Case studies

Nowcast storm predictor

  • Ongoing R&D
  • ML forecasting

A custom nowcasting model for short-range, localized storm prediction: what the sky above this site will do in the next hour, precisely enough to act on. Active research, with a direct roadmap into weather-aware autonomy.

The constraint

Conventional weather forecasting is excellent at tomorrow and poor at the next ninety minutes over your specific location. But the decisions that actually get made in real time (hold the crane or keep pouring, evacuate the site or wait, dispatch crews or stand down) live in exactly that window.

That short-range, hyper-local window, called nowcasting, is where general-purpose forecasts are weakest and where a purpose-built model can win.

What we're building

A custom nowcasting model focused on short-range, localized storm prediction: trained to forecast imminent severe weather for a specific area on a minutes-to-hours horizon, rather than the coarse grid and longer horizon of standard forecasting. The aim is a model that tells an operator what the sky above this site will do in the next hour, precisely enough to act on.

This is active research, and we present it as such: current results are research results, not production claims.

Why it matters

Nowcast compounds the rest of the portfolio: a construction client already running autonomous monitoring is the natural first customer for a model that warns before the storm hits. The roadmap tie-in is direct: weather-aware autonomy, feeding the same operations dashboards as the site-monitoring robots.

Resources

Building something in this shape?

Tell us about it. If it's complex, we're interested.

Start a project