ClimaCell: Building the ultimate weather engine

About ClimaCell

ClimaCell provides minute-by-minute, hyper-local weather forecasts by integrating proprietary data extracted from wireless networks and other new sensing technologies with data from traditional sensors.

Industries: Technology
Location: United States

ClimaCell produces some of the world’s most accurate and granular microweather data from millions of fresh outputs processed with graphics processing units (GPUs) on Google Cloud Platform.

Google Cloud Results

  • Improves performance by 30%, maintaining 5-minute runtime for machine-learning weather models
  • Helps increase accuracy 2x, going from 1-kilometer resolution to 500 meters
  • Increases efficiency and reduces costs to run machine-learning weather models

Processes data from millions of sensing points

There’s a reason why weather is often the topic of “small talk” — it impacts our lives every day, from rained out baseball games to delayed flights. For operators of aircraft, boats, and heavy equipment, having accurate and localized weather information is critical to safety. The weather also impacts the global economy. For farmers, it determines their livelihoods. For financial services and insurance companies, it can mean millions of dollars in investments, claims, or losses.

We can’t change the weather, but with more accurate and timely information, we can plan around it. Traditionally, weather data and forecast quality varied around the world, leaving large rural populations without accurate precipitation information and pilots at the mercy of radar to detect weather hazards.

ClimaCell is leading a revolution with its software-based weather engine. The engine combines proprietary data extracted from wireless networks and millions of other sensing points, including Internet of Things (IoT) sensors, with existing sources of data to produce timely, localized, high-resolution weather data. ClimaCell’s solution offers 90% correlation to ground truth versus radar’s 50% — a key measure of accuracy — making it extremely reliable.

ClimaCell is now on a mission to map the world’s weather data. In addition to its HyperCast software product with proprietary “nowcasting” capabilities for weather-sensitive industries, the company offers a Micro Weather API that delivers granular, minute-by-minute weather data for the next six hours out. To keep accuracy high and deliver forecasts based on fresh data, ClimaCell must maintain high performance for its machine learning (ML) algorithms and keep run times for its proprietary models under five minutes.

“When we moved to Google Cloud Platform, performance immediately improved by 30%. That’s helping us keep our promise to our customers, which is real-time weather data with minimal latency. We can tell the airlines exactly when the precipitation will hit.”

Shimon Elkabetz, CEO & Co-Founder, ClimaCell

To maintain its competitive advantage, ClimaCell moved from another cloud services provider to Google Cloud Platform (GCP), and began using graphics processing units (GPUs) for massive parallel computational power. Using Custom Machine Types configured for large amounts of RAM helps ClimaCell cache weather data in memory and serve it faster to customers.

“When we moved to Google Cloud Platform, performance immediately improved by 30%,” says Shimon Elkabetz, CEO and Co-founder of ClimaCell, who as a former pilot understands firsthand the challenges that unforeseen weather data can present. “That’s helping us keep our promise to our customers, which is real-time weather data with minimal latency. We can tell the airlines and other customers exactly when the precipitation will hit.”

“By optimizing our code for GPUs on Google Cloud Platform, we’ve increased efficiency and reduced the cost of running our machine learning weather models. That savings enabled us to run more models and double our accuracy, going from 1-kilometer to 500-meter resolution.”

Ian Roughley, VP of Engineering, ClimaCell

Reduced costs, improved accuracy

By shifting to GPUs with hundreds of cores, thousands of concurrent hardware threads, and the ability to maximize floating-point throughput, ClimaCell also reduced costs significantly. ClimaCell took advantage of the Google Cloud Startup Program, gaining access to Google experts in data processing, Python scripting, and other technologies.

“By optimizing our code for GPUs on Google Cloud Platform, we’ve increased efficiency and reduced the cost of running our machine-learning weather models,” says Ian Roughley, Vice President of Engineering at ClimaCell. “That savings enabled us to run more models and double our accuracy, going from 1-kilometer to 500-meter resolution.”

Real-time data processing

Weather data is stored on Cloud Storage, and both stream and batch processing are performed using Cloud Dataflow, a fully managed service. As real-time data comes in from networks and sensors, Dataflow grabs windows of data for batch processing as needed.

“Cloud Dataflow is very convenient because we’re able to have real time access and process data as it comes in without being forced into an initial batch mechanism,” says Ian.

Reliable, scalable service

To automatically scale its public API and web-based user interface, ClimaCell uses two Kubernetes clusters running on Kubernetes Engine. Container replication, monitoring, and automated repairs help ClimaCell offer a seamless experience and high availability to users. As ClimaCell maps the world’s weather, it will use Kubernetes Engine along with Terraform to automate consistent cloud infrastructure-as-code deployments in different regions.

“We’ve found that automation and reproducible infrastructure works best on Google Cloud Platform,” says Ian. “Kubernetes Engine helps us provide high availability to customers, allowing us to load balance data against multiple machines to offer a reliable, scalable micro-weather forecasting service to the world.”

“We believe we have the most accurate weather software product in the world, and one that will be critical to many emerging technologies. Our services require high performance and reliability from every component in our stack, from GPUs to networking, data processing, and storage. That’s why we use Google Cloud Platform.”

Shimon Elkabetz, CEO & Co-Founder, ClimaCell

Empowering a dynamic team

In addition to giving developers a user-friendly environment with GCP, ClimaCell gives all its employees access to
G Suite tools such as Gmail, Drive, Docs, Slides, and Sheets to help them collaborate and connect. As a small but growing company with remote employees, ClimaCell appreciates the ability to work together anytime, anywhere.

“G Suite makes it easy to share ideas and feedback,” says Shimon. “Documents are always up to date, so we can make sure everyone is using the same materials with customers. G Suite apps are great for remote employees and work very well on our mobile devices.”

Weather for the new economy

As more companies begin using ClimaCell’s microweather solutions — JetBlue and Autodesk have already signed on — ClimaCell will offer the default weather engine for the new economy. Ride-sharing services, travel apps, drones, and autonomous cars will benefit from better insights into ground-level weather systems delivered every minute by GPU-powered technology.

“We believe we have the most accurate weather software product in the world, and one that will be critical to many emerging technologies,” says Shimon. “Our services require high performance and reliability from every component in our stack, from GPUs to networking, data processing, and storage. That’s why we use Google Cloud Platform.”

About ClimaCell

ClimaCell provides minute-by-minute, hyper-local weather forecasts by integrating proprietary data extracted from wireless networks and other new sensing technologies with data from traditional sensors.

Industries: Technology
Location: United States
Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE
Google Cloud Platform logo

12 Months FREE TRIAL

Try Kubernetes Engine, BigQuery, and other Cloud Platform products with $300 in free credit and 12 months.

TRY IT FREE