CARTO 现支持直接在地理空间基础模型（geospatial foundation model）的嵌入向量（embeddings）上运行分析。通过可视化、聚类与变化检测，将空间数据转化为决策依据。

CARTO now lets you run analytics directly on geospatial foundation model embeddings. Visualize, cluster, and detect changes to turn spatial data into decisions.

CARTO 现在支持直接在地理空间基础模型（geospatial foundation model）嵌入向量上运行分析。通过可视化、聚类与变化检测，将空间数据转化为决策依据。 今天，我们很高兴推出 CARTO 的全新功能：支持直接在地理空间基础模型嵌入向量上运行分析！与传统模型不同——后者通常专为单一任务构建，且依赖少量人工筛选的数据集——基础模型（FM）是经海量数据预训练的大型人工智能模型，可适配广泛多样的下游任务。在地理空间领域，此类基础模型基于多元（通常为多模态）数据集进行训练，例如卫星影像、地图、兴趣点（POI）及线上行为数据，以理解物理世界与人类活动。通过学习上述信息，这些基础模型生成一组地理嵌入（geo-embeddings）：可将其视作地理位置的数字指纹，以紧凑的向量形式（例如一串数字）完整表征某地的上下文信息。 一个地理嵌入可编码多种关键细节，例如： 尽管这些嵌入本身即蕴含丰富的信息层，但其真正价值在于揭示位置内部的模式与关联，从而发掘原本难以察觉的趋势。CARTO 充当基础模型研究与现实世界分析之间的桥梁，使这些强大模型面向所有用户开放——而不仅限于机器学习或数据科学专家。CARTO 直接在您的数据仓库（例如 BigQuery）中运行，确保工作流具备可扩展性与可复现性，使业务人员无需掌握专业机器学习知识即可利用基础模型嵌入。 那么，这些地理空间嵌入如何助力您的团队与组织获取洞见？ 地理嵌入将复杂的空间模式转化为数值向量，从而更便捷地开展分析、比对与基于位置信息的行动。CARTO 全新推出的“嵌入分析扩展包”（Analytics on Embeddings Extension Package）支持您： 基础（此处双关，意指“基础模型”与“奠定基础”）已然夯实，接下来让我们探索实际应用场景，展示这些前沿模型如何为商业决策、城市规划与环境应用等真实场景创造价值。 通过可视化城市区域的地理嵌入，规划者可快速获得建成环境的直观空间概览，从而比仅依赖传统卫星影像更高效地指导城市绿化、基础设施布局及热缓释策略制定。图中呈现的色彩模式表明：红色高浓度区域对应高密度建筑区，绿色突出植被密集区，蓝色则标识具有高反射率屋顶的工业或商业建筑。因此，该地理嵌入能迅速揭示建筑密度与地表覆盖模式——而这些模式在原始卫星影像中并不明显，既节省时间，又不牺牲数据规模或质量。 此外，地理嵌入亦是捕捉移动模式与环境信号的有力工具——这两项因素对于识别具备相似网络需求特征的区域至关重要。这有助于电信运营商优化基站或 Wi-Fi 热点布设，并更高效地规划基础设施升级。借助地理嵌入的聚类分析，组织可凸显服务缺口与网络模式——而这些模式在原始数据中往往无法直接识别，从而支持其高效确定改进优先级并做出数据驱动型决策。 有趣的是，该算法还能识别未受 DANA 洪水影响的区域，因其可检测两年间任何显著变化——例如季节性植被更替或持续进行的城市开发。这种更广域的变化检测能力解锁了时间维度洞察，确保分析师能够结合背景理解风暴影响，提升驱动商业决策的评估准确性。 借助基于地理嵌入的相似性分析工具，投资者与开发商可快速识别最优机会，以更高精度与信心比对不同市场。 最后，基础模型还可作为强大的特征提取器，生成嵌入向量，进而输入至传统机器学习工作流——例如回归、分类或预测任务——并与其它地理空间变量及非空间变量协同使用。 若您希望深入了解该示例，可在此访问 CARTO 与 Google AlphaEarth 联合举办的完整直播研讨会！ Lucía 是 CARTO 的数据科学家，致力于研发空间统计与机器学习解决方案，挖掘基于位置数据的潜在价值，助力组织最大化地理空间信息的应用价值。 Lucía 是 CARTO 的数据科学家，致力于研发空间统计与机器学习解决方案，挖掘基于位置数据的潜在价值，助力组织最大化地理空间信息的应用价值。 Lucía 是 CARTO 的数据科学家，致力于研发空间统计与机器学习解决方案，挖掘基于位置数据的潜在价值，助力组织最大化地理空间信息的应用价值。 通过 CARTO 与巴塞罗那超级计算中心（BSC）联合研讨会的见解，深入探索地理空间基础模型的发展现状——从表征学习到人口动态建模。 将 Google 的 PDFM 嵌入集成至 CARTO 工作流，以增强空间模型性能。 本博客展示了两个用例，说明此类嵌入如何提升预测效果。

CARTO now lets you run analytics directly on geospatial foundation model embeddings. Visualize, cluster, and detect changes to turn spatial data into decisions. Today, we're excited to introduce new capabilities in CARTO that let you run analytics directly on geospatial foundation model embeddings! Unlike traditional models, which are usually built for a single purpose and depend on a small number of hand-selected datasets, foundation models (FM) are large AI models pre-trained on vast amounts of data that can be adapted to a broad range of tasks. In the geospatial world, these foundation models are trained on diverse (usually multimodal) datasets like satellite imagery, maps, points of interest, and online behavior to understand the physical and human world. By learning from this information, these foundation models generate a set of geo-embeddings: think of a digital fingerprint of a geographic location in compact vector representations (ex. list of numbers) that captures the full context of a place. A geo-embedding can encode different key details, such as: While these embeddings are rich layers of information, the real power lies in how they reveal patterns and relationships within a location, uncovering trends that might otherwise remain hidden. CARTO acts as a bridge between foundation model research and real-world analytics, making these powerful models accessible to everyone— not just ML or data science experts. Running directly in your data warehouse (e.g., BigQuery), CARTO ensures scalable, reproducible workflows that let practitioners leverage foundation model embeddings without needing specialized machine learning expertise. So, how can these geospatial embeddings help drive insights for your team and organization? Geo-embeddings translate complex spatial patterns into numerical vectors, making it easier to analyze, compare, and act on location-based information. CARTO’s new extension package for easy Analytics on Embeddings lets you: Now the foundation is set (no pun intended), let’s explore practical ways to leverage these embeddings, showing how these cutting-edge models can add value to real-world decisions across business, urban planning, and environmental applications. By visualizing geo-embeddings of urban areas, planners can have a very first intuitive spatial view of the urban built environment, guiding decisions on urban greening, infrastructure, and heat-mitigation strategies faster than traditional satellite imagery. The resulting color patterns suggest that high concentration of red corresponds to high building density, green highlights areas with dense vegetation, and blue indicates industrial or commercial buildings with reflective roofs. This geo-embedding thus quickly reveals building-density and land-cover patterns that are not apparent in raw satellite imagery – saving time without sacrificing scale or quality of data. In addition, geo-embeddings are powerful tools to capture mobility patterns and environmental signals — key factors to identify areas with similar network demand characteristics. This can help telecom providers optimize the placement of cell towers or Wi-Fi hotspots and plan infrastructure upgrades more efficiently. Leveraging clustering on geo-embeddings thus helps highlight service gaps and network patterns that are not immediately visible from raw data, allowing organizations to prioritize improvements and make data-driven decisions efficiently. Interestingly, the algorithm also highlights regions not impacted by the DANA floods since it detects any significant change between the two years— such as seasonal vegetation shifts, or ongoing urban development. This broader detection unlocks temporal insights and ensures that analysts can contextualize storm impacts, improving the accuracy of assessments that drive business decisions. By leveraging geo-embedding-based similarity tools, investors and developers can quickly spot the best opportunities, comparing markets with precision and confidence. Lastly, Foundation Models can also be used as powerful feature extractors, generating embeddings that can then feed into traditional machine-learning workflows— such as regression, classification, or forecasting— alongside other geospatial and non-spatial variables. If you want to take a closer look at this example, you can access the full live webinar with Google AlphaEarth here! Lucía is Data Scientist at CARTO, where she develops spatial statistics and machine learning solutions that unveil the hidden potential of location-based data, enabling organizations to maximize the value of geospatial information. Lucía is Data Scientist at CARTO, where she develops spatial statistics and machine learning solutions that unveil the hidden potential of location-based data, enabling organizations to maximize the value of geospatial information. Lucía is Data Scientist at CARTO, where she develops spatial statistics and machine learning solutions that unveil the hidden potential of location-based data, enabling organizations to maximize the value of geospatial information. Explore the state of geospatial foundation models, from representation learning to population dynamics, with insights from the CARTO & BSC workshop. Integrate Google's PDFM embeddings into CARTO Workflows for enhanced spatial models. This blog shows two use cases on how these embeddings improve predictions.