Running forecasts in Cloudrun’s Forecast Wizard is simple. Select a region on the map, a start and end time, and click the “Start Forecast Now” button. But the very text of this button illustrates a limitation: What if I don’t want to run my forecast models right now? What if I’m starting an expedition next week and I want to run micro forecasts for each day, 12 and 6 hours before dawn for best last-minute accuracy? What if I want a competitive advantage for my energy forecasting model by using time-lagged ensembles of weather prediction? This is where the new Cloudrun Forecast Scheduler comes in.
The Cloudrun Forecast Scheduler
The Cloudrun Forecast Scheduler is powerful and flexible enough to handle most scheduling and ensemble applications, but it’s dead-simple to use. Start with the Forecast Wizard, but instead of choosing “Start Forecast Now”, click the “Schedule Forecast(s)” button.
In the Forecast Scheduling page, add as many runs as you need by choosing a Run Start Time and clicking “Add To Schedule”. The Run Start Time is the time at which Cloudrun will start computing your forecast. If you’d like to create a staggered schedule or ensemble, just change the Forecast Timeframe instead of the Run Start Time. Or even change both! The scheduler is flexible enough to handle any of these configurations.
Last-minute Forecasting
Typically, the closer you run a forecast to its start time, the more accurate it’ll be. Forecasts are initialized from global models and real-world measurements. If your forecast was initialized with measurements taken 3 hours ago, it’s understandably more accurate than the same forecast initialized with older data from 3 days ago. This is particularly important in complex regions like steep mountains, which are prone to weather predictions changing at the last minute, sometimes drastically. Running your forecasts 3 to 12 hours before the start time can help tame these hard-to-predict areas, but some users have told us they’ve woken up in the middle of the night just to make these critical forecast runs. With the Cloudrun Forecast Scheduler, these last-minute runs can now be scheduled ahead of time.
Time-lagged ensembles
In weather prediction, it’s important to consider more than just a single, deterministic forecast, but also how forecasts change over multiple runs. This gives insights into the model’s nuances for a particular region and weather conditions, as well as the uncertainty associated with the large-scale weather patterns, and presents a more balanced aggregate view. Time-lagged ensembles have been traditionally used by professional meteorologists, but sailors, energy traders, and other weather-dependent industries are starting to use ensembles to gain a competitive advantage above and beyond high-resolution, deterministic forecasts alone. If this sounds interesting but you’re not familiar with time-lagged ensembles, contact us and we’ll give you our honest insight and help you decide if they can provide value for your use case.
Operational weather prediction
In operational weather prediction, a forecast is run recurrently once or multiple times per day as part of a larger real-time system. This is done routinely by national weather services around the world, but also by commercial weather prediction businesses, and even research groups in support of scientific field experiments. Such real-time weather prediction systems are difficult to develop and costly to maintain. While it’s been possible to develop operational, real-time systems with Cloudrun’s API, it’s now easier than ever with the Cloudrun Forecast Scheduler and its user-friendly interface. Once you set up your Forecast Schedule, you can get back to your work and consume the model outputs from Cloudrun as soon as they’re ready. You don’t have to worry about whether the forecasts will start on time or if you have enough disk storage for all the data. Cloudrun takes care of all of that for you.
How will you use it?
We’re thrilled to offer forecast scheduling at Cloudrun. Since our soft-launch a week ago, we’ve received overwhelmingly positive feedback about how flexible the system is. We’ve even been surprised at some of the use cases and configurations so far! If you have an idea for forecast scheduling, or if you have any questions, let us know! To start running high-resolution forecasts, check out cloudrun.co.
