider pays". Updates: - As of 2025-11-17, embedding layers have DATASET_VERSION 1.1. This includes a regenerated 2017 layer that incorporates additional Sentinel-1 acquisitions. Google is committed to ongoing production of annual Satellite Embedding layers and will provide at least one year of advance notice of any anticipated change in delivery, subject to the ongoing availability of input data streams from USGS and ESA that production of the dataset relies on. ### Bands **Bands** Pixel size: 10 meters (all bands) | Name | Units | Min | Max | Pixel Size | Description | |---|---|---|---|---|---| | `A00` | Dimensionless | -1 | 1 | 10 meters | The 0th axis of the embedding vector. | | `A01` | Dimensionless | -1 | 1 | 10 meters | The 1st axis of the embedding vector. | | `A02` | Dimensionless | -1 | 1 | 10 meters | The 2nd axis of the embedding vector. | | `A03` | Dimensionless | -1 | 1 | 10 meters | The 3rd axis of the embedding vector. | | `A04` | Dimensionless | -1 | 1 | 10 meters | The 4th axis of the embedding vector. | | `A05` | Dimensionless | -1 | 1 | 10 meters | The 5th axis of the embedding vector. | | `A06` | Dimensionless | -1 | 1 | 10 meters | The 6th axis of the embedding vector. | | `A07` | Dimensionless | -1 | 1 | 10 meters | The 7th axis of the embedding vector. | | `A08` | Dimensionless | -1 | 1 | 10 meters | The 8th axis of the embedding vector. | | `A09` | Dimensionless | -1 | 1 | 10 meters | The 9th axis of the embedding vector. | | `A10` | Dimensionless | -1 | 1 | 10 meters | The 10th axis of the embedding vector. | | `A11` | Dimensionless | -1 | 1 | 10 meters | The 11th axis of the embedding vector. | | `A12` | Dimensionless | -1 | 1 | 10 meters | The 12th axis of the embedding vector. | | `A13` | Dimensionless | -1 | 1 | 10 meters | The 13th axis of the embedding vector. | | `A14` | Dimensionless | -1 | 1 | 10 meters | The 14th axis of the embedding vector. | | `A15` | Dimensionless | -1 | 1 | 10 meters | The 15th axis of the embedding vector. | | `A16` | Dimensionless | -1 | 1 | 10 meters | The 16th axis of the embedding vector. | | `A17` | Dimensionless | -1 | 1 | 10 meters | The 17th axis of the embedding vector. | | `A18` | Dimensionless | -1 | 1 | 10 meters | The 18th axis of the embedding vector. | | `A19` | Dimensionless | -1 | 1 | 10 meters | The 19th axis of the embedding vector. | | `A20` | Dimensionless | -1 | 1 | 10 meters | The 20th axis of the embedding vector. | | `A21` | Dimensionless | -1 | 1 | 10 meters | The 21st axis of the embedding vector. | | `A22` | Dimensionless | -1 | 1 | 10 meters | The 22nd axis of the embedding vector. | | `A23` | Dimensionless | -1 | 1 | 10 meters | The 23rd axis of the embedding vector. | | `A24` | Dimensionless | -1 | 1 | 10 meters | The 24th axis of the embedding vector. | | `A25` | Dimensionless | -1 | 1 | 10 meters | The 25th axis of the embedding vector. | | `A26` | Dimensionless | -1 | 1 | 10 meters | The 26th axis of the embedding vector. | | `A27` | Dimensionless | -1 | 1 | 10 meters | The 27th axis of the embedding vector. | | `A28` | Dimensionless | -1 | 1 | 10 meters | The 28th axis of the embedding vector. | | `A29` | Dimensionless | -1 | 1 | 10 meters | The 29th axis of the embedding vector. | | `A30` | Dimensionless | -1 | 1 | 10 meters | The 30th axis of the embedding vector. | | `A31` | Dimensionless | -1 | 1 | 10 meters | The 31st axis of the embedding vector. | | `A32` | Dimensionless | -1 | 1 | 10 meters | The 32nd axis of the embedding vector. | | `A33` | Dimensionless | -1 | 1 | 10 meters | The 33rd axis of the embedding vector. | | `A34` | Dimensionless | -1 | 1 | 10 meters | The 34th axis of the embedding vector. | | `A35` | Dimensionless | -1 | 1 | 10 meters | The 35th axis of the embedding vector. | | `A36` | Dimensionless | -1 | 1 | 10 meters | The 36th axis of the embedding vector. | | `A37` | Dimensionless | -1 | 1 | 10 meters | The 37th axis of the embedding vector. | | `A38` | Dimensionless | -1 | 1 | 10 meters | The 38th axis of the embedding vector. | | `A39` | Dimensionless | -1 | 1 | 10 meters | The 39th axis of the embedding vector. | | `A40` | Dimensionless | -1 | 1 | 10 meters | The 40th axis of the embedding vector. | | `A41` | Dimensionless | -1 | 1 | 10 meters | The 41st axis of the embedding vector. | | `A42` | Dimensionless | -1 | 1 | 10 meters | The 42nd axis of the embedding vector. | | `A43` | Dimensionless | -1 | 1 | 10 meters | The 43rd axis of the embedding vector. | | `A44` | Dimensionless | -1 | 1 | 10 meters | The 44th axis of the embedding vector. | | `A45` | Dimensionless | -1 | 1 | 10 meters | The 45th axis of the embedding vector. | | `A46` | Dimensionless | -1 | 1 | 10 meters | The 46th axis of the embedding vector. | | `A47` | Dimensionless | -1 | 1 | 10 meters | The 47th axis of the embedding vector. | | `A48` | Dimensionless | -1 | 1 | 10 meters | The 48th axis of the embedding vector. | | `A49` | Dimensionless | -1 | 1 | 10 meters | The 49th axis of the embedding vector. | | `A50` | Dimensionless | -1 | 1 | 10 meters | The 50th axis of the embedding vector. | | `A51` | Dimensionless | -1 | 1 | 10 meters | The 51st axis of the embedding vector. | | `A52` | Dimensionless | -1 | 1 | 10 meters | The 52nd axis of the embedding vector. | | `A53` | Dimensionless | -1 | 1 | 10 meters | The 53rd axis of the embedding vector. | | `A54` | Dimensionless | -1 | 1 | 10 meters | The 54th axis of the embedding vector. | | `A55` | Dimensionless | -1 | 1 | 10 meters | The 55th axis of the embedding vector. | | `A56` | Dimensionless | -1 | 1 | 10 meters | The 56th axis of the embedding vector. | | `A57` | Dimensionless | -1 | 1 | 10 meters | The 57th axis of the embedding vector. | | `A58` | Dimensionless | -1 | 1 | 10 meters | The 58th axis of the embedding vector. | | `A59` | Dimensionless | -1 | 1 | 10 meters | The 59th axis of the embedding vector. | | `A60` | Dimensionless | -1 | 1 | 10 meters | The 60th axis of the embedding vector. | | `A61` | Dimensionless | -1 | 1 | 10 meters | The 61st axis of the embedding vector. | | `A62` | Dimensionless | -1 | 1 | 10 meters | The 62nd axis of the embedding vector. | | `A63` | Dimensionless | -1 | 1 | 10 meters | The 63rd axis of the embedding vector. | ### Image Properties **Image Properties** | Name | Type | Description | |---|---|---| | MODEL_VERSION | STRING | The version string uniquely identifying the model version used to produce the image. | | PROCESSING_SOFTWARE_VERSION | STRING | The version string uniquely identifying the model data processing software used to produce the image. | | UTM_ZONE | STRING | The UTM zone of the coordinate reference system used to produce the image. | | DATASET_VERSION | STRING | The dataset version. | ### Terms of Use **Terms of Use** This dataset is licensed under [CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/) and requires the following attribution text: "The AlphaEarth Foundations Satellite Embedding [dataset](https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL) is produced by Google and Google DeepMind." ### Citations Citations: - Brown, C. F., Kazmierski, M. R., Pasquarella, V J., Rucklidge, W. J., Samsikova, M., Zhang, C., Shelhamer, E., Lahera, E., Wiles, O., Ilyushchenko, S., Gorelick, N., Zhang, L. L., Alj, S., Schechter, E., Askay, S., Guinan, O., Moore, R., Boukouvalas, A., \& Kohli, P.(2025). AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data. arXiv preprint arXiv.2507.22291. [doi:10.48550/arXiv.2507.22291](https://doi.org/10.48550/arXiv.2507.22291) ### Explore with Earth Engine > [!NOTE] > **Important:** Earth Engine is a platform for petabyte-scale scientific analysis and visualization of geospatial datasets, both for public benefit and for business and government users. Earth Engine is free to use for research, education, and nonprofit use. To get started, please [register for Earth Engine access.](https://console.cloud.google.com/earth-engine) ### Code Editor (JavaScript) ```javascript // Load collection. var dataset = ee.ImageCollection('GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL'); // Point of interest. var point = ee.Geometry.Point(-121.8036, 39.0372); // Get embedding images for two years. var image1 = dataset .filterDate('2023-01-01', '2024-01-01') .filterBounds(point) .first(); var image2 = dataset .filterDate('2024-01-01', '2025-01-01') .filterBounds(point) .first(); // Visualize three axes of the embedding space as an RGB. var visParams = {min: -0.3, max: 0.3, bands: ['A01', 'A16', 'A09']}; Map.addLayer(image1, visParams, '2023 embeddings'); Map.addLayer(image2, visParams, '2024 embeddings'); // Calculate dot product as a measure of similarity between embedding vectors. // Note for vectors with a magnitude of 1, this simplifies to the cosine of the // angle between embedding vectors. var dotProd = image1 .multiply(image2) .reduce(ee.Reducer.sum()); // Add dot product to the map. Map.addLayer( dotProd, {min: 0, max: 1, palette: ['white', 'black']}, 'Similarity between years (brighter = less similar)' ); Map.centerObject(point, 12); Map.setOptions('SATELLITE'); ``` [Open in Code Editor](https://code.earthengine.google.com/?scriptPath=Examples:Datasets/GOOGLE/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL) [Satellite Embedding V1](https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL) The Google Satellite Embedding dataset is a global, analysis-ready collection of learned geospatial embeddings. Each 10-meter pixel in this dataset is a 64-dimensional representation, or "embedding vector," that encodes temporal trajectories of surface conditions at and around that pixel as measured by various Earth observation instruments and datasets, over a ... GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL, annual,global,google,landsat-derived,satellite-imagery,sentinel1-derived,sentinel2-derived 2017-01-01T00:00:00Z/2025-01-01T00:00:00Z -90 -180 90 180 Google Earth Engine https://developers.google.com/earth-engine/datasets - [](https://doi.org/https://earthengine.google.com/) - [](https://doi.org/https://deepmind.google/) - [](https://doi.org/https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL)