GWC’s technology improves existing forecasts by applying sophisticated post-processing algorithms to boost accuracy and granularity.
GWC’s technology employs sophisticated algorithms to post-process numerical weather prediction models. These are the models (such as GFS and ECMWF) computed by national weather services. GWC’s automated algorithms apply machine learning to fine-tune the accuracy. This replicates the role of the human meteorologist, aggregating all available models and data to present a more accurate forecast than is possible from any individual model.
The core technology starts by making improved forecasts at sensor locations. A significant strength of our approach is the ability to ingest data from customer-owned sensors and use that data to enhance the forecast at those locations. In real-world tests, our technology outperforms competing forecast.
The technology includes three major steps:
- A regression analysis to bias-adjust each model at the forecast location,
- A weighting of each model based on how well it is expected to perform under current conditions followed by combining the weighted model outputs, and
- A forward error correction that smoothly modifies the forecast over the next few hours to be consistent with current observation
Recently, this engine has been extended from sensor-based forecasts to one capable of providing forecasts globally at non-sensor locations – in other words, at any chosen latitude and longitude. The technology is updated on an ongoing basis through GWC’s R&D investments, including collaborative research with NCAR