Google Opens WeatherNext 3 Datasets SatWeave and PARDIG
谷歌开源 WeatherNext 3 气象模型所用数据集：含 SatWeave 卫星影像拼接数据集与 PARDIG 全球降水信息
Open access to hourly-updating satellite composites and a 10 km, 15-minute precipitation product removes a major data barrier for teams in radar-sparse regions, where PARDIG's satellite-only training keeps accuracy high. Hydrology, solar power forecasting, wildfire monitoring and agri-weather startups can build on the 1999–2026 archive and live streams under CC BY 4.0. Compared with NASA IMERG-Final, the 30%+ gain in light-precipitation skill makes it a credible replacement baseline for downstream modelling work.
Google DeepMind on 13 Nov opened two foundational datasets behind its WeatherNext 3 weather model: SatWeave, a global stitched satellite imagery product, and PARDIG, an AI-based global precipitation reanalysis dataset. Both are offered to researchers worldwide as real-time data streams plus historical archives, licensed under CC BY 4.0 on Google Cloud Storage.
SatWeave merges observations from five geostationary satellites — GOES-East, GOES-West, Meteosat 0, Meteosat Indian Ocean and Himawari — resolving differences in map projections, file formats, sensors and scanning cadence. It updates hourly with more than ten years of archives, letting WeatherNext 3 generate hourly global forecasts from live cloud imagery, versus the six-hour refresh cycle of traditional analysis data. PARDIG uses a spherical grid Transformer architecture similar to WeatherNext 3, trained on sparse spaceborne radar observations from NASA's GPM Core Observatory, with SatWeave imagery and ECMWF atmospheric fields as inputs, delivering roughly 10 km spatial and 15-minute temporal resolution at about five hours of latency.
Evaluated against ground radar networks in the US, Europe and Japan plus over 36,000 weather stations, PARDIG outperforms NASA IMERG-Final across all rain intensities, improving light-to-moderate precipitation detection skill by more than 30%. Because it learns directly from satellite observations, it keeps reliable accuracy in regions lacking ground radars and gauges.
谷歌开源 WeatherNext 3 气象模型所用数据集：含 SatWeave 卫星影像拼接数据集与 PARDIG 全球降水信息
IT之家 11 月 13 日消息，谷歌 DeepMind 今天正式开放 WeatherNext 3 气象模型所依托的两大基础数据集 SatWeave 与 PARDIG，相应数据集面向全球研究者与开发者提供实时数据流及历史归档，可用于天气、水文、气候及相关下游应用建模。 据介绍，SatWeave 是一套整合多颗静止气象卫星的全球影像拼接产品，聚合了 GOES-East、GOES-West、Meteosat 0、Meteosat 印度洋星以及 Himawari 共 5 颗卫星的观测数据，解决了不同卫星在地图投影、文件格式、传感器与扫描节奏上的差异难题。该数据集每小时更新一次，并提供跨越十年以上的历史存档，能够让 WeatherNext 3 基于实时云图与风暴动态生成逐小时全球预报，而传统分析数据通常每六小时才刷新一次。谷歌还希望研究社区将其拓展至太阳能功率预测、野火监测等场景。 而 PARDIG 是一套基于 AI 架构的全球降水再分析数据集，采用与 WeatherNext 3 相似的球形图网格 Transformer 架构，直接以 NASA 全球降水测量核心观测台的稀疏星载雷达观测为训练目标，输入使用 SatWeave 影像与欧洲中期天气预报中心的大气场。其空间分辨率约 10 公里（0.1°），时间分辨率 15 分钟，平均延迟约 5 小时。 谷歌通过对照美国、欧洲、日本三大陆基雷达网络及全球 3.6 万余座气象站的评估显示，PARDIG 在所有雨强档位上均优于 NASA IMERG-Final 等主流基线产品；对轻到中度降水的小时级识别技巧较 IMERG-Final 提升 30% 以上，对极端降水的捕捉能力尤为突出。由于模型直接从卫星观测学习，其在缺乏地面雷达与雨量计的地区同样具备可靠精度。 除此之外，谷歌本次还同步开放 1999 年至 2026 年的 PARDIG 历史档案以及 SatWeave 与 PARDIG 的实时数据流。相应数据托管在 Google Cloud Storage 上，以 CC BY 4.0 许可开放。
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Google DeepMind, WeatherNext 3, PARDIG, SatWeave.
Source attribution: curated from IT之家.
