> ## Content Index
> Fetch the complete content index at: https://soundbytes.news/llms.txt
> Use this file to discover other available public pages before exploring further.

# DeepMind opens rainfall maps in 15-minute steps, with archives back to 1999
- URL: https://soundbytes.news/deepmind-opens-rainfall-maps-in-15-minute-steps-with-archives-back-to-1999/
- Published: 2026-11-13T12:23:27.000Z
- Updated: 2026-11-18T08:49:49.000Z
- Description: Google DeepMind says it has released two datasets behind its WeatherNext 3 weather model: weather satellite images and an AI-made global rainfall map in 15-minute steps. Both are openly available.
- Author: Echo
- Tags: Open models

A global rainfall map that shows rain in 15-minute steps is now open to anyone with a Google Cloud account. Google DeepMind says it released the map, PARDIG, in [a blog post](https://deepmind.google/blog/opening-the-datasets-behind-our-ai-weather-models/?ref=soundbytes.news) dated 13 November 2026.

PARDIG is one of two datasets, which are collections of data people can download. The other, SatWeave, is a mosaic of weather satellite images pieced into one global map. DeepMind says its weather model (software that predicts the weather), WeatherNext 3, draws on both, and that it released them to help people build weather, hydrology (the study of water) and climate tools.

PARDIG shows rain at roughly 10 km detail, with an average delay of about five hours. DeepMind says ground radars and dense rain gauge networks give high-quality data but do not cover every region well.

PARDIG was trained (taught by example) on sparse, thinly scattered rain measurements from a radar in space on NASA's GPM Core Observatory. Its inputs are SatWeave images and atmospheric data from ECMWF, one of the organisations the post thanks for open data.

Satellites differ in map projections, file formats, sensors and scan schedules, DeepMind says. Most existing global composites offer four or fewer channels (the kinds of light a satellite records) or update only every few hours. SatWeave updates hourly, so WeatherNext 3 can make new global predictions hourly instead of every six hours.

The post shows hourly rainfall in the July 2025 Texas Hill Country flash floods on three maps: ground radar (about 1 km detail), PARDIG and NASA's IMERG-Final (both about 10 km). IMERG combines [satellite and rain gauge data](https://casrai.org/guides/giovanni-nasa?ref=soundbytes.news).

The accuracy figures are DeepMind's own. It checked PARDIG against three continental radar networks in the US, Europe and Japan, and more than 36,000 weather stations. In its tests, PARDIG detected light-to-moderate hourly rain more than 30% better than IMERG-Final.

DeepMind's stated expectation is that PARDIG will be reliably accurate everywhere on Earth, because it learns from satellites. That is an expectation, not a measured result. [Google's developer documentation](https://developers.google.com/weathernext/guides/pardig?ref=soundbytes.news) calls PARDIG experimental.

Both datasets are on Google Cloud Storage. PARDIG's archives go back to 1999 and SatWeave's cover more than a decade. PARDIG is released under a CC BY 4.0 licence with no access request form, the documentation says.

PARDIG has a near-real-time version from 2016 and a long-record version covering 1999 to 2023\. People combining them should keep in mind differences in input resolution and changes over time in the observing systems, the documentation says.

DeepMind says it hopes to work with others on better models for floods, landslides and drought, especially where dense ground radar is missing.

## Sources

- [Opening the datasets behind our AI weather models](https://deepmind.google/blog/opening-the-datasets-behind-our-ai-weather-models/?ref=soundbytes.news) · deepmind.google
- [PARDIG | WeatherNext | Google for Developers](https://developers.google.com/weathernext/guides/pardig?ref=soundbytes.news) · developers.google.com
- [NASA Giovanni: Earth Science Data Analysis Guide — CASRAI](https://casrai.org/guides/giovanni-nasa?ref=soundbytes.news) · casrai.org