
Google DeepMind has opened entry to a synthetic intelligence system that tracks rainfall worldwide, doubtlessly serving to researchers monitor floods and droughts in areas with restricted climate infrastructure.
The corporate introduced on Nov. 13 that researchers and builders can now entry historic data and constantly up to date information from PARDIG, its AI-powered world precipitation monitoring system.
The expertise combines satellite tv for pc observations and atmospheric information to estimate rainfall each quarter-hour throughout a worldwide grid with roughly 10-kilometer decision.
Its launch may increase entry to rainfall intelligence in creating international locations and distant areas the place restricted climate radar protection makes it tough to trace storms, flooding and extended dry spells.
Google mentioned PARDIG learns from precipitation measurements collected by NASA’s International Precipitation Measurement satellite tv for pc and makes use of further satellite tv for pc imagery and atmospheric data to estimate rainfall throughout the planet.
In contrast to standard monitoring programs that rely closely on ground-based climate stations, PARDIG can generate estimates over oceans and areas with sparse commentary networks.
The dataset additionally helps Google’s WeatherNext 3 forecasting mannequin, which makes use of PARDIG-generated rainfall data throughout coaching.
Researchers can entry historic rainfall data courting again to 1999, alongside a separate dataset overlaying 2016 to the current.
Google has launched each underneath a Artistic Commons license that allows industrial reuse with attribution, doubtlessly permitting firms to include the data into agricultural monitoring, water administration and disaster-risk functions.
Google’s AI beats NASA rainfall benchmark by greater than 30%
DeepMind mentioned PARDIG achieved greater than 30% increased hourly precipitation detection talent than NASA’s IMERG-Closing dataset when measuring light-to-moderate rainfall.
The corporate evaluated the system in opposition to radar networks in america, Europe and Japan, alongside observations from greater than 36,000 climate stations worldwide.
Its analysis paper additionally reported improved identification of intense rainfall occasions, which current world precipitation datasets can battle to seize precisely.
These capabilities may assist researchers develop extra detailed fashions of maximum climate occasions, together with flash floods, landslides and drought situations.
Google highlighted the July 2025 Texas Hill Nation floods as one instance, evaluating PARDIG’s estimates with ground-based radar measurements and NASA’s current rainfall dataset.
Nevertheless, the expertise has limitations that would constrain its use throughout quickly creating emergencies.
Though PARDIG estimates rainfall at 15-minute intervals, Google publishes its near-real-time information in six-hour batches, creating a median delay of about 5 hours between commentary and availability.
Superb time decision doesn’t imply fast supply.
Google’s documented supply schedule, not a measured uptime or pace take a look at. Latency is approximate at publication. PARDIG is experimental precipitation estimation and doesn’t exchange official warnings.
This implies emergency providers can’t use the dataset as a direct substitute for stay climate warnings.
Google additionally acknowledged that the system can underestimate snowfall in mountainous areas and is much less dependable in sure polar areas.
In the meantime, the corporate nonetheless classifies PARDIG as experimental and advises customers to depend on official meteorological companies when making public-safety choices.