NWS Transitions to Cloud-Based Supercomputing for Hyperspeed Weather Prediction

At Google Public Sector, we believe that operationalizing breakthrough mission projects should never be held back by legacy hardware limits. I’m incredibly proud that NOAA is migrating key workloads to Google Cloud Platform. With this shift, we are enabling the National Weather Service with the advanced cloud tools and resilience they need to protect communities nationwide. #GooglePublicSector #NOAA #GCP #PublicSectorInnovation

What begins as a low threshold for better quality data can sometimes leave you with very little coverage. Increase the threshold, and you may get better coverage, but at the cost of poorer data quality. This was something I came across while working with satellite data in Google Earth Engine, and it gave me a better understanding of an important concept in data cleaning: scene-level cloud filtering and spatial/pixel-level data availability are not the same thing. I was initially using scene-level cloudiness filtering for my Sentinel-2 data. The thinking was simple: a lower cloud threshold should give me cleaner images. But then I noticed the trade-off. With a lower threshold, I had fewer scenes, which left parts of my study area without enough data coverage. When I increased the threshold, I got enough scenes to cover the entire area, but the cloud contamination became much more obvious. That was where I had to look beyond just the overall cloud percentage of a scene. A scene can have an acceptable cloud percentage and still contain cloudy or unusable pixels over parts of the area I am interested in. So instead of only asking, “How cloudy is this scene?”, I started asking, Which pixels are actually usable?” This led me to pixel-level cloud masking, where cloudy pixels can be identified and removed while retaining clear observations from other scenes before creating a composite. It was a small change in my workflow, but it changed how I think about cleaning satellite data in Google Earth Engine. Sometimes, getting cleaner data is not about throwing away more images. It is about being more precise about which parts of the data are actually usable. #GoogleEarthEngine #RemoteSensing #Sentinel2 #Geospatial #GIS #EarthObservation #SatelliteData #DataCleaning #CloudMasking #GeoAI

Your AI agent can return HTTP 200 and still give a wrong answer. That is why agent monitoring has moved into mainstream observability tools this summer: → Grafana Cloud made Agent Observability GA on July 30 and calls agent sessions a "fifth telemetry signal". → Since August 7, Galileo is Splunk Agent Observability. → On September 22, OpenObserve v1.0 shipped open-source agent tracing and evaluations next to logs, metrics and traces. Under the hood, most of this relies on the OpenTelemetry GenAI conventions: invoke_agent, chat and execute_tool spans. They are still in Development status, so pin the version you use. My takeaway for teams: collecting traces is now the easy part. The real work is defining what "good" means — an eval dataset, quality thresholds and cost per session — so a green dashboard actually means good answers. What do you alert on for your agents today? #AIAgents #Observability #OpenTelemetry #LLMOps

Google is preparing to test AI computing in..... space. Project Suncatcher starts small, but the idea is anything but small. If data centres can eventually move beyond Earth, we may be looking at a fundamental shift in how... and where the digital world is powered. The cloud may soon become quite literal. #AI #DataCenters #SpaceTech https://lnkd.in/gEvvRVeT

A National Weather Service move to bring its high-performance supercomputing operations into a cloud environment will give the agency newfound ability to adapt as advancements from technologies — such as artificial intelligence — are expected to change the weather modeling landscape, officials told FedScoop. https://lnkd.in/dZK_qBZv

A National Weather Service move to bring its high-performance supercomputing operations into a cloud environment will give the agency newfound ability to adapt as advancements from technologies — such as artificial intelligence — are expected to change the weather modeling landscape, officials told FedScoop. https://lnkd.in/d4wVQY6x

A National Weather Service move to bring its high-performance supercomputing operations into a cloud environment will give the agency newfound ability to adapt as advancements from technologies — such as artificial intelligence — are expected to change the weather modeling landscape, officials told FedScoop. https://lnkd.in/eAJGCppe

Tech companies are expanding their horizons into space. As Google prepares to launch satellites for data processing, now's the time for engineers to hone skills in AI and satellite communications. Stay ahead by focusing on these emerging fields. #careeradvice #techtrends #spacei

Google DeepMind Drops WeatherNext 3 to Destroy Legacy Climate Models Google DeepMind has officially claimed the absolute global crown in meteorological computing by introducing WeatherNext 3. The groundbreaking model represents a massive structural leap in weather forecasting, aggressively taking the top spot on international leaderboards against both rival AI platforms and traditional government agencies like the U.S. National Weather Service and the European Centre for Medium-Range Weather Forecasts (ECMWF). While older AI forecasting architectures relied on static, historical physics simulations that arrived up to six hours late, WeatherNext 3 uses a dynamic, real-time pipeline that ingests live satellite images and active ground weather stations to refresh its global forecasts every single hour. The computational resolution of the model completely alters local climate tracking. WeatherNext 3 delivers a massive 5x upgrade in localized temperature detail, mapping conditions down to a highly granular 5-kilometer scale, sharp enough to calculate distinct, isolated microclimates inside individual mountain valleys and along jagged coastlines. Crucially, the architecture has conquered precipitation, historically the absolute weakest spot for machine learning weather systems. Google revealed that active forecasting errors dropped by up to 60% when compared directly to verified satellite observation data, boosting the overall accuracy of its day-ahead rain forecasts by up to 50%. The tech giant has already initiated the system's global deployment, integrating WeatherNext 3 natively across Google Search, Google Maps, Gemini, and Google Earth to serve billions of users worldwide. This sovereign infrastructure deployment is a direct strike against legacy supercomputer physics simulation models, slow government meteorological telemetry pipelines, and single-purpose commercial weather data providers. As Google DeepMind proves it can deliver hyper-local, real-time global weather updates updated hourly for pennies on the dollar, the multi-billion-dollar framework for maintaining expensive, slow-moving traditional forecasting grids faces immediate systemic downscaling. #GoogleDeepMind #WeatherNext3 #ClimateTech #AICompute