Satlas

HomeMapSuper-ResolutionAIDataFAQSatlasOpen Geospatial Data Generated by AIExplore MapSee examples of how our planet is changingThe monthly geospatial data in Satlas reveals changes in marine infrastructure, renewable energy infrastructure, and tree cover.China invests heavily in offshore wind farmsSince 2017, China's offshore windfarms have grown by over 10xA new windfarm near DenmarkIn recent years, a major offshore windfarm has been developed near DenmarkAmazon rainforest deforestationForests in this region of Brazil were recently converted to agricultureLogging in Washington StateSatlas reveals tree cover loss from commercial logging in the Olympic PeninsulaWind farms off ScotlandA timelapse of the development of the Beatrice and Moray wind farms off of ScotlandOil platforms in the Gulf of MexicoThere are extensive offshore platforms near Texas, Louisiana, and MississippiA new dam in the AmazonThe Sinop dam was completed along the Teles Pires river in 2019A new solar farm in VirginiaThe Pleinmont solar farm is developed in the forests of VirginiaWind farm near ShanghaiIn recent years, China has developed a major offshore wind farm near ShanghaiSolar development in BrazilBrazil has invested heavily in solar farms over the past few yearsWindfarm development in southern IndiaMany windmills have been developed recently in southern IndiaSuper-resolution of a marinaSuper-resolution highlights the details of the Everett waterfrontSuper-ResolutionDiscover how our AI models can enhance low resolution satellite imagery to produce high resolution images on a global scale.Explore Super-ResolutionSentinel-2Super-ResNakuru, KenyaAI ModelsAI models in Satlas employ state of the art architectures and training algorithms in computer vision. These models leverage a new large-scale remote sensing dataset called SatlasPretrain with over 30 TB of imagery and 300 million labels.Learn More about AI in SatlasGeospatial DataOur AI-generated geospatial datasets are freely and publicly available and can be downloaded for offline analysis. We release our AI models and training labels as well.Access DataTeamThis demo page is built and maintained by PRIOR and colleagues at the Allen Institute for AI. Our team seeks to advance computer vision to create AI systems that see, explore, learn, and reason about the world.Computer Vision TeamFavyen BastaniPiper WoltersAni KembhaviResearch Visualization TeamJon BorchardtArnavi ChhedaAaron SarnatMichael SchmitzShare this project: --- HomeMapSuper-ResolutionAIDataFAQ​ApplicationMarine InfrastructureRenewable EnergyTree CoverSuper ResolutionLayoutSingle MapSide by SideSplit ScreenChange Over TimeTime2016-012017-012018-012019-012020-012021-012022-012023-012023-12View TimelapseOptionsSatellite ViewHide AnnotationsMarine Infrastructure:Offshore Wind TurbinesOffshore PlatformsHeatmapDetailsLat, Lon: 55.960, 18.470 --- HomeMapSuper-ResolutionAIDataFAQSuper-Resolution Sentinel-2Super-ResNakuru, KenyaSuper-Resolution allows a very high level of detail compared to Sentinel-2 images.IntroductionGlobal low resolution satellite imagery is available on a weekly basis, but public high resolution imagery is very limited. High resolution imagery can help with tasks such as post-disaster building damage estimation and crop type classification, but since it is available infrequently, we cannot scale up promising AI methods to automate these tasks, especially in developing countries where it is needed most.We have trained super resolution models to generate high resolution imagery on a global scale.The Satlas Map allows users to explore globally generated high resolution imagery for 2023.Frequent UpdatesSuper-Resolution is generated from 2023 Sentinel-2 imagery, and is sometimes more up-to-date than Google Maps. This example of Cebu City, Phillippines shows evidence of coastal infrastructure in both Sentinel-2 and the generated Super-Resolution but not in the Google Maps image.2023 Sentinel-2 imageGoogle Maps image (outdated)2023 Satlas Super-ResolutionChange Over TimeWith increased resolution and cloud-free views, Super-Resolution's enhanced imagery makes it easier to see change over time.Sinop Dam in Brazil filling up, including removal of trees and dirt to build pathways for the water.Super-Resolution ModelThe Super-Resolution imagery is generated using an adaptation of the ESRGAN model. It is trained on pairs of corresponding Sentinel-2 (low resolution) and NAIP (high resolution) images. Our model inputs a sequence of between 6-18 Sentinel-2 images for each location.Super-Resolution Model ArchitectureThe training and inference code is open source and can be found at our Github. We note that AI models are not perfect and there are instances where the super-resolution outputs are incorrect. --- HomeMapSuper-ResolutionAIDataFAQAI in SatlasSatlas combines the power of modern AI with the scale of public domain satellite imagery to provide monthly monitoring of the planet.AI models in Satlas process freely available satellite images captured by the European Space Agency’s Sentinel-2 satellites. These images cover the majority of the planet every week, but are low in resolution, making them difficult to interpret even for humans.We leverage the latest advances in AI to robustly process images in this challenging domain and produce a variety of geospatial data products that we make freely available.SatlasPretrainOur AI models are pre-trained on a new large-scale remote sensing dataset called SatlasPretrain. This vast dataset contains over 30 TB of imagery with 302 million labels spanning 137 diverse categories, from tree cover and crop fields to wind farms and oil wells. Pre-training on SatlasPretrain teaches AI models to understand geographically and seasonally diverse satellite images.SatlasPretrain appeared at the 2023 International Conference on Computer Vision (ICCV 2023)Learn MoreRead the PaperDownload SatlasPretrainFine-tuning the ModelsThe AI models in Satlas are fine tuned on high-quality training datasets that we hand-label for each geospatial data product. In practice, we use multiple steps of fine-tuning, testing, and additional labeling to progressively improve the accuracy of our models.Access the Training DataSolar FarmsWind TurbinesOffshore PlatformsAI Model ArchitectureOur AI models use state of the art machine learning architectures and training methods. They input a sequence of the three most recent satellite images captured at each location. Each image is passed through a Swin Transformer backbone to extract features. These features are then combined via max temporal pooling, and passed to task-specific neural network heads to make the final predictions. Data AccuracyA per-continent breakdown of the accuracy of each Satlas geospatial data product is made available in our Data Validation Report. We also showcase examples of the limitations of our data, including erroneous and missing data. Overall, the data generated by our AI models has high accuracy, but AI systems are never perfect and several factors tend to degrade performance.View Data Validation Report